You are on page 1of 12

Novák and Vořechovský: Proceedings of the 11th International Probabilistic Workshop, Brno 2013

Structural reliability quantification and cost-based design opti-


mization of concrete elements subjected to fire

Ruben Van Coile, Robby Caspeele, Luc Taerwe


Magnel Laboratory for Concrete Research, Faculty of Engineering and Architecture, Ghent
University, Ghent
ruben.vancoile@ugent.be, robby.caspeele@ugent.be, luc.taerwe@ugent.be

Abstract: Fire is one of the most severe loads for concrete structures. Conse-
quently, fire resistance is an important consideration during design. However,
current design practice is generally governed by prescriptive design rules
which are based on experience and do not explicitly take into account e.g. the
damage costs due to a fire-induced failure and the probability of fire ignition
and successful fire suppression. A more rational approach would be to take into
account the characteristics of the structure and to determine the economic op-
timum fire safety design. As an example, the methodology is applied for de-
termining the economic optimum concrete cover of a simply supported
concrete slab, considering a more refined structural reliability calculation based
on a mixed-lognormal modelling of the bending resistance.

Keywords: cost optimization, fire, concrete cover, concrete slab

1 Introduction
Structural fire safety is a topic of general concern. However, current design rules and
common practice focus on meeting prescriptive requirements which are based on experi-
ence and do not necessarily correspond with the optimum investment in structural fire safe-
ty. A more rational approach would be to consider the specific characteristics of the
structure and its intended use, and to determine the economic optimum fire safety design.
This optimum is obtained by minimizing the total costs, explicitly taking into account e.g.
the fire ignition frequency, the probability of successful fire suppression, the probability of
structural failure in case of fire and the damage costs due to a fire-induced failure. Conse-
quently, this type of optimization balances up-front investments in additional safety with
the expected benefits from loss-prevention.
Van Coile et al.: Structural reliability quantification and cost-based design optimization of concrete elements

2 Cost-optimization of structural elements


The cost-optimization of structural elements can be based on the utility function Y(p) as
proposed by ROSENBLUETH AND MENDOZA [13] and elaborated by RACKWITZ [11]:

Y ( p) = B ( p) − C ( p) − D ( p) (1)
with
p an optimization design parameter
B ( p) the benefit function
C ( p) the initial cost of construction
D ( p) the costs due to failure

Maximizing Y(p), an optimum utility is achieved which balances current investments with
the expected future benefits from loss prevention. Therefore, all future costs and benefits
should be discounted. In accordance with [11], a continuous discounting function is used
relating a future monetary sum A(t) to its value at the time of construction A(0):

A ( 0 ) = A ( t ) exp ( −γ t ) (2)
with
γ the continuous discount rate
t the time (in years)
The theoretical optimum is determined by assuming a systematic replacement after failure
and an indefinite lifetime of the structure. While greatly simplifying the calculations, the
indefinite lifetime does not constitute a problem for most situations as the discounting
function (2) ensures that the results are not significantly affect by long-term costs.
The benefit function B(p) is related to a benefit rate b which can be assumed constant and
independent of the occurrence of failures if the reconstruction time is small compared to
the time between failures. Considering a systematic replacement after failure, B(p) can be
written as:


b
B ( p ) = ∫ b exp ( −γ t ) dt = (3)
0
γ
The construction cost C(p) can be considered to be composed of a cost C0 independent of
the design parameter p and a cost C1 which is function of p:

C ( p ) = C0 + C1 ( p ) = C0 (1 + ε ( p ) ) (4)
with
ε the ratio of C1(p) to C0
The failure cost D(p) is composed of the cost of reconstruction and the external damages
due to failure. These external damages incorporate both human casualties and indirect
costs such as business interruptions and social distress experienced by people affected by
Novák and Vořechovský: Proceedings of the 11th International Probabilistic Workshop, Brno 2013

the failure. A theoretical study of failure costs was made by KANDA and SHAH [8]. Based
on past experience from earthquakes and learned guesses the ratio ξ of the costs due to a
single failure to the initial construction cost C(p) was found to be in the range of 2 for pri-
vate houses, 7 for tall office buildings and over 2000 for nuclear power plants. Assessing
the monetary value of human life and indirect costs falls outside the scope of this paper and
hence optimum solutions will be calculated as a function of this cost ratio ξ. Considering
systematic replacement after failure, D(p) can be written as:

