Chapter 36

CHAPTER 36

RELIABILITY OF PLASTIC SLABS1

P. Thoft-Christensen, Aalborg University, Denmark

Abstract
In the paper it is shown how upper and lower bounds for the reliability of plastic slabs
can be determined. For the fundamental case it is shown that optimal bounds of a
deterministic and a stochastic analysis are obtained on the basis of the same failure
mechanisms and the same stress fields.

1. Introduction
Upper and lower bounds of the reliability of plastic slabs can be determined on the
basis of the upper and lower bound theorems of general plasticity. Upper bounds are in
this paper obtained on the basis of a geometrically possible failure mechanism and
lower bounds by considering statically admissible stress fields corresponding to
stresses within or on the yield surface. In the paper it is shown that upper and lower
bounds of the reliability index Ru ( x ) and the reliability R of plastic slabs can easily be
found. It is also shown that the best bounds of the stochastic analysis and of the
corresponding deterministic analysis are obtained for the same failure mechanisms and
the same stress fields in the so-called fundamental case.

2. Upper Bounds
Upper bounds of the reliability of a plastic slab can easily be obtained by using the so-
called yield line theory, see Thoft-Christensen & Pirzada [1]. In the yield line theory a
collapse pattern consisting of straight yield lines is assumed and the external work We
performed by the loads and the internal work Wi dissipated in the yield lines are
calculated.
For simplicity, let the loading and the bending moment capacities be modelled by
1
ICOSSAR 89, San Francisco, USA, 1989.

197
 Chapter 36

normally distributed random variables P1 ,..., Pn and M p1 ,..., M pn and let all remaining
parameters such as dimensions be deterministic quantities. A convenient safety margin
Mu then is
M u ( x ) = Wi ( x ) - We ( x )
m n (1)
= f i ( x ) M pi - gi ( x ) Pi
i =1 i =1

where fi and gi are deterministic functions of the variable x = ( x1 ,..., xk ) defining a set
of yield patterns. The safety margin Mu is linear in the normally distributed random
variables, so an upper bound bu ( x ) of the reliability index b ( x ) and an upper bound
Ru ( x ) of the reliability R ( x ) can easily be calculated for any admissible set of
variables a = ( a1 ,..., a10 )
Mu ( x )
E

bu ( x ) = (2)
Mu ( x )
1/ 2
D

Ru ( x ) = 1 - F ( - b ( x ) ) (3)
where E [
] , D[
] , and F are the expected value, the standard deviation and the one-
dimensional distribution function, respectively.
The most significant failure mode within a set of failure modes (yield patterns)
given by x is then obtained by minimizing bu ( x ) with regard to x .
In the corresponding deterministic analysis an upper bound pu ( x ) of the load-
carrying capacity pu ( x ) (only one load) is obtained for any set of parameters from the
work equation
Wi ( x ) = We ( x ) (4)
and the optimal yield pattern corresponds to the set of x parameters resulting in a
minimum value of pu ( x )
It is shown by Thoft-Christensen & Pirzada [1] that the deterministic optimal
yield pattern and the corresponding stochastic most significant failure mode (yield
pattern) are generally different. However, in the so-called fundamental case the two
yield patterns are equal. In the fundamental case M p1 ,..., M pn are fully correlated
random variables and the loading is modelled by a single random variable P. The proof
of this statement is straightforward and is similar to the corresponding proof for lower
bounds given in section 4.

3. Lower Bounds
It is generally believed that the optimal upper bounds bu and Ru determined by the
method in section 2 are so close to the exact values that they can be used for practical
design of slabs. From a theoretical point of view it is more satisfactory to use lower
bounds. The problem is, however, that good lower bounds bl and Rl for b and R are
generally more difficult to obtain.
The equilibrium equation in Cartesian coordinates for a slab can be written

198
 Chapter 36

2 mx ( x, y ) 2 mxy ( x, y ) 2 my ( x, y )
- 2 + = - p ( x, y ) (5)
x2
x y y2
where mx and my are bending moments, mxy is the torsional moment and p is the load
intensity perpendicular to the slab. For the sake of simplicity it is assumed that the load
p(x, y) is a constant p.
To illustrate how lower bounds of the load-carrying capacity p (the deterministic
case) or for b and R (the stochastic case) can be estimated a rectangular slab is
considered. For such a slab it has been suggested, see e.g. Nielsen [2], to construct safe
moment fields of the form
mx = a1 + a 2 x + a 3 x 2
my = a 4 + a 5 y + a 6 y 2 (6)
mx = a 7 + a 8 x + a 9 y + a10 xy
where a = ( a1 ,..., a10 ) are parameters which are chosen so that the boundary conditions
are satisfied in such a way that the moment field is safe, i.e. within or on the yield
surface at any point of the slab.
In a deterministic analysis a lower bound pl of the load-carrying capacity p can be
determined by inserting (6) into (5)
pl = -2 ( a 3 - a10 + a 6 ) (7)
where the parameters a 3 , bl and a10 will depend on the geometry of the slab and the
bending moment capacities. As an example the following expression
1 a b
pl = 8 1+ + mp (8)
ab b a
is obtained for a simply supported isotropic slab with side lengths a and b and bending
moment capacity mp, see Nielsen [2].
In general the equation (7) for the lower bound pl will have the form
m
pl = fi ( a ) m pi (9)
i =1

where m pi , i = 1,, m are bending moment capacities and where fi, i = 1,, m are
deterministic functions.
The optimal lower bound PI of a given set of moment fields as (6) is obtained by
choosing the parameters a so that the maximum of P is determined under the
constraints from the boundary conditions and the yield condition.
A lower bound of the reliability of a plastic slab with the load P and the bending
moment capacities m pi , i = 1,, m modelled as normally distributed random variables
can be estimated on the basis of the safety margin
m
M l ( a ) = fi ( a ) M pi - P (10)
i =1

