Professional Documents
Culture Documents
Reliability of Beams According To Eurocodes in Serviceability Limit State
Reliability of Beams According To Eurocodes in Serviceability Limit State
Division of Structural Engineering, Lund University, Box 118, SE-221 00 Lund, Sweden
Abstract
To achieve a relatively consistent probability of failure for structural elements, most design
codes apply reliability based code calibration process. Such approaches commonly focus on
the strength of the structural members, which is related to the ultimate limit state (ULS).
However in the design of beams the performance of the structural elements is often limited by
the serviceability requirements, which are related to the serviceability limit state (SLS) using
The current study aims to investigate the reliability for serviceability design for flexural
members made of different materials (steel, concrete and timber) according to the
determine the reliability index for different design situations for beams subjected to bending.
The probabilistic models of basic variables for time invariant analysis have been taken from
the JCSS Probabilistic Model Code. The characteristic, the frequent and the quasi-permanent
combination of actions are investigated and compared. The differences in service reliability
for different materials are discussed. The results show that there are differences between the
achieved reliability indices in the serviceability state between different materials and that for
the given load combinations in the Eurocode the reliability index is often below the one given
in the code.
*
Corresponding author. Tel.: +46 46 222 73 55; fax: +46 46 222 42 12.
E-mail addresses: daniel.honfi@kstr.lth.se (D. Honfi), annika.martensson@kstr.lth.se (A. Mårtensson)
sven.thelandersson@kstr.lth.se (S. Thelandersson)
1
© 2012. This manuscript version is made available under the CC-BY-NC-ND 4.0 license
http://creativecommons.org/licenses/by-nc-nd/4.0/
Keywords: Eurocode; Serviceability; Reliability
1. Introduction
Previous surveys showed that most structural defects are related to serviceability conditions
rather than strength, thus serviceability problems are economically really important and
horizontal displacements of structures and vibrations. The main type of serviceability non-
compliances are excessive floor and roof deflections which may cause damages to adjacent
floors; jammed doors and windows; slanting furniture; ponding; water/moisture penetration;
damage to services; or simply be aesthetically annoying or give the feeling of being unsafe.
Several studies were carried out to investigate the consistency of Eurocodes in terms of
probability of failure for different structural elements and materials [2,3]. However these
investigations – such as the reliability based code calibration process – commonly focused on
the ultimate strength of the structural members. The characteristic values of actions and the
combination factors – which are used in serviceability limit states as well – are mainly
developed and optimized for the ultimate limit state. Other studies investigated the
consistency of the American and Australian design codes for serviceability [1,4] but no such
research was made referring the European structural standard family. Furthermore previous
investigations considered the deflection limits as deterministic variables. The present paper
structural member – made of different materials (concrete, steel and timber) – subjected to
bending considering the deflection limits themselves also as random variables, giving a new
aspect to service reliability. It must also be said that although this paper is restricted to
deflections of beams thereby omitting the important field of vibrations, it is very often the
2
Published in Engineering Structures, Volume 35, February 2012, Pages 48-54.
The final publication is available at ScienceDirect via https://doi.org/10.1016/j.strusafe.2014.03.004
deflection limits that are used in order to give an estimate of the risk for vibrations in practical
design situations. It is also obvious that if the system effects would be considered the
reliability of the structural system could be increased, but then consideration was to be taken
to a number of elements and also to the connections between elements. This would increase
the uncertainty of the modeling thereby complicating the analysis. The intent here is to show
the discrepancies between different materials and the uncertainties that influence the results
The first part of the Eurocodes [5] defines the combinations of actions to be taken into
account in the relevant design situations. Three different load combinations are defined for the
serviceability limit states, which should be appropriate for the serviceability requirements and
performance. The code also distinguishes between irreversible and reversible limit states.
Irreversible serviceability limit states are those where some consequences of actions
exceeding the specified service requirements will remain when the actions are removed.
Contrarily at reversible limit states no consequences of actions exceeding the specified service
∑G
j ≥1
k, j + Qk ,1 + ∑ψ 0,i Qk ,i
i >1
(1)
where Gk,j denotes the characteristic value of the jth permanent action (i.e. the mean value if
the variability of G can be considered as small), Qk,1 is the characteristic value of the leading
variable action (i.e. the 98% fractile for a reference period of one year), Qk,i is the
characteristic value of the ith variable action and ψ0,i is the factor for combination value of a
3
© 2012. This manuscript version is made available under the CC-BY-NC-ND 4.0 license
http://creativecommons.org/licenses/by-nc-nd/4.0/
ith variable action. The characteristic combination is normally used for irreversible limit
states.
