You are on page 1of 15

TRM 41

TECHNICAL REFERENCE Rev 3

MANUAL Date 9/05


Civil & Structural
Page 1 of 15
COMPUTER MODELS OF STRUCTURES

INTRODUCTION

Computer models of structures need to be well thought out to avoid wasted time and in the
extreme potential failures due to the model not being appropriate for the structure under
consideration. This TRM provides guidance on the application of computer models and how the
best results can be obtained (this may mean the computer is not the most appropriate method of
design). Guidance is also given on the approach to be taken in assessing the validity of the
computer output.

GENERAL ADVICE

Do I need to use the computer?

Before starting work on a computer model, consider the alternatives. If the structure is relatively
simple, it may be easier to use design charts from the Civil / Structural Designer’s Data Pack, the
Reinforced Concrete Designer’s Handbook or the Steel Designers’ Manual, approximate methods
from codes of practice or simple rules of thumb. At scheme stage, when the form of the structure
is likely to undergo major changes, a computer model may not be justified - simplified hand
calculations may be more appropriate.

Use the computer if it saves time - if a number of similar calculations with minor variations are
expected (eg several continuous beams with variations in span/section/loading, or a structure with
a large number of different load cases and combinations) - or if a program automates a calculation
that would be long-winded if carried out by hand (eg design of composite beams, portal frames,
minimum weight/cost structures).

It is often necessary to use a computer to obtain very accurate results - although using a computer
does not of itself guarantee accuracy.

Which program should I use?

Again, consider the alternatives. Normally it is best to use a program designed for a particular
task rather than a more general package. For example, a specialist portal frame program should
allow investigation of plastic hinge formation and snap-through buckling which a linear elastic
frame analysis program would not. If the choice is borderline, use the program that is most
familiar to you. Consult colleagues and users in other offices (SITS can tell you which other
offices use the various programs).

Before starting to use a particular program, ensure that it is able to take into account all the design
requirements you have. For example, when checking crack widths in a reinforced concrete
retaining wall, the program should allow for the effect of axial compressive as well as lateral loads.

Check that the program is able to provide output in the form you require. For example, a grillage
frame analysis may not be able to provide Wood-Armer moments for a reinforced concrete slab.

Read the manual

Make sure you are thoroughly familiar with the program before you try to do anything complicated.
Ask other people who have used it. Try some simple structures first. If there is a tutorial, work
through the examples before starting on a “real” structure. Compare results with answers from

WSP Group
TRM 41
TECHNICAL REFERENCE Rev 3

MANUAL Date 9/05


Civil & Structural
Page 2 of 15
COMPUTER MODELS OF STRUCTURES

hand calculations and other programs. Check the underlying engineering assumptions made by
the programmer. Read the manual, use the on-screen ‘help,’ look up technical support on the
manufacturer’s web site or contact the manufacturer direct.

Limitations

Remember that in most cases the computer bases its analysis on a linear-elastic model. In other
words, deflections and other effects are directly proportional to applied loads. Sometimes the real
structure is not like this. Struts buckle, wires go slack and pin joints stick. Some programs allow
such behaviour to be modelled but great care is needed with non-linear models - for example,
results from different load cases cannot simply be added together to give combined results. In
addition, not all programs analyse structures with non-linear elements in the same way, and so
results may be different.

Linear elastic programs construct models based on the initial geometry of the structure and
assume small deflections. Real structures often function very differently after some deflection has
taken place. BS 5950: 2000 requires “P-delta” effects to be taken into account in certain
situations, and some programs are able to do this with more accuracy than others. A program
which applies the loads, calculates the deflected shape and then re-applies the loads produces
less accurate answers than one which uses a “true P-delta” calculation method.

Change in behaviour following deflection, requiring P-delta analysis

Computer programs calculate numbers to different numbers of significant figures. If there are
large differences in stiffness between elements (a thick pilecap on relatively flexible piles, for
example) then a program that employs “double precision” calculations should be used. If possible,
avoid large differences in stiffness between different parts of the model.

Again, the manual and other users will provide information about program limitations.

