of the South African Institution of Civil Engineering Volume 59 Number 2 June 2017

of the South African Institution of Civil Engineering

Volume 59 No 2 June 2017 ISSN 1021-2019

Publisher
South African Institution of Civil Engineering
Block 19, Thornhill Office Park,
Bekker Street, Vorna Valley, Midrand, South Africa
Private Bag X200, Halfway House, 1685, South Africa
Tel +27 (0)11 805 5947/48, Fax +27 (0)11 805 5971
http://www.saice.org.za
civilinfo@saice.org.za

Contents
Editor-in-chief
Prof Gerhard Heymann
University of Pretoria
Tel +27 (0)12 420 3627
gerhard.heymann@up.ac.za
joint Editor-in-chief
Prof Chris Clayton
University of Southampton
2 Characterisation of rigid polyurethane foam‑reinforced
c.clayton@soton.ac.uk ballast through cyclic loading box tests
MANAGING Editor
Verelene de Koker R F du Plooy, P J Gräbe DOI: 10.17159/2309-8775/2017/v59n2a1
Tel +27 (0)11 805 5947, Cell +27 (0)83 378 3996
verelene@saice.org.za
11 Fatigue behaviour in full‑scale laboratory tests of
journal editorial Panel
Prof G Heymann – University of Pretoria a composite deck slab with PBL reinforcement
Prof CRI Clayton – University of Southampton
Prof Y Ballim – University of the Witwatersrand P Lu, X Zhan, R Zhao DOI: 10.17159/2309-8775/2017/v59n2a2
Prof W Burdzik – University of Pretoria
Dr P Day – Jones & Wagener (Pty) Ltd
Prof J du Plessis – University of Stellenbosch 19 The influence of foundation stiffness on the
Prof GC Fanourakis – University of Johannesburg
Prof M Gohnert – University of the Witwatersrand
behaviour of surface strip foundations on sand
Prof PJ Gräbe – University of Pretoria
Dr C Herold – Umfula Wempilo Consulting H E Lemmen, S W Jacobsz, E P Kearsley DOI: 10.17159/2309-8775/2017/v59n2a3
Prof A Ilemobade – University of the Witwatersrand
Prof SW Jacobsz – University of Pretoria
Prof EP Kearsley – University of Pretoria
28 Revised regional sediment yield prediction
Prof C Roth – University of Pretoria methodology for ungauged catchments in South Africa
Prof W Steyn – University of Pretoria
Mr M van Dijk – University of Pretoria V C Msadala, G R Basson DOI: 10.17159/2309-8775/2017/v59n2a4
Prof JE van Zyl – University of Cape Town
Prof C Venter – University of Pretoria
Prof A Visser – University of Pretoria 37 The two-points condensation technique (TPC) for
Dr E Vorster – Aurecon South Africa (Pty) Ltd
Prof J Wium – University of Stellenbosch detection of structural damage due to vibration
Prof A Zingoni – University of Cape Town
Prof M Zuidgeest – University of Cape Town A N Al-Qayyim, B Ö Çağlayan DOI: 10.17159/2309-8775/2017/v59n2a5
Peer reviewing
The Journal of the South African Institution of 49 The application of continuous surface wave testing
Civil Engineering is a peer-reviewed journal
that is distributed internationally for settlement analysis with reference to a full-scale
Design and reproduction
Marketing Support Services, Ashlea Gardens, Pretoria
load test for a bridge at Pont Melin, Wales, UK
Printing G Heymann, J Rigby-Jones, C A Milne DOI: 10.17159/2309-8775/2017/v59n2a6
Fishwicks, Pretoria
Papers for consideration should be submitted 59 Effect of the minimum void ratio on the vertical
online at: http://journal.saice.org.za
intercept of the steady state line of non-plastic soils
The South African Institution of Civil Engineering accepts no
responsibility for any statement made or opinion expressed L A Torres-Cruz, S Geyer, P R Mackechnie DOI: 10.17159/2309-8775/2017/v59n2a7
in this publication. Consequently, nobody connected with
the publication of this journal, in particular the proprietor,
the publisher and the editors, will be liable for any loss or
damage sustained by any reader as a result of his or her action
upon any statement or opinion published in this journal.
© South African Institution of Civil Engineering

Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017 1
TECHNICAL PAPER Characterisation of rigid
Journal of the South African
Institution of Civil Engineering
polyurethane foam‑reinforced
ISSN 1021-2019
Vol 59 No 2, June 2017, Pages 2–10, Paper 1486 ballast through cyclic
RUDOLPH DU PLOOY obtained a BEng degree
loading box tests
in Civil Engineering from the University of
Pretoria. He followed that up with a BEng (Hons)
in Transportation Engineering in 2014 and an R F du Plooy, P J Gräbe
MEng (Transportation) in 2016, also from the
University of Pretoria.
As train speeds and heavy haul axle loads constantly increase due to market demands, so do the
stresses and strains experienced by track structures. This is especially true for track transitions
Contact details: that generate high dynamic forces on both the track and vehicles because of poor vertical
Transnet Freight Rail Chair in Railway Engineering track geometry and/or differing track stiffness values on either side of the track transition.
Department of Civil Engineering
Reducing differential settlement between the two track structures at a track transition is one
University of Pretoria
Pretoria 0001 method of improving the life of the track, and increasing maintenance intervals. In this study,
South Africa rigid polyurethane foam was used to reinforce ballast. Tests were conducted using a dynamic
T: +27 83 288 7669 hydraulic load frame and a single sleeper in a large ballast box subjected to heavy haul axle
E: rudidu@gmail.com loads. Unreinforced, reinforced and 50% reinforced ballast layers of 300 mm depth were tested
to approximately 5 000 000 load cycles. The results showed that rigid polyurethane foam-
PROF HANNES GRÄBE (Pr Eng, FSAICE) is a civil reinforced ballast exhibited in the order of 60% less settlement for a fully reinforced layer, and
engineer with experience in the fields of track
42% less settlement for a half reinforced layer. The use of rigid polyurethane foam (RPF) to
technology, geotechnology, advanced
laboratory testing, field investigations, reinforce ballast has a number of benefits that could result in better track geometry and longer
maintenance models and numerical analysis of maintenance cycles, in turn resulting in lower life cycle costs.
track structures. He is Associate Professor:
Transnet Freight Rail Chair in Railway
Engineering at the University of Pretoria, where
INTRODUCTION sample material tests were also conducted
he lectures under- and post-graduate courses in railway engineering. He is
also responsible for railway research, as well as continuing professional The objective of this study was to study to compare the material behaviour of rigid
development (CPD) in the form of short courses presented to industry. the settlement behaviour and other track polyurethane foam and ballast reinforced
parameters through long-term, large-scale, with rigid polyurethane foam.
Contact details:
Transnet Freight Rail Chair in Railway Engineering cyclic loading box tests subjected to heavy
Department of Civil Engineering haul axle loads (30 tons). It was hypothesised
University of Pretoria that the reinforcement of ballast using rigid BACKGROUND
Pretoria 0001
polyurethane foam would lead to a reduction The most widely used track structure world-
South Africa
T: +27 12 420 4723
in ballast layer settlement, an increase in wide is the conventional or ballasted track
E: hannes.grabe@up.ac.za stiffness, reduced ballast breakage and an structure. Typically, it consists of two main
overall improvement in track performance. parts – the superstructure (i.e. rail, fastening
Two main objectives of the study were, system and sleepers) and the substructure
firstly, to characterise the properties of the (i.e. ballast, subballast and subgrade) as
ballast, polyurethane foam and the ballast- defined by Selig and Waters (1994).
polyurethane-foam composite material and, Ballastless track structures are track
secondly, to characterise the settlement structures that have been developed to
behaviour of the ballast and polyurethane mitigate the problems relating to the slow
foam-reinforced ballast layer. deterioration and subsequent settlement of
Only a ballast layer was modelled in the the ballast material as a result of traffic load-
box test with no other foundation layers. ing. Ballastless track structures are typically
Various test configurations were prepared, divided into two main categories, namely
ranging from completely unreinforced bal- ballastless track that provides continuous
last to fully reinforced ballast. Each sample support to the rail, and ballastless track
was subjected to 5 000 000 load cycles at the that provides discrete support to the rail.
specified load. Ballastless track structures can result in
Settlement prediction models were maintenance costs that are 20% to 30% lower
compared to actual ballast settlement behav- than ballasted track (Esveld 2001).
iour, and the validity of these settlement Track transitions occur where a bal-
prediction models for use with polyurethane lasted track section changes to a ballastless
reinforced ballast samples was examined. track system or a ballasted track system
Resilient sleeper deflection, ballast layer on a structure. The abrupt change in track
Keywords: ballast, concrete sleepers, ballast settlement, ballast breakdown, stiffness and ballast layer strain were all support that occurs at these locations has
rigid polyurethane foam, cyclic loading determined and compared. A number of often been associated with accelerated rates