∞ ∞
D ( p ) = ξ C0 (1 + ε ( p ) ) ∑ ∫ exp ( −γ t ) f n ( t , p ) dt (5)
n =1 0

with

f n ( t , p ) the probability density function (PDF) describing the time to the nth
failure
Using Laplace transform and the convolution theorem, equation (5) simplifies to:


f * (γ , p )
D ( p ) = ξ C0 (1 + ε ( p ) ) ∑  f * ( γ , p )  =ξ C0 (1 + ε ( p ) )
n
(6)
n =1 1 − f * (γ , p )
with

f * ( γ , p ) the Laplace transform of the PDF of the time to first fire-induced


failure
Considering (3)-(6), (1) can be written as:

 f * (γ , p ) 
Y ( p ) = − C0 (1 + ε ( p ) ) 1 + ξ
b
 (7)
γ  1 − f * ( γ , p ) 

As the optimum design corresponds with the maximum of Y(p), this design solution is ob-
tained when the derivative of Y is equal to zero:

dY ( p ) d   f * (γ , p ) 
=  0(
C 1 + ε ( ) ) 
p 1 + ξ   = 0 (8)
dp dp   1 − f *
( γ , p )  

3 PDF of the time to first fire-induced structural failure


Evaluating the optimization criterion (8) requires calculating the Laplace transform of the
probability density function (PDF) describing the time to the first fire-induced failure:


f * ( γ , p ) = ∫ exp ( −γ t ) f1 ( t , p ) dt (9)
0
Van Coile et al.: Structural reliability quantification and cost-based design optimization of concrete elements

with
f1 ( t , p ) the PDF of the time to first fire-induced failure

The probability density function of the time to first fire-induced structural failure f1(t,p)
depends on the fire ignition frequency, the probability of successful fire suppression and
the probability of structural failure in case of a fully-developed fire.

3.1 The time to the first fully-developed fire


If fire ignitions are assumed independent, the time between fire ignitions Tig can be mod-
elled by an exponential distribution with parameter λ equal to the yearly probability of fire
ignition p1. Consequently, the time to the kth fire ignition can be modelled by a Gamma
distribution:

λ ( λt ) exp ( −λt )
k −1

fTig ,k ( t ) = (10)
( k − 1)!
The first fully-developed fire occurs when a fire ignition coincides with a failure of fire
suppression. This probability of fire suppression psup is modelled in accordance with
ALBRECHT and HOSSER [1] as:

psup = (1 − p2 p3 p4 ) (11)
with
p2 the failure probability of fire suppression by users of the building
p3 the failure probability of early fire suppression by the fire brigade
p4 the failure probability of the sprinkler system

For a building without sprinklers (i.e. p4 = 1), psup can be taken as 0.9. This value corre-
sponds with a common probability of fire suppression by the users (p2 = 0.5 according to
[1]) and a fire brigade intervention time of less than 15 minutes (i.e. p3 = 0.2).
The probability that the kth fire ignition results in the first fully developed fire is:

Pfi ,k = (1 − psup ) psup


k −1
(12)

with
Pfi , k the probability that kth ignition results in the first fully-developed fire

Combining equations (10) and (12), the PDF of the time to the first fully developed fire is
given by an exponential distribution with parameter (1-psup)λ:

λ ( λt ) exp ( −λt )
∞ k −1

fT , fi ( t ) = ∑ (1 − psup ) p
k =1
k −1
sup
( k − 1)!
(
= (1 − psup ) λ exp − (1 − psup ) λt ) (13)
Novák and Vořechovský: Proceedings of the 11th International Probabilistic Workshop, Brno 2013

3.2 The time to the first fire-induced structural failure


Given the occurrence of a fully-developed fire, the severity of the fire can be linked to an
ISO 834 standard fire duration tE by taking into account for example the fire load and ven-
tilation characteristics of the structure [9]. Alternatively, the standard fire duration tE can
be taken from national legal requirements which prescribe the resistance to fire as a func-
tion of the dimensions and the usage of the structure. For example in the UK, values of tE
can be found in Approved Document B (Fire Safety) [10].
Subsequently, the probability of failure Pf,tE in case of exposure to tE minutes of the stand-
ard fire should be assessed, using for example the advanced calculation tool described by
VAN COILE ET AL. [15]. This probability of failure will depend on the characteristics of the
structure.
Knowing Pf,tE, the probability that the kth fire ignition results in the first fire-induced struc-
tural failure can be calculated as:

(
Pf , fi ,k = Pf ,tE (1 − psup ) 1 − Pf ,tE (1 − psup ) )
k −1
(14)

with
Pf , fi , k the probability that the kth ignition results in the first structural failure

Equation (14) has the same layout as equation (12). Consequently, the time to the first fire-
induced structural failure is given by:

( )
fT , f , fi ( t ) = Pf ,tE (1 − psup ) λ exp − Pf ,tE (1 − psup ) λt = f1 ( t , p ) (15)

where Pf,tE is a function of the design parameter p.