∑G j ≥1
k, j + ψ 1,1Qk ,1 + ∑ψ 2,i Qk ,i
i >1
(2)
where ψ1,1 denotes the factor for frequent value of a leading variable action and ψ2,i is the
factor for quasi-permanent value of the ith variable action. The frequent combination is
∑G j ≥1
k, j + ∑ψ 2,i Qk ,i
i ≥1
(3)
The quasi-permanent combination is normally used for long-term effects and the appearance
of the structure.
2.2. Deflections
wc
w1
w2 wtot
wmax
w3
The definition of vertical deflections is shown in Fig. 1, where wc is the precamber in the
unloaded structural member; w1 is the initial part of the deflection under permanent loads of
the relevant combination of actions; w2 is the long-term part of the deflection under
4
Published in Engineering Structures, Volume 35, February 2012, Pages 48-54.
The final publication is available at ScienceDirect via https://doi.org/10.1016/j.strusafe.2014.03.004
permanent loads; w3 is the additional part of the deflection due to the variable actions of the
relevant combination of actions; wtot is the total deflection as sum of w1, w2, w3; wmax is the
partition walls, claddings) is being considered, the verification for deflection should take
account of those effects of permanent and variable actions that occur after execution of the
member or finish concerned, see [5] A1.4.3 (3). However the code does not say anything
about which load combination should be used. In terms of the previous definition of
deflections it is a limitation of the incremental deflections w2+w3. If the comfort of the user, or
the functioning of machinery are being considered, the verification should take account of the
effects of the relevant variable actions, see [5] A1.4.3 (6). This is a limitation of w3, however
the code doesn't indicate what the word relevant means in this context. In case of the
used, see [5] A1.4.3 (4). It is not stated but the investigation is usually carried out for the
remaining total deflection wmax. Long term deformations due to shrinkage, relaxation or creep
should be considered where relevant, and calculated by using the effects of the permanent
actions and quasi-permanent values of the variable actions, see [5] A1.4.3 (6). This basically
The next question is: what are the limiting values of the previously defined deflections? And
how should they be calculated? Which load combinations should be used? These prescriptions
vary from material to material and are given in the appropriate parts of EC. It must be
emphasized that although limit values in some cases are given in the codes, the limit values
differ between materials, there is an uncertainty about which load combinations that should be
© 2012. This manuscript version is made available under the CC-BY-NC-ND 4.0 license
http://creativecommons.org/licenses/by-nc-nd/4.0/
adopted in the calculations and the values given in the codes differs between different
2.3.1. Steel
In case of steel structures Eurocode 3 [6] states that the limits for vertical deflections
according to Fig. 1 should be specified for each project and agreed with the client and notes
that the National Annexes (NA) may specify these limits. However the values given in the
NAs are only suggested values and there are no compulsory rules given.
These prescriptions are quite varying and can be different depending on the function (e.g.
accessible/non-accessible roof, floor etc.), the importance (main girder, purlin), the type of the
carried material (plaster, brittle finish, non-brittle finish) or other conditions of the
investigated element. A typical value for wmax is L/250 and L/300 for w3.
2.3.2. Concrete
The deflection limits for concrete structures – given in Eurocode 2 [7] – should also take into
account the nature of the structure, of the finishes, partitions and fixings and the function of
the structure.
The appearance and general utility of the structure may be impaired when the calculated sag
of a beam, slab or cantilever subjected to quasi-permanent loads exceeds L/250, where the sag
is assessed relative to the supports. This criterion represents a limit for wmax.
Deflections that could damage adjacent parts of the structure should also be limited. For the
loads. It means that the damage criterion is considered by limiting w2+w3 here. Of course
6
Published in Engineering Structures, Volume 35, February 2012, Pages 48-54.
