Professional Documents
Culture Documents
Abstract
Assessing the service life of building products is relevant for all building actors
(insurers, manufacturers, building owners and architects). Indeed, the knowledge of
building products service lives leads to a reduction of maintenance costs and
environmental impact, and an improvement of safety. This paper deals with a
methodological approach for durability assessment. The major steps are :
Research of available durability data and their organization in a graph structure
followed by the assessment of belief and plausibility distribution of service life.
A Failure Mode and Effects Analysis, including a structural and a functional
analysis in order to search all potential failures (weathering factors, product
design and setting up).
The proposed method is a multi-model and multi-scale approach ; multi-model
in order to adjust the model with our knowledge and our aim (modelling real life of
building, but not a too complex and unusable model), multi-scale to take into account
the links between the three geometric scales materials/products/building.
Finally, it gives (1) a distribution of nominal service life, for normal weathering
processes, with corresponding belief and plausibility degrees, (2) details on the
design and setting up problems, on exceptional weathering phenomena, which could
lead to a shorter service life.
Keywords: belief, data quality, data sets, durability, FMEA, failure, functional
analysis, hypergraph, nominal service life, plausibility, service life, structural
analysis.
Introduction
Generalities
Bel ( A) = m(B ) B A
(1)
Pl(A) = 1 Bel(A)
The use of belief functions result from assigning probabilities to sets rather than
to individual points. With evidence theory, it is possible to model the narrowing of a
hypothesis set with the accumulation of evidence.
3
Phenomena
Building
actors
Actors
Models
Method
The three steps of the proposed method are (Lair et al. 1998):
Failure Mode and Effects Analysis (FMEA),
Data compilation and organization,
Service life and belief degree assessment.
Function
Modes
Boring
Solicitations
Defects
Loads
Reasons
Wear and tear
Inner
Effects
Permeability
Punctual stress
Gravel
Permeability
Punctual stress
Maintenance
load
Permeability
Outer
effects
Moisture in
insulation
Moisture in
insulation
Moisture in
insulation
The first column is filled with structural analysis data. Using functional analysis,
we complete columns number 2, 3 and 4. Experts imagine columns 5, 6 and 7.
1
1
1
1
2
2
3
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[55, 75]
[8]
Geometric
granularity
Temporal
granularity
Phenomenological
granularity
product
product
product
product
material (RC)
RC slab
steel, concrete,
paint
phenomenological
decomposition
Limit state
Failure time
failure time
failure time
failure time
failure time
failure time
not explicit
not explicit
not explicit
not explicit
not explicit
Carbonation
not explicit
limit state or
failure time
according to the
decomposition level
4
3
2
1
0
Established theory
Theoretically-based method
Computational model
Statistical processing
Definitions
Lets now assume that is the finite temporal scale ([0, tmax]). Each durability
data correspond to a support function on . The degrees of support are easy to
specify (in this case, the data points precisely and unambiguously to a single nonempty subset A of (Shafer 1976)):
M(B) =
(3)
We apply Dempsters rule (Shafer, 1976) to combine all support functions BEL1,
BEL2 (focusing respectively on A1, A2, ). The resulting support function is
independent of the combination order; and is given by :
BEL(A) = (((BEL1(A1) BEL2(A2)) ) BELn-1(An-1)) BELn(An)
(4)
However durability data are often conflicting data. In order to assess the validity
of Dempsters combination, lets define CON(BEL1, , BELn) the weight of conflict
(between evidences). The weight of conflict between BEL1 and BEL2 is the quantity:
1
with k =
1 k
m ( A ).m
1
(B j )
(5)
i, j
Ai B j =
We can now assess belief and plausibility degrees of service lives (BouchonMeunier, 1995). For example, considering the data A = [25, 100years] and its
corresponding mass m([25, 100]years) 0; we have:
If B = [0, 20[ years
then Bel(SL B) = 0 and Pl(SL B) = 0.
If B = [0, 50[ years
then Bel(SL B) = 0 and Pl(SL B) 0.
If B = [0, 110[ years then Bel(SL B) 0 and Pl(SL B) 0.
The decision criterion is the confidence interval [Bel, Pl] (Dromigny-Badin et
al., 1997). It expresses the knowledge we have about an assumption. The closer we
are to [1, 1], more confident we are about our assumption; the closer we are to [0, 0],
the less probable the assumption is; [0, 1] gives no information.
The methodological approach is shown in Fig. 2 : (1) compile available
durability data, (2) organize data in a hypergraph, (3) assess service life and mass of
each level, (4) combine them to obtain belief and plausibility of service life.
5 Results
Table 4 represents a non exhaustive list of available data. For the moment, we
have used arbitrary masses m. Lets note that m(.) is an intrinsic value of the data.
Thus, we could classify the data according to their own quality, and compare the
belief and plausibility degrees of different products.
Table 4 : Data compilation and mass assignment
Level
Durability data
Mass m(.)
11
12
13
14
21
22
3
4
20 years
More than 30 years
[25, 100 years]
150 years
Distribution (=100, =80)
[100, 120 years]
[5, 130 years]
[55, 75 years]
0.1
0.7
0.8
0.2
0.5
0.6
0.4
0.6
(1)
Data
Data
review
Data
review
Data
review
Scale :
Data
review Scale
:
Data review
Hypot
review Scale
:
Hypot
hesis:
Scale
:
Hypot
Scale : hesis:
Scale
: Forma
Hypot
hesis:
Forma
Hypothesis:
t
:
hesis:Hypot
Forma
Format
hesis:: t : Forma
:
Servicet life:
t : Forma
Reference:
t:
(2)
Levels 11, 12, 13, 14:
(3)
Level 3:
{concrete, steel, paint}
Level 4:
{carbonation, corrosion}
(4)
(SL, m)11
(SL, m)12
(SL, m)13
(SL, m)14
(SL, m)21
(SL, m)22
(SL, m)3
(SL, m)4
1,00
0,90
Plausibility
0,80
Belief - Plausibility
0,70
0,60
0,50
0,40
Belief
0,30
0,20
0,10
170
160
150
140
130
120
110
100
90
80
70
60
50
40
30
20
10
0,00