Sensitivity tests

It is usually possible to produce many pages of computer output giving results to several decimal
places. This is misleading, because the input to the program is rarely very accurate. Loads can
be different in magnitude and position, and the response of the structure may be less predictable
than the computer model implies.

On all except the most straightforward models, sensitivity tests are advisable. Adjust spring
stiffnesses, support locations, member properties and other parameters (in turn, not all together)
to observe the effect they have on the output. This will show up deficiencies in the model, and
give an indication of the level of accuracy that can be expected. If the structure itself is particularly
sensitive it may be an indication that it needs to be changed in some way to design out the
problem.

WSP Group
TRM 41
TECHNICAL REFERENCE Rev 3

MANUAL Date 9/05


Civil & Structural
Page 3 of 15
COMPUTER MODELS OF STRUCTURES

Record what you do

The input data for the computer model (eg node/member numbering, member/support properties,
and loads) must be printed out and kept in the calculation file, together with plots which present
this information visually. In addition, it is very important to record separately the data and
assumptions that are used to create the model, the derivation of the applied loads and the checks
that are carried out. If the model is being constantly ‘tweaked’, ensure the input data is from the
same version of the model as the output data. Do not rely on others being able to interrogate the
model using the computer, although it is advisable to record the file name of the model, and the
directory in which it is stored.

WHAT SORT OF MODEL IS NEEDED?

Frame models

These are usually the most simple computer models to input, check, understand, interpret and use
to design real members. Many programs can only work with frame models. Although clearly most
appropriate where well defined beams exist when the structure is built – eg steel frames, bridge
decks made from prestressed concrete beams - they can also be used (with care) for slab
structures by defining a grillage of equivalent beams.

Typical frame models

Finite element models

Finite element models are more complex to work with but are useful for complicated shapes made
of homogeneous parts. For example, a flat slab with irregularly spaced columns is best modelled
using 2D plate elements. A steel casting for a connection node would be suited to 3D solid
elements. It is very important to use the right type of element where there is a choice, and the
program manual or help file usually contains useful advice. If there is no choice it is still important
to recognise the limitations of the default element. Often this is a “shell” type which relies on the
assumption that each element is “thin” in relation to its length or breadth. If this is not true for the
structure the output may not be realistic.

WSP Group
TRM 41
TECHNICAL REFERENCE Rev 3

MANUAL Date 9/05


Civil & Structural
Page 4 of 15
COMPUTER MODELS OF STRUCTURES

Typical finite element models

Finite elements and beams may be combined in the same model - for example a composite
structure could be modelled using finite elements for the slab and beam elements for the steel
members (suitably offset vertically to model the difference in neutral axis positions).

How big should the model be?

No bigger than necessary. Big models take longer to build, take longer to solve, contain more
possibilities for error, need more checking and produce more output.

Model size and complexity can be reduced by considering one floor of a building at a time, by
using planar instead of 3D models, or by separating the structure into sections that can be
modelled individually.

However, errors can be caused by models which are too small. Satisfy yourself that the sections
of the structure which have been omitted can have only a negligible effect on the rest. For
example, column shortening may be significant if spans are very uneven, and if the columns are
represented as providing perfect vertical fixity the model will be inaccurate.

Sometimes it is best to build a broad-brush “global” model in order to determine overall behaviour;
smaller more refined models are then used to concentrate on areas of interest.

BUILDING THE MODEL

Nodes

Most analysis programs require that structures are defined as interconnected nodes, and so the
first step involves specifying the node locations. This is relatively simple, but consider the
following:
• Provide nodes at positions where results are required - it is usually easier to obtain results
at nodes than at intermediate points on beams.
• Similarly, specify a node half-way between support points to obtain the mid-span results.
• If beams or slab edges are curved, make sure that the nodes defining the curve are
reasonably close together. Otherwise the small angles at node points will introduce
unwanted secondary effects.

WSP Group
TRM 41
TECHNICAL REFERENCE Rev 3

MANUAL Date 9/05


Civil & Structural
Page 5 of 15
COMPUTER MODELS OF STRUCTURES

Supports, restraints and springs

Supports must be clearly defined. A model supported in a way which does not reflect the actual
structure on site will produce incorrect results. For example, if a truss is modelled with supports
which prohibit horizontal movement at both ends, then the top chord will act as a catenary.
Instead of being in compression for its whole length, it has lengths in tension at each end. Small
horizontal movements - similar to bolthole tolerances - release the catenary forces and so the
“fixed at both ends” condition will probably never occur. Similar problems arise if moment restraint
is wrongly assumed.