The final publication is available at ScienceDirect via https://doi.org/10.1016/j.strusafe.2014.03.004
2.3.3. Timber
In case of timber Eurocode 5 [8] is a bit confusing, since it gives a different definition of
deflections than was given in [5] (see section 2.2). The components of deflection resulting
from a combination of actions are shown in Fig. 2, where the symbols are defined as follows:
u0 is the precamber; uinst is the instantaneous deflection; ucreep is the creep deflection; ufin is the
The recommended range of limiting values of deflections for beams with span L is also given
in the code (see Table 1) depending upon the level of deformation deemed to be acceptable.
Table 1
Examples of limiting values for deflections of timber beams on two supports [8].
uinst unet,fin ufin
L/300 to L/500 L/250 to L/350 L/150 to L/300
Pf = Φ(− β ) (4)
© 2012. This manuscript version is made available under the CC-BY-NC-ND 4.0 license
http://creativecommons.org/licenses/by-nc-nd/4.0/
where Φ is the cumulative distribution function of standardised normal distribution. For the
Information about the target reliability index for serviceability limit state can be found only
for class RC2. The target reliability for irreversible (!) serviceability limit states in [5] Annex
C is set to be 1.5 for a 50 years reference period and 2.9 for a 1 year reference period for RC2
structural members i.e. structures with medium consequences of failure (e.g. residential and
3. Reliability analysis
In the present paper serviceability failure is deemed to occur when a deflection exceeds an
allowable deflection limit as a result of flexure. For probabilistic calculations this failure can
where X = x1,..., xn the vector of basic variables, δlimit is the allowable deflection limit –
considered as a random variable – and δmax is the maximum deflection of the beam for the
reference period. θE is the coefficient expressing the uncertainty of the action effect and θR is
the uncertainty of the resistance model (in this case of the deflection model). The probability
To calculate the reliability index β from the limit state function the probability distribution
functions of the different design variables are required. They were taken from the JCSS
8
Published in Engineering Structures, Volume 35, February 2012, Pages 48-54.
The final publication is available at ScienceDirect via https://doi.org/10.1016/j.strusafe.2014.03.004
Probabilistic Model Code [9] and [10]. The reliability was then calculated applying Second
Order Reliability Method using the structural reliability software Comrel 8.10 [11].
3.2.1. Loads
The dead load is assumed to be normally distributed with a coefficient of variation (COV) 0.1
and a mean value equal to the characteristic value of the action Gk [9],
The modelling of the imposed loads is the most uncertain of the stochastic variables. Different
studies use different coefficient of variations ranging from 0.2 to 0.6 [2,3,12,13].
Before the reliability analysis Monte Carlo simulation was carried out to estimate the
distribution of the variable loads with a 1 year reference period for floors with different user
category. The magnitude of the sustained load was assumed Gamma distributed with expected
2 2 A0
σ s = σ v,s + σ u ,s κ (7)
A
where A0 is the correlation area, A is the influence area (for beams 2 times the tributary area
AT) and the κ is peak factor (κ=1.4 for beams). The magnitude of the intermittent load is also
2 A0
σ i = σ u ,i κ (8)
A
Both loads – sustained and intermittent – are modelled as Poisson processes with intensities λ
and ν respectively. The arbitrary-point-in-time values of the variable loads for different
occupancy types – and other parameters required for the simulation – are given in Table 2.
Table 2
The parameters of the variable loads for floors [9].
A0 µs σv,s σu,s 1/λ µi σu,i 1/ν dp
[m2] [kN/m2] [kN/m2] [kN/m2] [a] [kN/m2] [kN/m2] [a] [d]
9
© 2012. This manuscript version is made available under the CC-BY-NC-ND 4.0 license
http://creativecommons.org/licenses/by-nc-nd/4.0/
office 20 0.50 0.30 0.60 5 0.20 0.40 0.3 1-3
residence 20 0.30 0.15 0.30 7 0.30 0.40 1.0 1-3
lobby 20 0.20 0.15 0.30 10 0.40 0.60 1.0 1-3
hotel room 20 0.30 0.05 0.10 10 0.20 0.40 0.1 1-3
patient room 20 0.40 0.30 0.60 5 0.20 0.40 1.0 1-3
The upper tail of the simulated results is fitted with a Gumbel distribution representing the
total variable load. Since the characteristic load is defined as the 98th percentile of the annual
µQ 1
= (9)
Qk 6
{− ln[− ln(0.98)] − 0.5772} VQ + 1
π
where µQ is the mean value, VQ is the coefficient of variation and Qk the characteristic value
The results of the simulations show that the mean value µQ and the coefficient of variation VQ
of the imposed load change with the tributary area. Eurocode 1 [14] recommends a reduction
factor αA to take into account the changing of the mean value. Fig. 3 shows, how the mean
value to the modified characteristic value changes with increasing tributary area for different
kind of floors.