2
Du Plooy RF, Gräbe PJ. Characterisation of rigid polyurethane foam-reinforced ballast through cyclic loading box tests.
J. S. Afr. Inst. Civ. Eng. 2017:59(2), Art. #1486, 9 pages. http://dx.doi.org/10.17159/2309-8775/2017/v59n2a1
of track geometry and component degrada- 1
tion, which in turn can lead to poor ride A A d95i
quality and increased maintenance demand BBI =
(Read & Li 2006). When maintenance at B A+B

these track sections is neglected, acceler- d max

ated deterioration in track geometry can be PSD = particle size distribution
2.36 = smallest sieve size
expected. Agreement in the literature exists d95i = d95 of largest sieve size
as to the possible mechanisms that affect the
differential movement at track transitions.

ge
The problems at track transitions can be

ka
ea
attributed to three primary factors (Sasaoka

br
um
Fraction passing
& Davis 2005):

im
ax
■■ Differential settlement

fm
■■ Differences in settlement characteristics

yo
ar
■■ Discrepancies in track damping proper-

nd
ou
ties between adjacent sections.

yb
ar
Nicks (2009) categorised remedial measures

it r
rb
aimed at reducing track transition impact

forces as follows:
■■ Limit approach track structure settlement.
■■ Decrease modulus on bridge deck/ballast­
less track structure.
■■ Increase the modulus for the approach
track.
■■ Reduce ballast wear and movement.
■■ Increase damping on bridge deck/ballast-
less track structure.
When ballast has been in service for a sig-
nificant period of time, it becomes damaged
and contaminated, resulting in a change in
the grading of the ballast which in turn leads
to a reduction in performance. The process
through which this occurs is known as
fouling. Selig and Waters (1994) list the five
causes of ballast fouling as:
■■ Ballast breakdown
■■ Infiltration from ballast surface
■■ Sleeper wear
■■ Infiltration from underlying granular layers
■■ Subgrade infiltration.
Ballast fouling prevents the ballast from
fulfilling its functions, and the extent of this
loss in function is dependent on the amount
and size of the fouling material. As the mass
of sand and fine-gravel-sized fouling particles
(0.075 mm to 19 mm) increases, the resilience
to vertical deformation, as well as the void
space, decreases. Fouled ballast is gap-graded.
This results in a reduction in drainage and
could make surface lining (i.e. tamping)
operations difficult. As the void space is filled,
the density of the ballast material increases.
Should this material then be tamped, a higher
rate of ballast settlement can be expected
after tamping. An increase in the mass of clay
and silt-sized particles (particle sizes smaller
than 0.075 mm) also results in reduced drain-
age, leading to ballast erosion and subgrade
attrition. When mixed with water, fine
particles may form a slurry that is abrasive to
the ballast material. Fouled ballast with high
water content leads to higher rates of plastic
strain in the fouled ballast.
A
0
0 2.36
Sieve size (mm)
Figure 1 Definition of ballast breakage index (Indraratna et al 2005)
Ballast breakdown can be quantified
by using a parameter known as the Ballast
Breakage Index (BBI), and the ballast breakage
can be quantified by comparing the plots of
particle size distributions. As the number
of breakages increase, the particle size dis-
tribution shifts to the left of the graph (see
Figure 1), and the area between this new line
and the original particle size distribution is
considered as the breakage zone. Indraratna
et al (2005) state that the breakage potential is
considered to be the area between the original
particle size distribution and an arbitrary
reference line connecting the point between
the intersection of d95 of the largest sieve size
and the minimum particle size of 2.36 mm.
Figure 1 shows the definition of the BBI
parameters which are used in Equation 1.
A overall trend was represented by the power
BBI = (1)
A+B
The permanent deformation behaviour of bal-
last under cyclic loading is usually in the form
of settlement. The settlement behaviour can
be both elastic (such as initial settlement due
to compaction) and plastic (due to breakage of
ballast particles). The number of load cycles
also has an effect on the permanent deforma-
tion, with the permanent deformation being
a function of the logarithm of the number of
cycles (Shenton 1974).
Initial PSD
Final PSD
53
The variation in settlement behaviour
between uncompacted and compacted
ballast samples was compared, with the
uncompacted ballast settling significantly
more than the compacted ballast, along
with the uncompacted ballast having greater
first cycle axial strain development (2.8%)
compared to 0.58% for the first cycle of the
compacted ballast. Each additional cycle of
loading caused an increment of plastic strain,
but at a diminishing rate, varying from 0.24%
after the first cycle to 0.001% after approxi-
mately 50 cycles (Indraratna et al 1997).
A number of equations have been pro-
posed to describe the settlement behaviour
of ballast. Selig and Waters (1994) compared
a number of settlement prediction equations
and found that, of the semi-log, hyperbolic,
parabolic and power relationships, the best
relationship of the form shown in Equation 2
and the logarithmic relationship shown in
Equation 3.
SN = S1 N b(2)
SN = S1(a log N + 1)  (3)
Where:
SN = settlement after cycles
S1 = settlement after the first cycle
a and b are shape parameters.

Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017 3
Granular materials, such as railway ballast,
experience problems over poor formations,
due to the development of tensile strains
at the formation interfaces. The continued
rotation of the principle stresses causes bal-
last densification, and hence settlement. This
increases over soft formations as a result of
low track stiffness, which then induces bal-
last plastic strains. One method of improving
the ballast layer, and reducing the amount of
plastic strain and settlement is to reinforce
the ballast layer (Woodward et al 2009).
A method of reinforcing or stabilising the
ballast layer is the use of polyurethane foam.
Keene et al (2012a) conducted a study into
the characterisation of rigid polyurethane
foam (RPF) reinforced ballast. RPF is an
expanding polyurethane which is placed
in the ballast layer (either through pouring
or injection) and is then allowed to fill the
ballast layer voids, preventing the infiltra-
tion of water and other contaminants. The
RPF would further protect the structure
from rearrangement and settlement. RPF
is a cross-linked, closed-celled, thermoset
material with a low density, and a number of
the RPF material’s properties are dependent
on the density. RPF is typically supplied in
two separate components – an A-component
(polyether polyol) and a B-component (poly-
isocyanate) are then mixed and applied.
Keene et al (2012b) summarised the
mechanical property results of the work
on polyurethane stabilised ballast (PSB) as
follows:
■■ PSB outperforms other track-substruc-
ture materials.
■■ PSB typically had higher elastic deforma-
tional behaviour.
Furthermore, Keene et al (2012b) concluded
the following with regard to the feasibility
of using polyurethanes to stabilise track
substructure:
■■ Stabilisation does not have a negative
impact on elastic response.
■■ Injection methods that are cur-
rently employed are feasible for track
stabilisation.
■■ PSB can greatly increase the track mecha-
nistic life cycle.
Keene et al (2013) conducted numerical
modelling into the effect of polyurethane
stabilisation on rail track response. The
numerical model was used to determine the
effects of various parameters, such as loca-
tion, thickness and polyurethane properties,
on the ballast layer and how resilient behav-
iour is influenced. A larger range of PSB
moduli than observed in the laboratory was
also incorporated into the numerical model.
The simulations by Keene et al (2013)
showed that there are minimal changes
in the strain of each substructure layer
4 Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
(i.e. ballast, subballast and subgrade) and no
negative effect on the overall elastic response
under loading due to the change in stiff-
ness of the polyurethane reinforced areas.
Specific material properties and structural
components of the track system appear to
have far greater influence on substructure
elastic strain and track modulus than the
integration of polyurethane reinforcement
into the ballast layer.
Keene et al (2012b) report that, while
rigid-compact polyurethane in rail infrastruc-
ture has been used, very few experimental and
empirical methods have been developed for
ascertaining the mechanical properties and
life cycle characteristics of rail substructures
that have been stabilised with polyurethane.
The investigation into the injection of RPF
into the rail substructure is an uncharted
area. Standard laboratory tests for the fabrica-
tion and characterisation of polyurethane
stabilised ballast (PSB) are aimed at contribut-
ing to research infrastructure. The overall
objective is to reduce maintenance life cycle
costs, increase rail freight load capacity and
provide maintenance techniques that are not
disruptive to railroad traffic.
EXPERIMENTAL WORK
The characteristics of the foam and ballast
materials that were subjected to cyclic load-
ing box tests and the details of the experi-
mental work are described in the following
sections of the paper.
Material properties
The rigid polyurethane foam that was used to
reinforce the ballast was Elastopor® H 1311/1.
The foam is supplied in two separate com-
ponents, namely a Polyol-component (A) and
an Iso-component (B). The polyol component
is a mixture of polyetherpolyolls, stabiliser,
catalyst, flame retardant and water. The iso-
component is polymeric diphenylmethane
Figure 2 B
 allast box test setup before being placed into the hydraulic load frame
diisocyanate (IsoPMDI 92140). The material
properties of the rigid polyurethane foam
(RPF) and the ballast material were deter-
mined prior to testing. An initial grading
analysis and a loose bulk density determina-
tion were conducted on the ballast. RPF
cylinders, and RPF reinforced ballast cylinders
were cast and their load-deflection behaviour
was tested in a universal testing machine. The
ballast material properties are summarised
in Table 1, with a summary in Table 2 of the
various foam material properties.
Table 1 Ballast material properties
Ballast Material Quartzite
ACV (%) 14.5
Water absorption (%) 0.3
Loose bulk density (kg/m3) 1 511
Void ratio (loose) 0.857
Porosity (loose) 0.461
Table 2 Rigid polyurethane foam properties
RPF density (kg/m3) 36.9
PRB density (kg/m3) 1 135.5
RPF Compressive strength 138 kPa – 140 kPa
Mixing ratio (A:B) 100:107
Cyclic loading ballast box tests
A hydraulic MTS load frame with a maxi-
mum actuator capacity of 500 kN was used
to apply the cyclic loading to the ballast
samples. A large steel box was constructed,
with internal dimensions of 2 400 mm long,
600 mm wide and 400 mm high. Each test
consisted of a 300 mm ballast layer with a
PY-sleeper for ballast loading. Linear variable
differential transducers (LVDTs), with a full
scale of 20 mm, were mounted onto the test
frame at each sleeper end for local sleeper
displacement measurement in addition to the

