Académique Documents
Professionnel Documents
Culture Documents
2012/1(6)/tp
Introduction
A common misconception is that the permanent and the variable load are independent. However, these loads
are independent only during one year and almost fully dependent during the normal service time of structures, 50 years.
Another misconception is that structural loads are combined dependently if the loads are dependent and
independently if the loads are independent. However, loads must always be combined dependently.
Extreme distribution
In the structural design the load distributions denote the extreme values of the loads during a selected reference time. Consequently, the load combination must also consist of the extreme values of the combination
loads. When the loads are combined independently, the load distribution is the sum of extremes but it should
be the extreme of the extremes.
load [kN]
G
1
0
0.2
0
0.2
Q
0
1
0
0.2
0.2
stress [N/mm2]
depend. independ.
1
1
1
1
0.2
0.2
0.2
0.2
0.4
0.37
16.5.2012/2(6)/tp
G G QQ
M
The equality equation makes definite dependence and correlation between the distributions G, Q as the material property M is constant. G and Q have a full negative correlation when one load is increased the other
decreases correspondingly. Due to this correlation the loads G and Q must be combined dependently though
the loads were independent.
permanent load, G
1
0.091471
0.5
normal
1
0.09147
variable load, Q
1
0.4
0.98
gumbel
0.4909
0.1964
The loads can be combined independently by using the convolution equation below, = 1.
FGQ ( x)
FG x r G G fQ r Q Q dr
In this equation, F denotes the cumulative function and f the density function respectively. This combination
results in an equal outcome as the Ferry Borges Castanhetas combination.
The dependent combination is obtained from the same equation with = 1.47672. The - factor fits the
combination distribution to cross the crossing point (1.2413, 0.995831) of the partial distributions G and Q.
1
This value corresponds to G = 1.35, VM = 0.
16.5.2012/3(6)/tp
0.8
0.995
0.8G
GQi
GQd
0.6
0.2Q
0.99
0.8G
0.4
0.985
GQi
0.2
Q
GQd
0.2
0.4
0.6
0.8
0.98
1.2 0.95
1.05
1.1
1.15
1.2
1.25
We study the load values in the fractile 0.98, i.e. in a variable load return period 50 years and in a normal
service time of a structure. The load values in this fractile are tabulated at Table below:
0.2Q0.98
0.2
Q0.98
1
0.8G0.98
0.9503
GQi,0.98
1.0742
GQd,0.98
1.15823
G0.98
1.1879
As explained above the permanent load value and the variable load value are simultaneous in each fractile
when the time is long. Accordingly the loads G0.98 and Q0.98 plus 0.8G0.98 and 0.2Q0.98 occur at the same time
and the loads are combined dependently by adding up the values as such. The combination distribution has
the same fractile as the partial ones. We see at Table, that 0.2.G + 0.8.Q is equal to the dependent combination.
Common sense reasoning
The permanent load G and the variable load Q can be deduced dependent by common sense reasoning: Assume a material (or a structure) has the survival probability S and the resistance 1 for the permanent load G
alone and the variable load Q load alone. Now, if the material is loaded by 0.5G and 0.5Q and the loads are
combined independently, the material has the resistance of ca 1.14 and if combined dependently, the resistance is 1. The independent combination is impossible. There is no link between the loads which would affect one load if the other were present. This is shown numerically below in the demonstration 1.
2
This value applies the load ratio = 0.2
3
This value has been calculated from the convolution equation. This equation is approximate when the distributions are
normal and gumbel. The quantile sum equation is correct and results in 1.1503, i.e. 0.7 % less value.
4
in Eurocode more precisely 1.0646, VG,normal = 0.09147, VQ,gumbel = 0.4
16.5.2012/4(6)/tp
Numerical demonstration 1
Assume a material has the coefficient of variation VM = 0.2057. In Eurocode G = 1.35, G = 1.5, this material
has the material safety factor MG = 1.2309 when it is loaded by the permanent load G and MQ = 1.1078 when
loaded by the variable load Q. We find that MG / MQ = Q / G, i.e. a structural component of this material,
e.g. a tension bar, has equal resistance in permanent and in variable load. Assume, for instance, the material
has characteristic material property xk = 1 and the bar has a cross section area 1, the bar resists a permanent
and a variable load 0.6018.
