Вы находитесь на странице: 1из 10

A STOCHASTIC APPROACH TO THE FACTOR METHOD FOR

ESTIMATING SERVICE LIFE

L.I.AARSETH and P.J.HOVDE


Dept. of Building and Construction Engineering, NTNU, Trondheim, Norway

Durability of Building Materials and Components 8. (1999) Edited by M.A. Lacasse


and D.J. Vanier. Institute for Research in Construction, Ottawa ON, K1A 0R6,
Canada, pp. 1247-1256.
 National Research Council Canada 1999

Abstract

The ISO/CD 15686-1 “Service life planning” describes a deterministic method that
allows an estimate of the service life to be made for a particular component or assem-
bly in specific conditions. In “real life” the service life has a big scatter and should be
treated as a stochastic quantity. In this paper we introduce the “step-by-step” principle
as a stochastic approach to the ISO factor method. The “step-by-step” principle pro-
vides a more systematic approach to the estimating process and makes possible a sto-
chastic handling of the factors. For each factor three estimates shall be made; the
minimum, maximum and the most expected value of the factor. In this way the un-
certainty is identified and estimated for each factor. The most uncertain factors
should, if possible, be divided into sub-elements and more information should be gath-
ered in order to reduce the uncertainty. In this stochastic approach the “factors” are
treated as elements which finally are summed up. Also unlike the proposed ISO factor
method, the estimates are expressed in years, instead of in numbers close to 1. These
changes facilitate seeing the consequences of the estimates during the estimating proc-
ess. After a statistical calculation the estimated service life is expressed as three fig-
ures; the expected value plus/minus one standard deviation. Two examples are shown
where the service life is estimated for a window: first in a deterministic way according
to the proposed ISO factor method, then in a stochastic way according to the proposed
“step-by-step” principle.

Keywords: Estimating service life, "step-by-step” principle, stochastic approach, ISO


factor method
1 Introduction

The ISO/CD 15686-1 “Service life planning” describes a method that makes pos-
sible to estimate the service life of a particular component or assembly under specific
conditions. An important shortcoming in this method is that the service life is treated
as a deterministic value. In reality the service life has a big scatter and should be
treated as a stochastic quantity (Siemes 1997). A stochastic treatment requires the aid
of a statistical tool, and preferably a tool designed for practical utilisation. In this
report the “step-by-step” principle is introduced as a stochastic approach to the ISO
factor method.

2 The ISO factor method

The following definitions are taken from “ISO 6707 Vocabulary of construction
terms – Part 1 General terms” (as referred in the ISO/CD 15686 – 1; 1997).

Table 1: Definitions

Period of time after installation during which all essential properties of an item meet or exceed
Service life (SL)
the required performance.
Period of use intended by the design, e.g. as established by agreement between the client and
Design life (DL)
the designer to support specification decisions.
Reference service Service life established for a class of building or for parts of a building for use as a basis for
life (RSL) estimating service life in specific items in specific conditions.
Estimated service Reference service life multiplied by factors related to specific conditions, e.g. materials, de-
life (ESL) sign, environment, use and maintenance
The method allows an estimate of the service life to be made for a particular
component or assembly in specific conditions. The method is based on a reference
service life and a series of modifying factors that relate to the specific condition of the
case. The reference service life is the expected service life in the conditions that gen-
erally apply to that type of component or assembly.

Table 2: Deterioration factors of materials and components

Factors Relevant Conditions (examples)


Quality of compo- Manufacture, storage, transport, materials, protective coatings (fac-
Agent related A
nents tory-applied)
to the inherent
B Design Level Incorporation, sheltering by rest of structure
quality char-
acteristics Work execution Site management, level of workmanship, climatic conditions during
C
level the work execution
D1 Indoor environment Aggressiveness of environment, ventilation, condensation
Environment Outdoor environ- Elevation of the building, micro-environment conditions, traffic
D2
ment emissions, weathering factors
Operating E In use conditions Mechanical impact, category of users, wear and tear
Conditions F Maintenance level Quality and frequency of maintenance, accessibility for maintenance
Any one (or any combination) of these variables can affect the service life. The
factor method can therefore be expressed as a formula in which:

ESLC = RSLC x A x B x C x D1 x D2 x E x F”
The output of the equation is a single number; the estimated service life of the
component (ESLC). The following quotation from chapter 7 in the ISO/CD 15686 – 1
indicates the intended use level of the factor method (in 1997):

“The factor method is a way of bringing together the consideration of each of the
variables that is likely to affect service life. It can be used to make a systematic as-
sessment even when little or no reliable test data is available. Its use can bring to-
gether the experience of designers, observations, intentions of managers and manu-
facturers assurances as well as some data from test houses.
The factor method does not provide an assurance of a service life - it merely
gives an estimate based on what information is available. It is less reliable than a fully
developed prediction of service life. The distinction between estimated and a predicted
service life should be made when a forecast of service life is given. The information
taken into account should also be recorded, so that it is clear whether the estimate is
particularly robust or not.
It may be desirable to consider the consequences of failure when estimating
service lives using the factor method. It may be used as a guide both to those compo-
nents which should be included in estimates (it is not necessary to estimate every com-
ponent’s service life) and in deciding those where failure would be most critical. For
these it may be necessary to be extra cautious, either by raising the design life (the
“target”) or by being particularly rigorous in assessing the value of factors.
The factor method can be applied to both components and assemblies. When ap-
plied to assemblies it is necessary to consider the interfaces (e.g. joints) between com-
ponents as well as the components themselves. For example, different external envi-
ronment and maintenance factors may apply to a whole assembly that relies on sea-
lants to weatherproof the joints between factory made cladding units than that which
would apply to the each of the individual cladding units.”

3 The “step-by-step principle”

The conceptual context in the original description of the “step-by-step” principle


is project planning. In this paper most notions are left unchanged but, where neces-
sary, their relevance for estimating service life is emphasised. The “step-by-step”
principle is based on “Successivprincippet”, which was developed by Lichtenberg
(1990) at DTU in Denmark in the early 70’s. The “step-by-step” principle was devel-
oped at NTNU in Norway (Klakegg 1993) as a tool for improving the quality of the
basis for decisions in project planning under uncertainty. As with estimating service
life, a basic problem in project planning at an early stage is the lack of relevant infor-
mation.
In “The step-by-step principle – a systematic approach to project planning under
uncertainty” Klakegg (1993) states that the "step-by-step” principle, when applied to
e.g cost planning, puts focus on the way the plans are made, and not only on the con-
tents of the plans. Improved communication and awareness of the process of planning
is vitally important in order to improve the plan itself. This is why the “step-by-step”
principle is applied in the planning session with a resource group, consisting of plan-
ners and project members, guided by a moderator. The “step-by-step” principle is
built on knowledge from professions like psychology, sociology, pedagogy, creative
problem solving, and management formed into a consistent and flexible working pro-
cedure. This procedure offers support to the process of documenting the basis for de-
cisions in accordance with the modern philosophy of quality management. The aim of
implementing a systematic approach to planning under uncertainty is normally to im-
prove the economic benefit of the project. This is achieved by improving the basis for
decision-making and improving control and management in the project by focusing on
the situation in which decisions are made. Practical use of the “step-by-step” principle
has demonstrated that the plan is realistic and relevant for decision-makers.
In traditional project planning, estimates of costs, time consumption etc. are han-
dled as deterministic values. The uncertainty of these estimates is taken into account
after the main calculations are done, and then added to this sum as a “safety value”. In
the “step-by-step” principle, the uncertainty is identified and estimated for each ele-
ment (cost, time etc.) during the calculation. Thus the “step-by-step” principle identi-
fies the most uncertain elements. These elements may then, if possible, be divided into
sub-elements and/or more information can be gathered in order to reduce the uncer-
tainty.

3.1 The main steps in the “step-by-step” principle


Klakegg (1994) describe the main steps in the “step-by-step” principle as shown
in Figure 1.

1 General 3 Dividing
Problem
definition 2 conditions Structure
4

Estimate
Detailing
7 6 Calculation
Conclusion
Evaluation 5

Fig 1: Elements of the iteration process in the “step-by-step” principle

3.1.1 Problem/task definition


Before the planning process can start, the project and the approach to the prob-
lem must be defined. This is important to ensure an efficient project work without
disturbing discussions about irrelevant issues. To assist and clarify the definition
phase, a situation map can be drawn.