Figure 3 T est apparatus and test setup
Table 3 Test record summary
Test Type
1 Reinforced ballast
2 Reinforced ballast
3 Unreinforced ballast
4 Reinforced ballast (50%)
Figure 4 F ully reinforced ballast layer sample (upside down) after testing, showing full void filling
by RPF
Comment
Free expansion
Expansion limited
– 10
Expansion limited
Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
internal actuator displacement measurement.
The setup of a single sleeper on a 300 mm
ballast layer in a steel box, simulating a sec-
tion of track, is shown in Figure 2.
Support column Each test was divided into primarily two
phases, namely the initial phase and the
consolidation phase. During the initial phase,
Crosshead large settlement values were recorded as a
result of the settlement of the uncompacted
Load frame swivel/hinge
ballast and the tests had to be paused to
Hydraulic actuator readjust the LVDTs to their measuring range.
The loading during the initial phase consist-
Load cell
ed of a number of sub-phases where the load
Loading beam
and loading frequency were incrementally
LVDT for sleeper displacement
measurement increased during the initial 57 000 cycles.
Loading block The final test setup is shown in Figure 3.
LVDT support frame The second phase of the test consisted of
Ballast box the consolidation phase where a sinusoidal,
cyclic 260 kN load (30 ton/axle loading) was
applied to the sleeper for a total of 5 000 000
cycles. The magnitude of this cyclic load
was obtained using the AREMA Manual of
Railway Engineering standard specification.
Load frame test bed A maximum rail seat load of 60% of the
axle load is recommended for a pre-stressed
Hydraulic ram to raise/lower
crosshead concrete sleeper at 760 mm sleeper spacing
(Figure 30.4.1 in AREMA 2010). A dynamic
factor of 1.46 was applied to the static load of
147 kN. A total of four tests were conducted,
all at a test frequency of 10 Hz. The complete
test record, summarising the various tests, is
shown in Table 3.
During initial sample preparation it was
Frequency (Hz) Cycles (No)
discovered that, without restricting the
10 5 058 308 movement of the ballast, the foam would
expand and push the ballast particles apart.
10 5 040 036
In Test 1, the 300 mm ballast layer was
4 787 750 reinforced using RPF, and the expansion of
the ballast and foam was not limited after
10 4 967 957
pouring. During the preparation of Test 2,
the expansion of the RPF in the ballast layer
was limited by the placement of a sleeper on
top of the ballast layer during the pouring
process. Test 3 was a ballast test without
any reinforcement applied. Test 4 was
conducted using 50% reinforcement (i.e.
bottom 150 mm reinforced and top 150 mm
unreinforced).
Figure 4 shows the complete void filling
of the RPF of the ballast layer to produce the
reinforced ballast composite structure.
RESULTS AND DISCUSSION
Ballast settlement
The settlement results from the initial phase
of testing are shown in Figure 5. A significant
amount of settlement occurred in the initial
phase of testing, due to the ballast layer being
uncompacted at the start. From Figure 5 it
can be observed that the unreinforced ballast
(UR) experienced the greatest amount of
5
settlement, as the particles were free to settle
and undergo reorientation. The reinforced 0