Assume the same bar is a tension member of a truss and it is loaded by 80 % permanent and 20 % variable load, = 0.2. If the loads are combined independently, the material safety factor is 1.0964 and the bar
resists a load 0.6624 i.e. ca 10 % higher load than G or Q alone. The independent combination cuts down a
part of the load. Therefore the strength is increased. The reason for the error is that the loads must be combined by accumulation, not by random. When the loads are combined dependently, the bar resists load
0.6018 in all load combinations, = 0...1, of G and Q.
Numerical demonstration 2
A permanent load G and a variable load Q of Eurocode are combined here. To simplify calculations, we assume that the distribution of the variable load is normal (gumbel in Eurocode) and the design point value of
the permanent load is 0.98 (0.5 in Eurocode). The design point is set at unity. The coefficients of variation
are VG = 0.0915 and VG = 0.4, i.e. the distribution parameters are G = 0.842, G = 0.077, Q = 0.549,
Q = 0.220. The loads are combined in the serviceability state, i.e., G = Q = 1. We claim that the design
must be fulfilled by 98 % probability. We select the loads to be 1, i.e. the loads are at the design point and at
the validation cut-off point, 0.98. The loads are combined in a load ratio = 50 %, i.e. the loads are combined 50 % + 50 %.
In Eurocode, permanent loads are combined dependently (both loads have the highest value in all fractiles). The combination load distribution is calculated (in this case) by a normal distribution N:
NGQ d x
N x G 1 Q G 1 Q
2
2 2 Q G 1
N x G 1 Q G 1 Q
0.99
0.98
0.97
The figures below show the load distributions, the horizontal axis denotes the load, the vertical axis the fractile,
The combination load value at the validation cut-off point, i.e. in 0.98
fractile, shown by the horizontal solid line, is the key issue. One would
0.96
0.9
0.95
1.05
0.8
0.6
0.4
0.2
0.2
0.4
0.6
0.8
16.5.2012/5(6)/tp
be a common sense value (1 = 0.5 + 0.5). This is true, as the permanent load prevails one during a whole
time and the variable load hits the 0.98-fractile, i.e. the unity load, at least once (by a high probability, this
topic is discussed later) during the service time, 50 years. The full loads, and their halves, are simultaneous
and added up as such. If the loads are combined independently, the combination load is 0.934. In this model,
the load combination lessens the load, 0.5 + 0.5 < 1, which is wrong and results in an unsafe design. In this
combination, the G and Q loads occur simultaneously in fractile 0.98 by a negligible probability. Thus, the
loads must be combined dependently.
Summary
Loads must always be combined dependently in structural design.
In current codes the loads are sometimes combined independently or semi-dependently, which results in
an unsafe error in comparison with the target reliability:
The combination rules including two permanent load factors e.g. the rules 6.10a,b and 6.10a,mod of
Eurocode are induced from the independent combination and these combination rules should be deleted.
The total safety factor, t = (G, Q, M) is up to 10 % too low, if the permanent and the variable load is
combined independently.
Two variable loads are currently combined semi-dependently, which results in up to ca 30 % too low
combination factors 05.
The combination factor 0 for snow is currently 0.5...0.7 but should be 1 as the current snow load distribution is equal to its active time distribution.
The current imposed load factor is 0.7 but should be 1. The reason is that the imposed loads on floors
are dependent and proportions of the whole live load of the house6.
The combination factor 0 of variable loads with short duration, e.g. live load and wind load is ca. 0.6,
i.e. the present values are correct.
The unsafe error of the reliability model due to the independent and the semi-dependent load combination is up to ca 15 % in two loads and up to ca 20 % in three loads. The unsafe error of the actual code is
less, up to ca 15 %, as the safety factors are normally selected higher than obtained from the independent and semi-dependant calculation.
Tampere 16.5.2012
Tuomo Poutanen
Docent, D.Sc. (Civ. Eng.)
Tampere University of Technology
Department of Civil Engineering
P.O. Box 600, FI-33101 Tampere
mobile: +358 40 849 0900
fax: +358 3 3115 2811
email: tuomo.poutanen@tut.fi
www.tut.fi/rtek/poutanen/en
Two variable loads are combined dependently when the distributions are altered in a way the loads are simultaneous
with one on the other.
6
The characteristic live loads in current codes are high enough to compensate the 0 error.