Note that the safety margin Ml in (10) has the same form as Mu in (1). It is
therefore a trivial matter to estimate the lower bounds bl ( a ) and Rl ( a ) as functions
of a for the reliability index b and the reliability R of a plastic slab.
The optimal lower bound within a set of moment fields is then obtained by
maximizing e.g. bl ( a ) with regard to a under the boundary conditions and the yield

199
 Chapter 36

condition as constraints. An interesting question is then whether the optimal lower

bounds pl and a are obtained for the same a .

4. Lower Bounds for the Fundamental Case

Let the bending moment capacities of the plastic slab be modelled by fully correlated
random variables M p1 ,..., M pm and the load by a single random variable P. Furthermore,
assume that a set of moment fields is defined by a set of parameters a given by (6).
The corresponding reliability problem is called the fundamental case. It will now be
shown that the corresponding deterministic analysis (when all variables are
deterministic) has the same optimal moment fields (or a values) as the stochastic case.
The proof is almost identical to the proof for the upper bounds given in Thoft-
Christensen & Pirzada [1]. The safety margin M can be written
M l = f ( a ) M p - P = k ( M p , P, a ) (11)

where f is a function only depending on a and M p = M p1 . The sensitivity of bl with

regard to each of the parameters a i is then determined by, see Madsen, Krenk & Lind
[3]
bl ( a 0 ) k ( m*p ) 1
= (12)
ai a i k ( m*p , p * , a 0 )

where m p and p * signify the design point and a 0 are the values of a for which bl is
*

calculated. The maximum value of bl is obtained for

bl ( a 0 )
f ( a0 )

=0 =0 (13)
ai ai
In the deterministic analysis the lower bound pl for the load-carrying capacity p
for the corresponding problem is given by
pl ( a ) = f ( a ) m p (14)
it is seen that the optimal lower bound obtained by
pl ( a )
=0 (15)
ai
results in the same equation (13) as for the stochastic analysis. Therefore, the optimal
moment fields are equal.

5. Example
Consider a square slab supported at two adjacent edges (see figure 1). An upper bound
for the reliability can be obtained by a simple yield pattern given by a single yield line
defined by the variable x0 shown in figure 1. The external and internal work
corresponding to a unit downward displacement of point A is
1
We ( x ) = a (3a - x0 ) p (16)
2

200
 Chapter 36

a x
Wi ( x ) = + 0 mp (17)
x
0 a

Figure 1. Square slab supported at two adjacent edges. Uniformly distributed

load P ~ N(4; 0.4) and isotropic bending moment capacity Mp ~ N(1; 0.1).
Setting We ( x0 ) = Wi ( x0 ) the deterministic load-carrying capacity p as a function
of x0 can be determined. The minimum value of p is obtained for x0 =0.72a. The
corresponding optimal yield pattern gives the optimal upper bound for the reliability
index for this set of yield patterns. The upper-bound safety margin is
M u = Wi ( 0.72 a ) - We ( 0.72 a ) = 5.55 M p - P (18)
and the corresponding reliability index bu = 2.26.
A lower bound can be determined on the basis of the moment field given by (see
Nielsen [3])
x x 2
mx = 2 a - 2 (19)
a a
y y 2
my = 2 a - 2 (20)
a a
x xy y
mxy = a - 2 + (21)
a a a
where a is a parameter. By inserting (19) - (21) into the equilibrium equation (5) the
load-bearing capacity pl is obtained as a function of a
1
pl ( a ) = 6a 2 (22)
a
The boundary conditions are fulfilled by (19) - (21). It can be shown (see Nielsen
[3]) that the maximum numerical value of the principal moments is equal to 4a / 3 . The
optimal lower bound pl corresponding to the moment field (19) - (21) is then obtained
from (22) by inserting a = 3m p / 4 . The result is pl = 9mp/2a2.
The optimal lower bound for the reliability index can then be calculated from the
corresponding safety margin (a = 1)
M l = 4.50 M p - P (23)

201
 Chapter 36

The corresponding reliability index is bl = 0.83. Note the big gap between the optimal
lower bound PI and the optimal upper bound bu = 2.26 .

6. Conclusions
In the paper it is shown that upper and lower bounds can easily be determined on the
basis of the upper and lower bound theorems of general plasticity. For the fundamental
case where the bending moment capacities are modelled by fully correlated random
variables and the load by a single random variable it is shown that optimal (bent)
bounds of the stochastic analysis and the corresponding deterministic analysis are
obtained for the same yield patterns and moment fields.

7. References
[1] Thoft-Christensen, P. & G. B. Pirzada: "Upper-Bound Estimate of the Reliability of
Plastic Slabs". In Probabilistic Methods in Civil Engineering (editor P. D.
Spanos), ASCE, N. Y., 1988, pp. 98 - 103.
[2] Nielsen, M. P.: "Limit Analysis and Concrete Plasticity". Prentice-Hall, Inc.,
Englewood Cliffs,1984.
[3] Nielsen, M. P.: "Limit Analysis of Reinforced Concrete Slabs". Acta Polytechn.
Scand., Civil Eng. Build. Constr. Ser. No 26, 1964.

202