ballast (RB) test, where the foam was allowed

to freely expand, showed similar initial set-
–10
tlement values, only exhibiting 3.5% less set-

Initial settlement (mm)

tlement than the unreinforced ballast during
the initial phase of testing. Reinforcing half –20
of the ballast layer (bottom 150 mm) resulted
in a reduction in initial settlement of 54%.
Limiting the expansion of the RPF reinforced –30
ballast layer by placing a sleeper on the top of
the layer during the pouring process resulted
–40 RB – Free expansion
in the lowest initial settlement (33.8 mm) RB – Expansion limited
which represents 34% of the settlement Unreinforced
50% RB – Expansion limited
observed in the unreinforced ballast test. –50
Following the initial phase, the main
(consolidation) phase of the test was initiat- 10 0 101 102 103 10 4 105
ed. This phase consisted of the full test load Cycles (log scale)
being applied for 5 million load applications
at a frequency of 10 Hz. During the initial Figure 5 I nitial phase settlement of the polyurethane stabilised ballast samples
phase of testing, the unreinforced ballast test
produced the highest settlement. However,
as shown in Figure 6, during the main phase –0 RB – Free expansion
of testing the unreinforced ballast layer RB – Expansion limited
settled the least (3.7 mm). The reinforced Unreinforced
–2 50% RB – Expansion limited
ballast layer with free expansion showed
Consolidation settlement (mm)

the largest amount of settlement during

–4
the main phase, settling a total of 12.9 mm.
This is most probably as a result of the lack
of inter-particle contact between the ballast –6
stones. This lack of inter-particle contact can
be explained by the fact that, during the RPF –8
curing process, expansive forces were gener-
ated and these forces were significant enough –10
to lift the ballast stones. Only once the foam-
filled voids between the ballast stones were
–12
sufficiently compressed, was inter-particle
contact restored.
0 1 × 10 6 2 × 10 6 3 × 10 6 4 × 10 6 5 × 10 6 6 × 10 6
Fully reinforcing the ballast layer resulted
Cycles
in a total settlement of 4.3 mm, while the
ballast layer with 50% reinforcement set-
Figure 6 M
 ain phase settlement of the polyurethane stabilised ballast samples
tled 4.9 mm. The unreinforced, reinforced
and 50% reinforced ballast layer tests all
produced settlement values within 1.3 mm of
0
one another. RB – Free expansion
RB – Expansion limited
The two settlement test phases were com- Unreinforced
bined to obtain a total settlement value. The –10 50% RB – Expansion limited
largest combined settlement was observed in
the test with RPF reinforced ballast with free
Total settlement (mm)