3.3 Evaluating the Laplace transform and the optimization criterion


As the time to the first fire-induced structural failure is described by an exponential distri-
bution, the Laplace transform (9) can be evaluated analytically:

(
f * ( γ , p ) = ∫ exp ( −γ t ) Pf ,tE (1 − psup ) λ exp − Pf ,tE (1 − psup ) λt dt )
0

Pf ,tE (1 − psup ) λ
(16)
=
γ + Pf ,tE (1 − psup ) λ

This allows to elaborate the optimization criterion (8):

dε ( p )  Pf ,tE ( p ) (1 − psup ) λ  ξ dP ( p )
1 + ξ  + (1 + ε ( p ) ) (1 − psup ) λ f ,tE =0 (17)
dp  γ  γ dp
 
The derivative of Pf,tE may be evaluated numerically.
Van Coile et al.: Structural reliability quantification and cost-based design optimization of concrete elements

4 Application example: simply supported solid slab


A simply supported solid slab with a thickness of 0.2 m and a free span of 6 m is load-
bearing in one direction and carries an uniformly distributed permanent load gk consisting
of the self-weight of the slab, and a uniformly distributed variable load qk of 3.8 kN/m².
According to EN 1990 [2], the design value of the bending moment MEd induced by these
loads equals 50.4 kNm. Consequently, the design value of the bending moment capacity
MRd calculated in accordance EN 1992-1-1 [3] should be larger than or equal to 50.4 kNm.
As many publications emphasize the importance of the concrete cover for structural fire
resistance, e.g. [16], choosing a large concrete cover may correspond with additional safety
in case of fire. However, for a constant value of MRd, increasing the concrete cover necessi-
tates the placement of additional reinforcement bars. Consequently, an economic optimum
concrete cover may be determined for which the up-front investment in additional rein-
forcement is balanced with a reduction of the expected losses due to fire-induced failures.
The basic variables used for describing the slab are given in Tab. 1, as well as the associat-
ed probabilistic models. These models are chosen in accordance with the study performed
by HOLICKY AND SYKORA [6] and HOLICKY ET AL. [7].

Tab. 1: Probabilistic models for the basic variables involved in the analysed concrete slab
Symbol Name Distri- Mean µ Standard
bution deviation
σ
MRd [kNm] design value of the bending moment capacity in DET 50.4 -
normal design conditions
h [mm] thickness N 200 5
fc [MPa] 20°C concrete compressive strength (fck = 30 MPa) LN 42.9 6.4
fy [MPa] 20°C reinforcement yield stress (fyk = 500 MPa) LN 581.4 40.7
c [mm] concrete cover {min = µ-3σ; max = µ+3σ} Beta cnom 5
As [mm²] bottom reinforcement area per m (Ø = 10 mm) N 1.02As,nom 0.02µ As
kfc(θ) [-] concrete compressive strength reduction factor at Beta θ depen- θ de-
temperature θ (see also [16]) dent pendent
kfy(θ) [-] steel yield stress reduction factor at temperature θ Beta θ depen- θ de-
(see also [16]) dent pendent
KR [-] model uncertainty of the resistance effect LN 1.1 0.11
KE[-] model uncertainty of the load effect LN 1 0.1

The nominal reinforcement area As,nom is defined by:

π∅2 1000
As ,nom = ⋅ (18)
4 s
with
s horizontal spacing of the reinforcement bars
∅ reinforcement bar diameter (10 mm)
The parameter s is varied as a function of the concrete cover in order to maintain a constant
design value for the bending moment capacity MRd.
Novák and Vořechovský: Proceedings of the 11th International Probabilistic Workshop, Brno 2013

4.1 The probability of structural failure Pf,tE(p)


The probability of structural failure associated with an ISO 834 duration of tE minutes is
related to the limit state function:

Z = K R M R , fi ,t − K E ( M Q + M G ) (19)

with
M R , fi ,t bending moment capacity at t minutes of ISO 834 duration
MG bending moment induced by the permanent load
MQ bending moment induced by the variable load

For MG a normal distribution is assumed, while for MQ a Gumbel distribution is considered


[6]. The model uncertainties KR and KE are described by a lognormal distribution (Tab. 1).
Traditionally, a lognormal distribution is also considered as appropriate for the stochastic
representation of the resistance effect [14]. However, as elaborated in [16], the uncertainty
with respect to the concrete cover necessitates that the bending moment capacity MR,fi,t of
concrete slabs during fire is modelled by a mixed-lognormal distribution:

M R , fi ,t = ∑ P [ ci ] M R , fi ,t ( ci ) (20)
i

with
P [ ci ] the probability of a concrete cover in the range ci ± ∆c/2
M R , fi ,t ( ci ) the lognormal distribution of MR,fi,t for a fixed concrete cover ci
∆c the concrete cover interval width (1 mm)
Consequently, the probability of failure (i.e. P[Z < 0]) can be written as:

P [ Z < 0] = ∑ P [ ci ] P  K R M R, fi ,t ( ci ) − K E ( M Q + M G ) < 0 = ∑ P [ci ] Pf ,t ( ci ) (21)


i i

with
Pf ,t ( ci ) the probability of failure for a fixed concrete cover ci

For each of the constituent functions with lognormal distributed MR,fi,t(ci), a FORM analy-
sis can be applied together with the Rackwitz-Fiessler theorem to determine the reliability
index βfi,t(ci). This reliability index can be associated with the failure probability Pf,t(ci)
through [2]:

Pf ,t ( ci ) = Φ ( − β fi ,t ( ci ) ) (22)

with
Φ ( .) the standardized cumulative normal distribution
Van Coile et al.: Structural reliability quantification and cost-based design optimization of concrete elements

Evaluating equations (21) and (22) for the slab configuration of Tab. 1 allows to visualize
the probability of failure as a function of the ISO 834 fire duration tE for different values of
the nominal concrete cover cnom (Fig. 1).

Fig. 1: Probability of failure for the example slab as a function of the ISO 834 exposure
duration for different nominal concrete covers c

4.2 The ratio of additional investments in safety ε(p)


As defined above, ε(p) is the ratio of the additional cost C1(p) associated with the design
parameter p to the basic construction cost C0. For the specific situation under considera-
tion, the additional costs refer to the additional reinforcement bars necessary to compensate
an increase in nominal concrete cover.
For the slab configuration of Tab. 1, Fig. 2 visualizes the relationship between an increase
in concrete cover ∆c and the corresponding increase in reinforcement area ∆As, with cref a
reference concrete cover of 15 mm and As,ref the associated reinforcement area (683 mm²).
As illustrated in Fig. 2, the following linear approximation can be used:

∆As ∆c c − cref
≈ 0.11 = 0.11 (23)
As,ref cref cref
Novák and Vořechovský: Proceedings of the 11th International Probabilistic Workshop, Brno 2013

Fig. 2: Additional reinforcement area as a function of the increase in concrete cover

If the costs with respect to the additional reinforcement can be considered proportional
with the reinforcement area, then:

C1 ( p ) a∆As As ,ref aAsref ∆c


ε ( p) = = = 0.11 (24)
C0 C0 As ,ref C0 cref
with
a the proportionality constant

4.3 Evaluating the optimization criterion


Considering equation (24), the optimization criterion (17) can be evaluated as:

As ,ref  Pf ,tE ( p ) (1 − psup ) λ   ∆c  ξ dPf ,tE ( p )


(1 − psup ) λ
aA
0.11 1 + ξ  + 1 + 0.11 s ,ref  =0
C0  γ   C0 cref  γ dp (25)
 

This optimization criterion gives the value of ∆c which results in the lowest overall cost
with respect to fire-induced structural failure. Fig. 3 visualizes this optimum nominal con-
crete cover cnom for different ignition frequencies λ, as a function of the reinforcement cost
ratio aAsref / C0. Other variables are taken for a high-rise office building: tE = 120 minutes,
psup = 0.9 and ξ = 7. Furthermore, the discount rate γ is assumed to be 0.03, as was used in
[4] for optimizing the number of escape routes from road tunnels.
Van Coile et al.: Structural reliability quantification and cost-based design optimization of concrete elements

Fig. 3: Optimum concrete cover as a function of the reinforcement cost ratio and the fire
ignition frequency λ, (ξ = 7, psup = 0.9, γ = 0.03, tE = 120 minutes)

Fig. 3 illustrates how the fire ignition frequency can have an important influence on the
optimum concrete cover. Similarly, the influence of the failure cost ratio ξ can be investi-
gated. Assuming an ignition frequency of 2.5 10-3 per year for an office building, as indi-
cated by RAHIKAINEN AND KESKI-RAHKONEN [12], the optimum concrete cover is given in
Fig. 4 as a function of the reinforcement cost ratio for different values of ξ.