3.1.2 Situation map


A situation map should highlight the essential elements of the project, such as
clarity of the objective, acceptance of the objective, organisation, market conditions,
innovation, complexity, size, duration. When estimating service life, other elements
will be more relevant. This could be a list of the factors affecting the service life.

3.1.3 General conditions


This is a main point in the “step-by-step” principle. All external and internal
factors likely to influence the project itself, as well as the project execution, must be
highlighted. This should normally be done in two separate steps:

1. Brainstorm; identify all conditions of relevance


2. Select and evaluate/consider particular important conditions

In the proposed ISO factor method the factors are supposed to cover all condi-
tions affecting the service life, including “general conditions” and synergy effects.
“General conditions” and synergy effects can be numerous and difficult to account for
in a reasonable and correct way. The total effect of the assumed contributions to each
factor might also be rather arbitrary and hard to foresee. In section 4.2.1 we suggest a
new, separate factor, in order to cover general conditions and synergy effects in a
clearer and more separate manner.

3.1.4 Division, structure


The project must be divided into separate, independent parts, suitable for the in-
dividual project. In the first loop of the “step-by-step” cycle, the division should be
rather coarse. In later loops (if necessary) one or more parts/groups can be divided
into sub-groups, until the division is satisfactory. (This explains why the “step-by-
step” expression is used.)
The seven factors in Table 2 form the first division in a service life estimation
process.

3.1.5 Triple estimate


The quality of the estimates of costs/time consumption etc. is a crucial step in the
“step-by-step” principle. Three values must be estimated, and preferably in the fol-
lowing order:

1. The minimum estimate (the 1% percentile)


2. The maximum estimate (the 99%percentile)
3. The (most) expected estimate (not necessarily the statistical expected value, see
Fig. 2

By focusing on the extreme estimates first, the estimates will be conservative and
more realistic. Too much focus on the expected estimate could yield too narrow and
definite estimates. Wide exterior values will ensure:

• A wide range of estimated values that will contain the “real value”, even if this
value is quite different from the most expected estimate
• Lower risk of hidden uncertainty

Good estimates can be achieved by a “neutral” (but subjective) evaluation, free


of prejudice, of all input data. A group of individual persons (the resource group) are
supposed to make individual estimates. After a discussion, the three values are fixed
for each factor to estimate. This group process will even out individual elements of
pessimism or optimism. Expertise within the resource group can be utilised for better
estimates of the most expected value.
Bourke and Davies (1996) state that for practical use of the factor method, the
accuracy and range of the factors should be considered, e.g. by a panel of experts for
individual materials.

3.1.6 Calculation
The typical triple estimate will form a skew probability density function as
shown in Fig. 2.

f(x) "Statistical" expected value (f)

x
“Minimum” “Expected” “Maximum”
estimate ( l ) estimate (m) estimate (h)

Fig. 2: Probability density function

The “Erlang” density function with k = 10 is applied in the “step-by-step” princi-


ple, which is assumed to give a reasonably good statistical representation of the triple
estimates (Klakegg 1993). The statistically expected value f and the standard deviation
σ can be calculated from the following equations:

l + 2.95 ⋅ m + h h−l
f = σ = (1, 2)
4.95 4.6
where
l = minimum estimate (low)
m = expected estimate (medium)
h = maximum estimate (high)

When the plan/project has been divided into suitable parts and the triple esti-
mates have been done for each part, the whole plan must be thoroughly calculated.
Special software for this purpose has been developed at NTNU (Klakegg 1993).
When the total project values are calculated, the sum probability density function
is assumed to be a normal distribution. This assumption can be done when adding up a
number of independent elements of reasonably uniform size (according to the Central
Limit Theorem). This assumption will not always be fulfilled, and the sensitivity of
the calculations might need to be evaluated in some cases.

3.1.7 Evaluation
After the calculation, the results must be evaluated. Some questions to answer
are listed below:

• Does the plan contain unreasonable or unrealistic elements (definition, division,


general conditions or synergy effects)?
• Does the plan include all general conditions and synergy effects (-of relevance)?
• Is all information taken into account, that is known at the moment when the cal-
culations are done?
• Does the result satisfy the needs?

3.1.8 Detailing and iteration


Normally it will be necessary to do one or more iterations if the plan is too
coarse. Some of the elements might need to be divided into smaller sub-elements that
are more clearly defined. This can enable better triple estimates.