Effects of support fixity

It is only necessary to consider support fixity when the structure is sensitive to the effects it
causes. Forces in simply supported structures - such as the truss above - are unaffected by
vertical movements of the supports and so they can be fixed vertically in the computer model.
Horizontal fixity does affect the structure and so it is necessary to think about horizontal supports
more carefully. Greatest care is needed with 3D structures, as would be expected.

Columns are often described as having “pinned” or “fixed” bases but in practice both completely
free rotation and total moment fixity are impossible to achieve. Reference 1 includes some
recommendations for modelling columns where it is particularly important to take realistic
behaviour into account.

WSP Group
TRM 41
TECHNICAL REFERENCE Rev 3

MANUAL Date 9/05


Civil & Structural
Page 6 of 15
COMPUTER MODELS OF STRUCTURES

Modelling of column base fixity

Where supports are liable to move, and where this could affect the structure, spring supports
should be used. Forces in statically indeterminate structures can change a great deal if supports
move and so it is important to obtain correct values for spring stiffnesses. Advice can be obtained
from WSP Environmental on the stiffness of piles and other foundations; if the spring models the
support from another piece of structure a value can be obtained from a hand calculation or another
computer model. Sensitivity checks are vital whenever springs are used.

Effect of spring supports

As noted above, there may be a great difference in stiffness between a deep pilecap and the piles
which support it, and a “double precision” computer model will be called for.

It is important to restrain the model so that it is not able to move or rotate as a whole. For
example, if the model was free to move along the X axis the computer may calculate enormous
deflections even though no force is applied to the model in that direction. This is because
rounding errors lead to a non-zero force being divided by a zero stiffness. A similar problem can
occur with pin ended beams that are left free to rotate about their own axis. Sometimes the
program will provide restraints for parts of the structure which are free to move in this way, but the
results can be unpredictable. If the program identifies errors through singularities or lack of
restraints, correct them yourself.

Beams

Beam elements are reasonably simple, particularly when the real structure is made of them. Portal
frames and trusses are good examples of this. Position beam elements on the centroid of the
member - the depth of a truss in a computer model is less than the overall depth.

WSP Group
TRM 41
TECHNICAL REFERENCE Rev 3

MANUAL Date 9/05


Civil & Structural
Page 7 of 15
COMPUTER MODELS OF STRUCTURES

Beams can also be used to model slab structures, and if this approach is adopted the results are
often easier to interpret than those from finite element models. If the slab contains “real” beams -
precast bridge beams, for example - follow them. Transverse members should be at 90º to
longitudinal members unless there is a good reason for another angle, such as real beams or
skewed reinforcement. If there are no real beams - as with a solid slab - space the beams at as
near a regular spacing as possible. One grillage member should be fairly close to the edge of the
slab (within 0.3 × depth) in order to pick up the torsion stresses.

Slab modelled as grillage (transverse members omitted for clarity)

Finite elements

With finite elements, the advice is not “read the manual” but “read and understand the manual.”
Finite elements can be very useful tools for modelling structures, but there are many pitfalls for the
unwary.

First of all, make the elements the right shape. Triangular elements should be equilateral or nearly
so; quadrilateral elements should be approximately square. Quadrilateral elements can have a
length to breadth ratio of up to 3 without much loss of accuracy, and corners do not have to be
exactly 90º. Always avoid internal angles greater than 120º.

Remember that the elements are only joined at nodes so that if you use a finer grid - near an
opening or corner, for instance - it will be necessary to use triangular elements to reduce spacings
without loss of accuracy.

It is often sensible to reduce element sizes at stress concentrations and other highly stressed
areas. A rule of thumb is to halve the node spacing in these locations.