Fig. 4: Optimum concrete cover as a function of the reinforcement cost ratio and the failure
cost ratio ξ, (λ = 2.5 10-3 per year, psup = 0.9, γ = 0.03, tE = 120 minutes)

Fig. 4 indicates that a higher failure cost ratio ξ results in a larger optimum concrete cover.
However, for large reinforcement cost ratio aAs,ref / C0, the optimum cover decreases.
Novák and Vořechovský: Proceedings of the 11th International Probabilistic Workshop, Brno 2013

Furthermore, there is no consensus considering the discount rate γ. The influence of the
discount rate on the target reliability level for structures was investigated by HOLICKY in
[5]. The effect of the discount rate on the optimum concrete cover for the slab configura-
tion of Tab. 1 is investigated in Fig. 5.

Fig. 5: Optimum concrete cover as a function of the reinforcement cost ratio and the dis-
count rate γ, (ξ = 7, λ = 2.5 10-3 per year, psup = 0.9, γ = 0.03, tE = 120 minutes)

It is concluded that the discount rate is an important parameter with respect to the cost-
optimization of structural elements. As expected, a higher discount rate favours less up-
front investment in safety, while a low discount rate corresponds a with higher concrete
cover as the optimum design.

5 Conclusions
• Mathematical formulations for deriving optimum design solutions for structural ele-
ments exposed to fire are presented. The optimization balances up-front investments in
structural safety with the expected benefits from loss prevention.
• As an example application, the economic optimum concrete cover is derived for a spe-
cific simply-supported concrete slab configuration as a function of the fire ignition fre-
quency, the failure cost ratio and the cost of additional reinforcement. Parameter
studies are presented demonstrating the effect of changes in key parameters on the op-
timum design solution.
• The derived formulations are generally applicable and can be used for other design
problems as well, e.g. the optimum reinforcement ratio of a concrete column, or the op-
timum investment in fire suppression measures in a wooden building.
Van Coile et al.: Structural reliability quantification and cost-based design optimization of concrete elements

Literature
[1] Albrecht, C., Hosser, D., Risk-informed framework for performance-based structural
fire protection according to the eurocode fire parts, Proceedings of the 12th Interflam
Conference, 2010, p. 1031-1042.
[2] CEN, EN 1990: Basis of structural design, European Standard, 2002
[3] CEN, EN 1992-1-1: Design of concrete structures: Part 1-1: General rules and rules
for buildings, European Standard, 2004
[4] Holický, M., Probabilistic risk optimization of road tunnels, Structural Safety 31
(2009), p. 260-266
[5] Holický, M., Optimisation of the target reliability for temporary structures, Civil En-
gineering and Environmental Systems 30 (2013), p. 87-96
[6] Holický, M., Sýkora, M., Stochastic models in analysis of structural reliability, Pro-
ceedings of the International Symposium on Stochastic Models in Reliability Engi-
neering, Life Sciences and Operation Management, 2010
[7] Holický, M., Retief, J.V., Dunaiski, P.E., The reliability basis of design for structural
resistance, Proceedings of the third International Conference on Structural Engi-
neering Mechanics and Computation (SEMC), 2007
[8] Kanda, J., Shah, H., Engineering role in failure cost evaluation for buildings, Struc-
tural Safety 19 (1997), 79-90
[9] Kodur, V.K.R., Pakala, P., Dwaikat, M.B., Energy based equivalent approach for
evaluating fire resistance of reinforced concrete beams, Fire Safety Journal 45
(2010), p. 211-220
[10] National Building Specification, Approved Document B – Fire Safety: Volume 2 –
Buildings other than dwellinghouses (2013 edition), 2013
[11] Rackwitz, R., Optimization – the basis of code-making and reliability verification,
Structural Safety 22 (2000), p. 27-60
[12] Rahikainen, J., Keski-Rahkonen, O., Statistical Determination of Ignition Frequency
of Structural Fires in Different Premises in Finland, Fire Technology 40 (2004), p.
335-353
[13] Rosenblueth, E., Mendoza, E., Reliability Optimization in Isostatic Structures, Jour-
nal of the Engineering Mechanics Division 97 (1971), p. 1625-1642
[14] Torrent, R.J., The log-normal distribution: a better fitness for the results of mechani-
cal testing of materials, Matériaux et Constructions 11 (1978), p. 235-245
[15] Van Coile, R., Caspeele, R., Taerwe, L., Quantifying the structural safety of concrete
slabs subjected to the ISO 834 standard fire curve using full-probabilistic FEM, Pro-
ceedings of the 10th International Probabilistic Workshop, 2012, p. 313-330
[16] Van Coile, R., Caspeele, R., Taerwe, L., The mixed lognormal distribution for a
more precise assessment of the reliability of concrete slabs exposed to fire, Proceed-
ings of the European Safety and Reliability Conference (ESREL 2013), in press

You might also like