Fig. 3. Changing of the mean value of the fitted Gumbel distribution with the tributary area.
10
Published in Engineering Structures, Volume 35, February 2012, Pages 48-54.
The final publication is available at ScienceDirect via https://doi.org/10.1016/j.strusafe.2014.03.004
From the above investigation it is clear that µQ and the VQ depend on the occupancy and the
tributary area. In the following 4 different design situations will be studied with µQ/Qk equal
to 0.4, 0.5, 0.6 and 0.7 representing the situation with a large tributary area for a lobby, an
office, a residence and a hotel room respectively. The coefficient of variations belonging to
these mean values are 0.578, 0.386, 0.257 and 0.165 respectively. To model the action effect
θE is chosen lognormally distributed with a mean value equal to 1 and COV=0.1 [9].
The material models and the applied resistance factors are given in Table 3 for steel, Table 4
for concrete and Table 5 for timber. Since the resistance factor for midspan deflections is not
given in the JCSS code the values for the midspan moment were applied.
Table 3
The simplified probabilistic models of basic variables for steel.
Description X Distribution µX σX
Young’s modulus E Normal En 0.04µX
Moment of Inertia I Normal In 0.03µX
Resistance factor θR Lognormal 1 0.05µX
Table 4
The simplified probabilistic models of basic variables for concrete.
Description X Distribution µX σX
Young’s modulus of reinforciment Es Normal Es,n 0.04µX
Width of the beam b Normal bn-0.003bn 4mm+0.006bn
Height of the beam h Normal hn-0.003hn 4mm+0.006hn
Effective depth d Normal dn 0,02µX
Reinforcement area As Normal As,n 0.02 µX
Concrete compressive strength fck Lognormal fck+2σX 0.17µX
Resistance factor θR Lognormal 1 0.10µX
Table 5
The simplified probabilistic models of basic variables for timber.
Description X Distribution µX σX
Young’s modulus E Normal En 0.13µX
Height of the beam h Normal hn 0.005µX
Width of the beam b Normal bn 0.02µX
Resistance factor θR Lognormal 1 0.10µX
11
© 2012. This manuscript version is made available under the CC-BY-NC-ND 4.0 license
http://creativecommons.org/licenses/by-nc-nd/4.0/
3.2.3. Deflection limits
A critical part of the analysis is modeling the uncertainties of the deflection limits. There
exists very little information about this in literature, primarily due to the fact that it is very
hard to measure and that the acceptance of deflections depend a lot on the situation. Hossein
and Stewart [15] describe a probabilistic model for deflection limits. They consider two types
of damaging deflections: perception damage and partition wall damage. The former is related
to the total deflections (wmax) while the latter to the incremental deflections, where
incremental deflection is that part of the total deflection that takes place after the construction
of partition walls (w2+w3). The parameters of the probabilistic model of the deflection limits
are given in Table 6. Since measuring the incremental deflections is quite difficult, the field
measurements are usually restricted to detect the total deflections thus the proposed model for
partition wall damage corresponds to total deflection values. The authors in [15] suggest that
the incremental deflections may be around 80-85% of total deflections. Hence in the current
probabilistic study the limit for incremental deflection is multiplied by 1.25 to make them
Table 6
The simplified probabilistic models of basic variables for the deflection limits.
Description X Distribution µX σX
Perception damage wmax Lognormal 0.0077L 0.42µX
Partition wall damage wmax Gamma 0.0054L 0.57µX
4. Results
To investigate the effect of the variable actions a load ratio χ is defined representing the ratio
Qk
χ= . (10)
Gk + Qk
12
Published in Engineering Structures, Volume 35, February 2012, Pages 48-54.
The final publication is available at ScienceDirect via https://doi.org/10.1016/j.strusafe.2014.03.004
The partial safety and combination factors applied in the calculations are the ones given by
Eurocode: γG=1.35, γQ=1.5, γM0=1.0, ψ0=0.7, ψ1=0.5 and ψ2=0.3 (for office, domestic and
residential floors).