Possible finite element grid reduction near internal corner

WSP Group
TRM 41
TECHNICAL REFERENCE Rev 3

MANUAL Date 9/05


Civil & Structural
Page 8 of 15
COMPUTER MODELS OF STRUCTURES

The length and breadth of the elements should not be less than about twice their thickness. If you
find that this gives you too few elements, then the structure is probably not suited to 2D elements
because the assumption in finite element theory that the elements are “thin” will break down.
Having said that, the basic rule for accuracy is “the more elements the better.” The graph below
shows the tip deflection of a cantilever composed of varying numbers of finite elements compared
with the result from simple beam theory. Results only approach the correct value when fairly large
numbers of elements are used. (This is an easy experiment to do yourself.)

Effect of number of elements on accuracy of model

However, it is possible to produce enormous models which take ages to solve, with results files
occupying many megabytes of disc space. The two constraints - accuracy and model size - have
to be balanced. Model sizes may be reduced by modelling the area of interest with a fine mesh,
and other areas rather more coarsely. Symmetry can be exploited to reduce the model size.

Local axes

It is important, just as in real life, for beam elements to be the right way round. For 2D models this
is not too difficult - you tell the computer whether the beam is being bent about its x-x or its y-y
axis. With 3D models the beam’s neutral axes are always at 90º to the beam and to each other,
but can be at any angle to the horizontal. The computer program will choose a default angle - how
it does this is usually explained in the manual or the on-line help - but it is important to check that it
is correct. Always show the local axes on the screen and look at the model from at least two
directions. Some programs are able to display a rendered view of the model and this can provide
a useful check.

Finite elements also have local axes and it is important that they are correct. One of the axes will
be at right angles to the plane of the element; whether it points “up” or “down” depends on the
order in which the corner nodes are specified. Again, the direction of the axes should be checked.

Section properties

Most programs allow section property values to be entered directly - as area, I value etc - or to use
standard shapes such as BS steel sizes, rectangles or tubes. Obviously it is important to check
the units - is the program expecting inertia values in cm4 or m4?

A special approach is needed to torsional inertia if a slab is being modelled as a grillage of beams.
Torsion moments in the two directions of the grillage are not independent - a torsion moment about
one axis automatically leads to a torsion moment in the orthogonal direction (see figure below).
Because of this it is necessary to halve the torsional inertia of the grillage members, or the slab
would be too stiff in torsion.

WSP Group
TRM 41
TECHNICAL REFERENCE Rev 3

MANUAL Date 9/05


Civil & Structural
Page 9 of 15
COMPUTER MODELS OF STRUCTURES

Equality of orthogonal torsion moments in slabs

For example, a stand-alone beam with a rectangular cross section of width b and depth d has a
torsional inertia of bd3/6, but if it forms a grillage member in a slab model a value of bd3/12 should
be used. It will be necessary to edit the torsion properties manually if the program has calculated
them from the dimensions of the section.

Manual calculation of A, Ixx and Iyy is not difficult, and for reasonably regular shapes with no re-
entrant angles the torsional inertia value J is (Ixx + Iyy)/2. However if the cross-section is at all
complex it is best to use a commercial program such as Prokon to calculate section properties.

Many structures can be modelled successfully by setting torsional inertia values to zero. If the
structure is reasonably ductile the lack of torsional resistance will be accommodated by the
bending moments increasing. If the strength of the structure is in equilibrium with the applied
loads it will continue to stand. This approach avoids the problems that arise in designing for
torsion but can cause unwanted cracking or deflection as the moments redistribute.

Effect of zero torsional inertia

Shear deformations occur when rectangles become parallelograms under shear load. The amount
of deformation depends on the proportion of the cross-sectional area that is effective in resisting
shear - in I beams only the web is effective, whereas the whole area contributes in rectangular
beams. Program help files often give guidance regarding this proportion - often called a ‘shear
factor.’

Bending and shear deformation

WSP Group
TRM 41
TECHNICAL REFERENCE Rev 3

MANUAL Date 9/05


Civil & Structural
Page 10 of 15
COMPUTER MODELS OF STRUCTURES

Shear flexibility may often be ignored in longer beams because its effects are often very small
compared to bending flexibility. Most programs recognise that a shear factor of zero indicates that
shear deformation is to be ignored.