3.1.9 Conclusion and complementary work


If the preceding analysis is carried through in a systematic and well-documented
manner, much important information for future activities will be gathered. That is in-
formation for maintenance planning, quality management of design and work execu-
tion.

4 How can the “step-by-step” principle improve the proposed ISO factor
method?

4.1 Possibilities
Two major aspects of the “step-by-step” principle could be useful for the service
life estimation process:

1. The factors will be handled as stochastic variables. The uncertainty of each single
factor will be estimated.
2. All conditions that are assumed to influence the service life can be identified and
quantified in a more systematic way.

4.2 Changes to the ISO factor method

4.2.1 Removing interactions between the factors


The statistical treatment in the “step-by-step” principle is based on a structure
where each cost element or factor can be assumed independent. In the ISO factor
method there normally will be strong interactions between the factors. The way these
effects are taken into account in the factor method gives results that are difficult to
foresee. The problem can be handled in two different ways:

1. Altering the statistical treatment. This can hardly be done simply enough for prac-
tical use.
2. “Remove” every assumed synergy effect or interaction between factors. Evaluate
the effect of the interactions in a new factor, such as the general factor in the “step-
by-step” principle.

The latter alternative will be used in this paper, and the new factor, G, is named
“general conditions”.

4.2.2 Turning factors into elements and percentages into years – the element method
Several major problems entail the use of factors:

1. It is difficult to find or calculate a reasonable accurate number for the modifying


factors.
2. It is difficult to evaluate the consequences of each factor in service life/years.
Conversion calculations must be done in order to evaluate these.
3. The effect of multiplying factors is difficult to foresee. In the “step-by-step” prin-
ciple, the estimates are added up; this is far more observable.

Problem number 1 in the section above will not be addressed further in this pa-
per. Problem 2 and 3 may be reduced by:

1. Expressing the estimates in years, and


2. Adding up the estimates for each factor instead of multiplying them.

If the estimates are expressed directly in years it will be easier to see the effect of
the estimates. The estimates should be made relative to the RSLC, namely the number
of years that the given conditions are assumed to shorten and/or prolong the service
life. Finally the l-, m-, and h-estimates may be added up in order to find the ESLC and
the total uncertainty, see Table 3. After such a change, the factors should be named
elements. In order to avoid mixing up the two methods, the altered method will be
named the element method.

5 Examples of ESLC calculations

In the ISO/CD 15686 – 1 an example is shown where the service life of a soft-
wood window is estimated. The important conditions that influence each modifying
factor are listed in a matrix. Their assumed contribution to the service life is divided
into three groups: poor, average and good. 0.8, 1.0, and 1.2, respectively, is proposed
as default values for the modifying factors. In the example the RSLC is 25 years, and
the seven modifying factors are 1.2, 1.2, 1,2, 1, 1,2, 1 and 1.2.

5.1 ESLC calculation according to the ISO factor method


ESLC = 25 x 1.2 x 1.2 x 1.2 x 1 x 1.2 x 1 x 1.2 = 62.2 years. In the ISO-example
the result is rounded off to 60 years.

5.1.1 Discussion
The factor method lacks information about uncertainty. An error range should be
given for both the result (ESLC) and for the individual factors upon which the results
are based.

5.2 ESLC calculation according to the “step-by-step” principle; the element


method
The factors from the ISO-example will be transformed into years relative to the
RSLC of 25 years. The m-estimates correspond to the factors in the example above,
and are found by multiplying the factors (ranging from 1.0 to 1.2) to the RSLC and
then subtracting the RSLC ((25 x 1.2) – 25 = 5, and (25 x 1.0) – 25 = 0).
The l-estimate is found by subtracting (0.2 x 25 = 5 years) from the m-estimate.
This corresponds to one step “down” in the three-divided range for factors in the ma-
trix in the ISO example. The h-estimate is found by adding 5 years to the m-estimate.
A more moderate addition of e.g. 2.5 years would be more realistic, since it is harder to
improve a top score than reduce a high/medium score. However, by choosing a sym-
metric range of ± 5 years, the examples will be more comparable. For the same rea-
son, the G-element is left inactive (set to 0).

