Sei sulla pagina 1di 3

Use of stress distributions obtained from linear elastic finite element models in code

checks of concrete bridges

Use of stress distributions obtained from linear elastic finite


element models in code checks of concrete bridges
1. Introduction
The use of un-cracked linear elastic FE shell and brick models is becoming
increasingly common for the analysis of concrete bridge structures, but the
interpretation of results continues to cause problems. Such models produce good
results prior to concrete cracking but are rarely realistic after cracking of concrete or
yielding of reinforcement would have occurred in the real structure, or at high
compressive stress such that the true concrete behaviour becomes non-linear. Code
formulae make allowance for this behaviour, permitting redistribution of stresses
across the cross-section. The simplest illustration is for flexure where, on cracking,
flexural stresses are shed from the web to the flanges where the main reinforcement is
provided. Also, code formulae for limiting shear stress are calibrated against average
shear stresses, rather than peak elastic shear stresses as determined in an FE model, so
the use of the latter in a code check is conservative. Conversely, compressive flexural
stresses in the concrete are generally underestimated by linear elastic un-cracked
models when the concrete cracks. Care is therefore needed in interpreting results.
This note is not intended to be a comprehensive guide to finite element modelling of
concrete structures; it gives no guidance at all on constructing such models. The
focus is on interpreting output from un-cracked elastic models.
2. Potential conservatism (and vice versa)
Stress distributions from un-cracked linear elastic finite element models with brick or
shell elements reflect true behaviour up to the onset of cracking. They are therefore
good predictors of peak elastic stresses at serviceability prior to cracking and also the
location of formation of the first crack. They are however poor predictors of
behaviour after cracking (and hence poor predictors at the ultimate limit state) unless
the non-linear behaviour of the materials and concrete cracking can be modelled.
This is because changes in internal lever arm and rotation of web compression struts
due to cracking and plasticity across sections is not modelled. It is not, for example,
correct to take peak elastic shear stresses from such a model and compare them
against the codified limits (in BS5400 or other ultimate limit state codes), because
these make allowance for such re-distribution and are calibrated against average shear
stresses. Analysis of a rectangular section under shear, V, and torsion, T, provides the
simplest example:

V
b

1 of 3

Use of stress distributions obtained from linear elastic finite element models in code
checks of concrete bridges

For shear, the peak elastic shear stress in the un-cracked condition occurs at the elastic
neutral axis and is equal to 1.5 V / bh. The average shear stress used in code checks is
V / bd. Assuming d is about equal to h for a deep beam, the elastic shear stress is
therefore 50% greater than the value to be used in comparing against the code limits.
A rectangular cross section provides the greatest difference in these stresses.
For torsion, the peak elastic shear stress in the un-cracked condition, assuming h >> b,
3T
is
. The plastic shear stress, used in BS5400 Part 4 for comparison with its
hb 2
2T
limits for shear stress, is 2 . The elastic stress is therefore again 50% greater than
hb
the value to be used in comparing against the code limits. (The plastic distribution is
not the correct one, but that which has been used to calibrate against tests).
Despite the above, it should not be assumed that the stresses from a linear elastic
finite element model can always conservatively be compared against code limits.
Whilst, for simple beams, the use of peak shear stresses from such a model in code
checks is usually conservative, if d is in fact much smaller than h, then the elastic
shear stress from the un-cracked model could become un-conservative. For flexural
compressive stresses, it is almost always un-conservative to use the un-cracked values
as cracking reduces the section moduli and increases compressive stresses. Also, in
the web, reinforcement directions generally do not align with the direction of
principal tensile stress just before cracking, so rotation of the stress field needs to
occur. This means that it is not possible simply to compare the principal compressive
stress in the web against the same compression limit as for flexure as a substitute for a
codified check on the concrete shear stress. The angle of the plastic compression field
needs to be considered, which usually requires a sandwich model in conjunction with
reinforcement design equations see 5 below. Code values of v max allow for this
behaviour.
3. Overcoming the problems of un-cracked elastic analysis
To produce stresses and stress resultants more suitable for codified checks at ULS, the
following are possible:
(i)
Refine the model to include the effects of cracking and reinforcement
yielding this requires complex models which often experience
convergence problems, so this approach is not recommended;
(ii)
Isolate elements (such as the rectangle above) or entire beam cross
sections, determine the stress resultants within them (moment, shear,
torque, axial force) and perform code checks using these values as
appropriate. Most software packages, including LUSAS, will allow the
user to take a slice through a cross-section and compute the overall
resultant forces and moments in this way. If stress resultants are obtained
for the whole cross section, code checks will implicitly be allowing for
redistribution across the cross section. If stress resultants are obtained for
individual elements of a cross section, such as webs and flanges,
subsequent code checks on these individual elements will not allow for
redistribution across the cross section.
(iii)
Process the results with a sandwich model see 5 below. Design of
reinforcement via a sandwich model does not allow for redistribution of
2 of 3

Use of stress distributions obtained from linear elastic finite element models in code
checks of concrete bridges

(iv)

(v)

stresses around the cross-section after cracking e.g. shedding of web


flexural stresses to flanges.
Use simpler 3D models with beam elements (rather than shells) where
output is straight away in the form of stress resultants. A further advantage
is that redistribution between elements can be achieved by modifying
section properties of overstressed elements whilst equilibrium is still
maintained (although a check of rotation capacity may become necessary).
The torsionless grillage represents an extreme example.
For complex geometries (e.g. diaphragms) it may be better to design the
reinforcement required using strut and tie models and use the FE model
only to determine the flow of force and guide the construction of an
appropriate strut and tie model.

4. Prestressing ducts in solid models


Where prestressing ducts are present, these are often omitted from the model, thus, in
opposition to the above, making the stresses obtained potentially un-conservative.
They need to be included when performing codified stress checks in accordance with
2(ii) or 2(iii) above.
5. Conversion of concrete tensile stresses from an un-cracked elastic FE analysis
into reinforcement stresses and checking the compression field
For general stress fields, a sandwich model can be used, such as that in Annex LL of
Eurocode 2 Part 2 (EN 1992-2) Concrete bridges, Design and Detailing rules.
Commentary on this can be found in the Designers' Guide to EN 1992-2 Eurocode 2:
Design of concrete structures Part 2, Concrete bridges, Design and Detailing rules.
ISBN: 0727731599. Thomas Telford. Clarks book Concrete Bridge Design to
BS5400 also provides equations for reinforcement design. Without the ability to
model the redistribution across the section which occurs at cracking, it will often be
found that significant reinforcement may appear to be needed in zones which would
ordinarily not be significantly reinforced e.g. longitudinal reinforcement in webs.
For elements with predominantly transverse loading, such as in deck slabs, the
resulting stress resultants can be converted into reinforcement moments using the
Wood-Armer equations see in-house concrete training course session on WoodArmer.
6. Other problems encountered in the finite element analysis of concrete bridges
A variety of other problems, such as dealing with peak values of stress at stress
concentrations, may be encountered. These will be the subject of future guidance but
any immediate issues should be raised with Chris Hendy for incorporation.
Written by: Chris Hendy
Date:
3/7/06
Reviewed by: Mike Chubb
Date:
3/7/06

3 of 3

Potrebbero piacerti anche