If the structure is made of concrete it is sometimes necessary to use the properties of the cracked
section, allowing for the transformed area of reinforcement. Because the amount of cracking
depends on the applied bending moment, an iterative approach is required. Most codes (eg
BS 8110-1, BS 5400-4) allow the properties of the gross section (ignoring reinforcement) to be
used and this simplification is often acceptable. However, it may seriously under-estimate
deflection and create a less accurate distribution of bending moments and shear forces (the
concept of redistribution of support moments into the spans is a crude method of allowing for this).

Young’s modulus (E) values

Structural analysis programs usually have a built-in library of materials which have been assigned
default values for Young’s modulus. Generally the defaults can be used without modification, but
in some cases special values may need to be assigned. For example, the Young’s modulus of
timber depends on the species and grade used.

For concrete structures a further complication arises when long-term behaviour is being modelled,
as creep increases deformations; the effect is usually modelled by reducing the E-value. This is
explained in TRM 109 Movements in concrete structures. Most structures carry a mixture of short-
term and long-term loads and so an intermediate value should be chosen. If different parts of the
structure carry different ratios of dead and live load it will be necessary to assign different modulus
values to the different elements.

LOADS

Loads must be applied carefully. Check that the individual figures and the totals are correct and
that the loads are applied in the right direction. The sign convention in some programs is that
gravity loads are positive, in others they are negative. Check that the vertical axis in the model
corresponds to the vertical direction assumed by the program. Some programs assign loads at
90º to the member (“local axis loading”) by default and for sloping members this may not be
correct - “global axis” loading should be specified instead. Programs usually allow loading to be
displayed visually and this permits an easy check to be made.

If the model is linear - that is, if it does not contain tension-only, compression-only or other non-
linear elements - results can be combined to give factored moments, shears and the like.
However, if the model is non-linear the loads must be combined before the model is solved or the
results will be incorrect.

WSP Group
TRM 41
TECHNICAL REFERENCE Rev 3

MANUAL Date 9/05


Civil & Structural
Page 11 of 15
COMPUTER MODELS OF STRUCTURES

Load combinations in non-linear structures

Remember that the worst combination for, say, moment may not be the worst combination for
shear as well. Some programs, when enveloping output, will correctly list the maximum moment
together with the concurrent shear force. Others merely list the maximum moment and the
maximum shear, even though they derive from different load cases. If you simply take the worst
results in each case to design an element you may be providing more strength than is strictly
required.

PRINTING AND INTERPRETING THE RESULTS

General

When all the geometrical and loading data have been supplied to the program the model can be
solved. If the program offers to save the matrix, do not do so - stiffness matrices use large
amounts of disk space and the saving in time in future runs is rarely significant. Error messages
are often displayed as the program is run. These are usually caused by the model - or part of it -
being a mechanism. The program may state that it has restrained the model in some way to
“resolve” the problem. If this happens do not use the results. It is unlikely that the program will
have restrained the model in a way that will reflect the reality on site. Return to geometry
definition, find the problem and correct it.

Then view the results critically on the computer screen. It is worth spending some time on this
check, because errors found later - when members have been designed and drawings have been
produced - will be more costly to correct. Are the deflections as anticipated? How do they
compare with approximate hand calculations? Is one part of the structure deflecting more than the
rest, and if so is this to be expected? Similarly, do the reactions, moments, shears and axial
forces appear sensible? Is the total of the reactions equal to the sum of the applied loads in each
load case? If the model is complex it may be necessary to inspect different parts separately.

Beam elements

Programs typically give moments, shears and axial forces at each end of beam elements. Where
results within the element are larger than at the ends, some - but not all - programs will give the
larger result too. If the program does not provide intermediate results, split the beam into two or
more elements to provide nodes near the critical locations. Shears and moments corresponding to
the local axis directions and the sign of the results will be explained in the manual or help file. If
the explanation is unclear, run a simple model to see what the convention is.

WSP Group
TRM 41
TECHNICAL REFERENCE Rev 3

MANUAL Date 9/05


Civil & Structural
Page 12 of 15
COMPUTER MODELS OF STRUCTURES

Finite elements

Finite element results can be difficult to interpret, although the program manual or help file should
provide an explanation. The program will usually provide bending moments about two axes and of
course for design purposes it is essential to ensure that the axes are interpreted correctly, or
structural strength may be designed at 90º to the direction in which it is required.