4.1. Steel
In case of steel beams the design equation for the remaining total deflection is formulated as:
5 qL4 L
wmax = ≤ (11)
384 EI 250
5 qL4 L
w3 = ≤ (12)
384 EI 300
q is the uniformly distributed load − calculated from the appropriate load combination −, L is
the span, E is the modulus of elasticity and I is the second moment of inertia.
Fig. 4 presents the reliability index (with a reference period 1 year) for a steel beam with
varying load factor for the characteristic load combination. The values are below the target
reliability index (β=2.9) and are very low for small values of variable loads. Furthermore the
© 2012. This manuscript version is made available under the CC-BY-NC-ND 4.0 license
http://creativecommons.org/licenses/by-nc-nd/4.0/
Fig. 5. Reliability indices for a steel beam for w3 (characteristic combination).
For w3 the load ratio χ does not influence the results, since the deflection is given by the
variable loads only. However, applying stricter limits increases the reliability – as shown in
Fig. 5. Increasing µQ/Qk ratio of the variable load decreases the reliability. One should note
that the target value is given for irreversible limit states only, thus values far below the dashed
5.2. Concrete
The remaining total deflection of a simply supported concrete beam considering creep and
5 qL4 5 qL4 L
wmax = ζ + (1 − ζ ) ≤ (13)
384 Ec ,eff I 2 384 Ec ,eff I1 250
5 qL4 5 qL4 L
w2 + w3 = ζ + (1 − ζ ) ≤ (14)
384 Ec ,eff I 2 384 Ec ,eff I1 500
where I1 and I2 are the moment of inertia of the uncracked and fully cracked cross-section
14
Published in Engineering Structures, Volume 35, February 2012, Pages 48-54.
The final publication is available at ScienceDirect via https://doi.org/10.1016/j.strusafe.2014.03.004
Ecm
Ec , eff = (15)
1 + ϕ (∞, t 0 )
where Ecm is the secant modulus of elasticity of concrete, φ(∞,t0) is the final creep coefficient
Fig. 6 presents the reliability index (with a reference period 1 year) for a concrete beam with
varying load factor for the characteristic load combination. The results are very similar to
those calculated for a steel beam, however in case of concrete the long-term effects are more
considering the quasi-permanent combination the effect of increasing the variable load is
totally different than for the characteristic combination as shown in Fig. 7. Increasing χ the
reliability decreases especially if the µQ/Qk value is high (e.g. residence, hotel room). That
means that if the dimensions of the beam are governed by the deflection criteria, the limits
will be exceeded with a quite high probability. If we think about the damage criterion it can
be a problem, but on the other side it should be mentioned that the consequences of exceeding
a deflection limit with just a little is usually different than exceeding an ultimate limit state.
15
© 2012. This manuscript version is made available under the CC-BY-NC-ND 4.0 license
http://creativecommons.org/licenses/by-nc-nd/4.0/
Fig. 7. Reliability indices for a concrete beam (quasi-permanent combination, wmax=L/250).
Fig. 8 presents the situation for the frequent combination. If the ratio of µQ/Qk is high, the
reliability is quite consistent. With higher µQ/Qk of the variable load (e.g. hotel room) β
decreases with increasing load ratio, however the effect is not that significant like in case of
consist the effect of the permanent load, thus the load ratio χ influences the results. Fig. 9
shows that the behaviour is very similar to those experienced for wmax, although the β values
are somewhat higher and do not fall below zero. The results presented in this case are
16
Published in Engineering Structures, Volume 35, February 2012, Pages 48-54.