5.2.1 Result

Table 3. Calculation sheet

Estimates Statistical
Elements l m h calculation f s Part of s
A Material and components 0 5 10 f= 5.00 2.17 14 %
B Design 0 5 10 (l+2.95 x m+h) 5.00 2.17 14 %
C Sitework/execution 0 5 10 /4.95 5.00 2.17 14 %
D1 Indoor environment -5 0 5 s = 0.00 2.17 14 %
D2 Outdoor environment 0 5 10 (h - l)/4.6 5.00 2.17 14 %
E Operating characteristics -5 0 5 0.00 2.17 14 %
F Maintenance level 0 5 10 5.00 2.17 14 %
G General conditions 0.0 0.0 0 0.00 0.00 0%
Sum l, m, h = -10 25 60 Sum f = 25.0 15.22 100 %
RSLC + sum l, m, h = 15 50 85 Sqrt-sum σ = 5.8
ESLC, "statistical values" = 44 50 56 RSLC + sum f = 50.00 ESLC = 50 years
RSLC = 25 years Result +/- 6 years

Table 3 shows the estimates that are made and how the ESLC are calculated.
The ESLC are 50 years ± 6 years (± 1 standard deviation or ± 12%).

5.2.2 Discussion
The element method gives a more conservative result than the factor method, due
to the different nature of addition and multiplication (62 and 50 years, respectively).
With a lower and more realistic h-estimate, this difference would have been even big-
ger (an h-estimate of m + 2,5 years would give a RSLC of 46 ± 4 years).
The scattering of ± 12 % indicates that the estimates are too definite and narrow.
If the estimates had been made in a practical case, preferably in a resource group, the
range of the estimates would most probably have been wider due to the present lack of
reasonably accurate “service life” data.
As mentioned earlier, the result from the first calculation step would normally
have an unacceptable amount of scattering. In this example “environment” is already
divided into “indoor” and “outdoor”. Another sub-division could be within the A-
element, e.g. A1 “material” and A2 “protection” (painting). “Outdoor environment”
could also be divided into “driven rain”, “time of wetness”, “air temperature”, “pollu-
tion” etc.
Making estimates in “years” instead of in percentage of the RSLC might be con-
fusing in some situations, e.g. when estimates are made for similar components with
different RSLC. This possible drawback should be considered in further work.
6 Summary

By use of the “step-by-step” principle, as shown in the element method, the un-
certainty is identified and estimated for each factor/element, as well as for the ESLC.
This gives an opportunity to identify the most uncertain elements for further detailing,
and for making better estimates.
The estimating process can be improved by changing the factors into elements
that are added together. The estimates can also be made directly in years. These
changes will make it easier to see the consequences of the estimates during the estima-
tion process.
The statistical calculation makes the element method more complicated than the
factor method. However, a computer with any kind of spreadsheet software should be
available, at least for reporting, regardless of which estimating method that is used.

7 Conclusions

The “step-by-step” principle enables a stochastic handling of the modifying fac-


tors in the ISO factor method by performing a triple estimate for each factor. After the
statistical calculation the estimated service life is expressed as three figures: the ex-
pected value pluss/minus one standard deviation.
The “step-by-step” principle also demonstrates a systematic approach to the es-
timating process, including the process of defining, subdividing and estimating the
modifying factors in the ISO factor method.

8 References

Bourke, K. and Davies H. (1996) Factors affecting service life predictions of build-
ings: a discussion paper. Watford, Construction Research Communications Ltd.,
England.
ISO/CD 15686 – 1 (1997) Buildings: Service Life Planning, part 1 – general princi-
ples.
Klakegg, O.J. (1993) Trinnvisprosessen. Dept. of Building and Construction Engi-
neering, NTNU, Norway (in Norwegian).
Klakegg, O.J. (1994) The step-by-step principle – a systematic approach to project
planning under uncertainty”. Department of Building and Construction Engi-
neering, NTNU, Norway.
Lichtenberg, S. (1990) Projektplanlægning i en foranderlig verden. 2. Ed. Polyteknisk
forlag, Lyngby, Denmark, (in Danish).
NS 3420 (1972) Beskrivelsestekster for bygg og anlegg. Norwegian standard, Norges
Byggstandardiseringsraad, NBR, Oslo, Norway, (in Norwegian)
Siemes, T. (1997) Probability and reliability in service life prediction. TNO Building
and Construction Research, Delft, The Netherlands.

Вам также может понравиться