Most programs provide an option to display “Wood-Armer” moments, which utilise a set of
equations devised by Mr Wood and Mr Armer. These combine bending moments in two directions
with torsion moments to provide a design moment that can be used - for reinforced concrete - to
determine the area of reinforcement needed in each of two chosen directions. Note that, because
of the effects of torsion in the slab, the Wood-Armer moments sometimes require both top and
bottom reinforcement.

Large numbers of other types of result can usually be displayed. It is essential to understand the
program manual or help file in order to interpret the results properly. For example, the program
may display stresses parallel to the X axis in the top face of the elements, but this information
cannot be used unless you understand the assumptions made by the program about the direction
of the X axis, and which face of the element it assumes is uppermost.

Finite element models often show very large stresses at internal corners, supports and other
significant changes of stiffness. These should be reviewed carefully, but if the material is ductile
and the load is not cyclic, the peak values can often be reduced substantially for design purposes.
This is because the areas of high stress are small, and a minor amount of local yielding, crushing
or other plastic deformation will relieve the stress concentration.

General advice

If the model is large, do not print all the numerical results. Graphical output of bending moment
diagrams, shear force diagrams, contours of stress and the like use less paper and reveal peak
results (and errors) far more readily. There are often options allowing the user to “zoom” in to
show only part of the structure, or to display the results for some beams in a certain view and not
others, or to plot the results diagrams to a convenient scale.

VERIFICATION OF COMPUTER CALCULATIONS

The Standing Committee on Structural Safety (SCOSS) have highlighted the need for guidance on
the use of computers with the shift of emphasis of engineering design from detailed calculations
towards computer modelling. This has led on occasions to computer assisted errors and in the
worst cases to structural failures. These guidance notes set out the approach to be taken when
using computers for structural analysis. It is assumed that validation of the software has been
completed within the procedures set out in Project Life Cycle. The use of spreadsheets in design
has become a simple popular means of designing elements, but these are not to be used unless
they have been validated independently of the author.

Verification records should be maintained covering the whole process of analysis from the
inception of the structural model, through the input data through to the implementation of the
results. The level of verification will depend upon several factors, including the complexity, the
degree of innovation and the level of experience, and should be agreed with the project director
and recorded in the project plan.

WSP Group
TRM 41
TECHNICAL REFERENCE Rev 3

MANUAL Date 9/05


Civil & Structural
Page 13 of 15
COMPUTER MODELS OF STRUCTURES

The various levels of verification may be considered as:


• Self-assessment checking by the engineer.
• Detailed or sample check by a second party within WSP.
• Independent assessment incorporating analysis using separate software.
• Independent assessment by an external engineer to WSP with appropriate experience and
track record.

The resources on the project will also have an impact upon the level of verification and should be
agreed with the project director at the inception of the project:
1. Has the designer sufficient experience and training to create a structural concept model
and develop it through the software package?
2. Is the structural model appropriate for the structural form being considered?
3. Are there any limitations in the software being considered to provide the analysis? And is
the software appropriate for the analysis?
4. Are there the skills and experience within WSP available to complete the verification at the
various stages?

The following should be considered as a minimum requirement for checking of computer outputs:
• Check the input data including loads and geometry of the frame.
• Consider the compatibility of the data electronically transferred into the analysis package.
• Have the correct material properties been selected?
• If eccentricities have been ignored will there be significant discrepancies?
• Understand how the loading conditions are applied by the program; simple matters such as
the self weight, has it been calculated by the computer?
• Are the applied loads and reactions balanced?
• Are the internal forces and moments in balance?
• Use simple calculations to 10% of the members to estimate the magnitude of forces and
moments, investigate further if the results are not within 20% of the output.
• Draw upon previous experience and rules of thumb for simple structures, eg refer to the
WSP C&S designer pack together with published safe load tables.
• Have the warnings or error messages been resolved satisfactorily?