The final publication is available at ScienceDirect via https://doi.org/10.1016/j.strusafe.2014.03.004
5.3. Timber
For timber the deflection limits applied and how they were calculated are as follow:
5 L4 L
u inst = (Gk + Qk ) ≤ (16)
384 E 0, mean I 400
5 L4 L
u fin =
384 E 0,mean I
[ ]
Gk (1 + k def ) + Qk (1 + ψ 2Q k def ) ≤
200
(17)
5 L4 L
u net , fin =
384 E 0,mean I
[ ]
Gk (1 + k def ) + Qk (1 + ψ 2Q k def ) − u 0 ≤
300
(18)
where kdef is the creep factor depending on the type of the wood-based material and the
service class. This definition is not equivalent to calculating the deflections from the quasi-
permanent load combination! In case of the variable load it includes the rare and the quasi-
permanent part of the variable action and hence leads to a significantly higher load level than
© 2012. This manuscript version is made available under the CC-BY-NC-ND 4.0 license
http://creativecommons.org/licenses/by-nc-nd/4.0/
Although it is not indicated in EC5 but from the EC0 prescription it seems to be logical to use
an additional criterion as suggested in [16]. For the deflection from the time dependent part of
5 L4 L
u 2, fin =
384 E 0,mean I
[ ]
Gk k def + Qk (1 + ψ 2Q k def ) ≤
400
(19)
In case of timber all of the previously defined deflections contain the effect of the permanent
loads, thus the effect of load ratio is of interest in all cases. Fig. 10 shows the reliability
indices for the different deflections considering the deflection limit values given in section
2.3.3 with µQ/Qk=0.5 (e.g. office foor) for a reference period 1 year. For all 4 cases the
reliability decreases by increasing the ratio µQ/Qk. This is significantly different from what is
observed for concrete and steel. The reason is the different interpretation of the deflections
and the related load combinations. It is interesting how close the reliability indices for unet,fin
and u2,fin are to each other, even though the former is based on the damage perception
criterion, while the latter on the partition wall damage criterion. The calculations are made
assuming no precamber. Note that if that is the case ufin and unet,fin are the same and since the
18
Published in Engineering Structures, Volume 35, February 2012, Pages 48-54.
The final publication is available at ScienceDirect via https://doi.org/10.1016/j.strusafe.2014.03.004
5. Conclusions
The results of the current paper show that the reliability in SLS is not consistent for different
load ratios and is below the target value (β=2.9), which is given in the code for irreversible
serviceability limit states. In general we can say that the reliability of serviceability of
Eurocodes seems not to be always consistent − since different load levels are used for
• In case of steel the reliability increases with higher portion of variable loads in the
total load which is a favorable situation, since steel structures are usually light and
of exceeding the given deflection limit is quite high especially for higher variable
loads, which can cause problems for innovative light-weight concrete structures.
• For timber members the EC prescriptions to calculate the long-term deflections differ
from those given for concrete which results in different effect of increasing the
There is obviously a need for further investigation in this field, this analysis indicates that
some of the main objectives with future studies should be to discuss relevance of limit values
and also what are the relevant reliability indices in serviceability. System effects must also be
discussed since all structural elements are parts of larger system. The latter point is, however,
difficult and to perform such analysis sound knowledge about the individual elements is
needed.
19
© 2012. This manuscript version is made available under the CC-BY-NC-ND 4.0 license
http://creativecommons.org/licenses/by-nc-nd/4.0/
References
[1] Stewart MG. Optimization of serviceability load combinations for structural steel beam
[2] SAKO 1999. Basis of design of structures – proposals for modification of partial safety
factors in Eurocodes. Joint committee of NKB and INSTA-B. NKB Committee and work
reports, 1999:01 E.
[4] Stewart MG. Serviceability reliability analysis of reinforced concrete structures. Journal of
Belgium; 2002.
[6] EN 1992-1-1 Eurocode 2 – Design of concrete structures: General rules and rules for
[7] EN 1993-1-1 Eurocode 3 – Design of steel structures: General rules and rules for
[8] EN 1995-1-1 Eurocode 5 – Design of timber structures: General rules and rules for
[9] Model Code. Joint Committee of Structural Safety, JCSS; 2001. Available from:
http://www.jcss.ethz.ch.
[10] Sorensen JD, Svensson S, Dela Stang B. (2005) Reliability-based calibration of load
[12] Galambos TV, Ellingwood B. Serviceability limit states: Deflection. Journal of Structural
[13] Melchers RE. Structural Reliability: Analysis and Prediction. 2nd Edition, John Wiley
Belgium; 2002.
[15] Hossain NB, Stewart MG. Probabilistic models of damaging deflections for floor
[16] Blass HJ. Timber Engineering STEP 1: Basis of design, material properties, structural
21
© 2012. This manuscript version is made available under the CC-BY-NC-ND 4.0 license
http://creativecommons.org/licenses/by-nc-nd/4.0/