These notes are not meant to be definitive and each structure will need to be considered on its
merits at the concept review with the project director.

CASE STUDY OF INAPPROPRIATE MODELLING

By way of example this is an instance where things did not go to plan with the analysis of a simple
frame which resulted in significant remedial costs. The project was a three storey reinforced
concrete frame designed using a 3D computer model (the geometry cannot be shown to protect
confidentiality).

1. Long span beams which were supported at the tips of cantilever beams at right angles were
under-designed. The model assumed the cantilever beams to have high torsional resistance,
which led to the span moments in the long span beams being reduced by fixed-end moments
from the cantilever beams. In reality the beams did not have the ability to develop the torsion
moments. Inspection of the moments by an experienced engineer would have identified the
shortcomings.

WSP Group
TRM 41
TECHNICAL REFERENCE Rev 3

MANUAL Date 9/05


Civil & Structural
Page 14 of 15
COMPUTER MODELS OF STRUCTURES

2. The floors were precast concrete designed by others. They were modelled into the computer
assuming an equivalent thickness diaphragm; this resulted in the floor loads being distributed
in two directions, when the precast floors could only span in one direction. This resulted in
certain beams being under-designed with others over-designed. A simple arithmetic check to
balance loads would have identified the error. We have a responsibility to integrate designs by
others into our model.

3. The model was assuming uncracked sections which coupled with the above grossly
underestimated the deflections; this became clear during the construction. In this case a
simple span to depth ratio check would have flagged up there was going to be a problem.

4. Somehow the computer was indicating a staircase area remote from columns and walls was
attracting significant loads at each level. The contractor decided not to cast the staircase with
the floors as the engineer did not specify this requirement. As engineers we have a
responsibility to transfer knowledge of our designs to others when they may have an impact on
the method of working on site.

This is a case of misuse of a computer model in the extreme and disregard for simple verification
that would have avoided the problems encountered on site.

SUMMARY

• Use the right program.

• Read the manual, understand the program’s limitations.

• Do sensitivity tests if appropriate.

• Provide an explanation in the calculation file of the assumptions made and the data used.

• Decide whether beam elements or finite elements are most suitable.

• Produce a model of the right size - possibly by building a big (but coarse) model first and then
smaller more refined models to show areas of interest.

• Plan the model carefully - consider standardising node and beam numbering.

• Take great care with supports, restraints and springs.

• Make sure that beam positions and orientations are sensible.

• Use a sufficient number of properly sized and shaped finite elements.

• Check local axis orientation.

• Check section properties - use cracked section properties if appropriate.

• Use suitable values for Young’s modulus - allow for creep in concrete under long-term loading.

• Ensure load values and directions are correct and that combinations are properly defined.

WSP Group
TRM 41
TECHNICAL REFERENCE Rev 3

MANUAL Date 9/05


Civil & Structural
Page 15 of 15
COMPUTER MODELS OF STRUCTURES

• Record input data for the model and the applied loads.

• Correct errors discovered when the model is solved.

• Review the results critically - are the results as expected?

• Read and understand the manual’s advice on the interpretation of finite element results.

• Plan results printing so that the volume of paper is not overwhelming - but don’t miss vital
areas.

• Finally, don’t believe a word of it! Check all input and output for mistakes and unexpected
results. Only use the computer output when you are completely convinced that it is correct.

FURTHER READING

Modelling of steel structures for computer analysis, Steel Construction Institute, 1995.

Guidelines for checking computer analysis of building structures, CIRIA Technical Note 133, 1988.

Powerful analysis software needs skilled users, The Structural Engineer, v 78 n 12, 20 June 2000,
pp 11-13.

Second order analysis of building and industrial structures, The Structural Engineer, v 79 n 3,
6 February 2001, pp 19-21.

IStructE Guidelines for the use of computers for engineering calculations 2002.

Biennial reports from SCOSS.

KEYWORDS

Analysis; case study; computers; finite elements, grillages; models; validation; verification.

Author: John Parker, WSP Cantor Seinuk


Sponsor: Group Technical Centre
Revision record
6/01 Rev 2.
9/05 Rev 3, section on checking computer calculations added (John Brighton, GTC)

WSP Group

You might also like