–20
expansion, where a total combined settle-
ment of 63.0 mm was observed. The unrein-
–30
forced ballast layer exhibited the second larg-
est combined settlement of 55.6 mm, the vast
majority of which occurred during the initial –40
phase (92.9%). Figure 7 shows the combined
settlement results, with the 50% reinforced –50
ballast layer showing the second lowest total
settlement (32.0 mm). The reinforced ballast –50
layer with the RPF expansion limited showed
the lowest combined settlement, with a total 0 1 × 10 6 2 × 10 6 3 × 10 6 4 × 10 6 5 × 10 6 6 × 10 6
settlement of 22.3 mm. Cycles
The use of RPF as a means of reducing
ballast settlement is a possibility, provided Figure 7 T otal settlement results of the polyurethane stabilised ballast samples

6 Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
 70 that the expansion of the foam into the voids
of the ballast does not result in significant
60 ballast particle uplift which could remove
the inter-particle contact between the stones.
50
Settlement (mm)

By reinforcing 50% of a ballast layer, the total

40 settlement experienced by the layer can be
reduced by 58% compared to unreinforced
30
ballast. Comparing the completely reinforced
20 ballast layer (expansion limited) with the
50% reinforced layer, the former produced a
10 45% reduction in total settlement. A partially
foam-reinforced layer would have the benefit
0
Unreinforced 100% RB 100% RB 50% RB of retaining the drainage, maintainability
Free expansion Expansion limited Expansion limited and other ballast characteristics that are lost
Test when fully reinforcing the ballast layer.
Initial settlement Consolidation settlement Although some of the reinforced tests
showed more settlement during the main
Figure 8 F inal comparison of settlement results phase of the test, the large reduction in
initial consolidation settlement should
also be noted as a primary contributor to
0 the reduced total settlement. Allowing
Actual settlement the free expansion of the polyurethane
Power function foam reinforcement should be limited, as
–1 Log function
increased settlement in all phases of testing
Consolidation settlement (mm)

was observed and was even outperformed

by the unreinforced ballast. For use in the
–2 field, the foam should be poured or injected
SN = 0.0963N0.2551 into an already existing track structure in
SN = 0.0963(7.4749 log N + 1)
order to prevent the occurrence of ballast
–3 uplift from the foam expansion. It should
be mentioned that clean ballast would be an
essential requirement to allow penetration
–4 of the foam into the ballast layer. When bal-
last uplift or foam expansion is limited, the
ballast would perform significantly better.
–5 Reinforced ballast exhibits lower settlement
0 1 × 10 6 2 × 10 6 3 × 10 6 4 × 10 6 5 × 10 6
values and, as a result, the ability of the
Cycles
track to maintain good geometry in the field
can be expected. The result of this could
Figure 9 S emi-reinforced settlement results compared with logarithmic and power prediction
be longer maintenance intervals and lower
functions
dynamic forces at track transitions as a result
of the reduction in differential settlement. A
0 final comparison of the total settlement and
the relative contribution of each phase to the
Actual settlement
–5
combined settlement are shown in Figure 8.
Power function
Log function Results are shown for unreinforced ballast
Combined total settlement (mm)

(UR), reinforced ballast (RB), free expansion

–10
(FE) and expansion limited (EL) samples.
The actual settlement results for each test
–15
SN = 4.019N0.135
were also modelled with the relationships
SN = 4.019(1.040 log N + 1) described in Equations 2 and 3. The actual
–20 settlement data was compared to power
and logarithmic functions. The results from
–25 these comparisons for the semi-reinforced
ballast test are shown in Figure 9 (consolida-
–30 tion phase) and Figure 10 (combined initial
and consolidation phase).
–35
The coefficient of correlation values (R2)
0 1 × 10 6 2 × 10 6 3 × 10 6 4 × 10 6 5 × 10 6 6 × 10 6 for the power functions used to theoreti-
Cycles cally predict the total settlement of each
test were significantly lower than the R2
Figure 10 S emi-reinforced combined total settlement results compared with logarithmic and values for the logarithmic prediction func-
power prediction functions tions. For the total settlement R2 values, the

Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017 7
power functions were not adequate and the Table 4 Comparison of R2 values for different test phases (consolidation and combined phases)
logarithmic prediction function gave better
Initial and
correlation. The lowest R2 value for the total Initial Phase
Consolidation Phase
Test
settlement logarithmic prediction functions
Power Logarithmic Power Logarithmic
was 0.77, and for the same test the R2 value
for the power function was 0.22. A summary Fully reinforced – free expansion 0.81 0.63 0.38 0.88
of these results is provided in Table 4.
Fully reinforced – expansion limited 0.77 0.43 0.64 0.96
From these results, and from those in the
previous section, it would appear that when Unreinforced 0.78 0.31 0.22 0.77
there is significant initial settlement in the
Semi-reinforced 0.97 0.42 0.29 0.88
early load cycles (such as in uncompacted
ballast), the logarithmic prediction function
would be the ideal relationship to use, as it
provides the closest estimate in these cases.
For test samples where initial settlement 220
is not that significant, the power function
tended to provide superior results. For this 200
reason, the prediction function chosen
should be selected based on the present
Resiliant modulus (MPa)

180
testing conditions and behaviour of the
test sample. 160

Ballast layer resilient modulus 140

The ballast layer resilient modulus (Er) was
calculated by dividing the applied ballast 120
stress by the ballast layer strain for that
RB – Free Expansion
cycle. The stress directly below the sleeper 100
RB – Expansion limited
was approximately 390 kPa, assuming a
Unreinforced
sleeper area of 2 200 mm x 300 mm and a 80
50% RB – Expansion limited
load of 260 kN. The ballast layer resilient
modulus results can be seen in Figure 11. 0 1 × 10 6 1 × 10 6 1 × 10 6 1 × 10 6 1 × 10 6
The unreinforced ballast layer had the Cycles
highest initial modulus, with a modulus
of 184 MPa. The resilient modulus of the Figure 11 R
 esilient modulus results of the polyurethane stabilised ballast samples
unreinforced ballast layer increased up
until 500 000 cycles, after which a gradual
100
decrease in stiffness can be observed, most
Initial grading
likely as a result of ballast breakdown, caus-
Left side
ing the ballast layer to become fouled and
Middle
leading to a loss in strength. 80
Right side
The partially reinforced ballast layer
Maximum breakage line –
(50% reinforcement) showed an increase Christie et al (2007)
Percentage passing (%)

in modulus throughout the duration of the 60

test. The other two fully reinforced bal-
last tests showed lower initial moduli than
the partially reinforced and unreinforced
40
ballast tests. The 50% reinforced ballast
sample commenced with an initial modulus
of 121 MPa and this increased to 186 MPa
during the course of the test. At the end of 20
the test, the 50% reinforced ballast sample
had a modulus that was approximately equal
to that of the unreinforced ballast sample 0
(186 MPa vs 189 MPa respectively). 0 10 20 30 40 50 60 70 80
The completely reinforced ballast sample Nominal sieve aperture (mm)
that had the polyurethane foam expansion
limited, achieved a 41% increase in resilient Figure 12 S emi-reinforced (50%) ballast grading test results
modulus during the course of the test.
Even the test where the polyurethane foam In the case of the fully reinforced bal- between the ballast stones becoming fully
reinforcement was allowed to freely expand last, the resilient modulus of the layer compressed, allowing the ballast stones
(which performed very poorly with regard gradually increased with the increase in to regain contact with one another. The
to settlement) achieved a 72% increase in load cycles. This increase in modulus could final modulus values of the fully reinforced
modulus over the course of the test. be as a result of the small foam-filled gaps ballast layers were found to be significantly

8 Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
 1.4
1.3
1.2
Resiliant sleeper deflection (mm)
1.1
1.0
0.9
0.8
0.7
0 2
Figure 13 R
 esilient sleeper deflection and ballast resilient modulus as a function of test frequency
lower than those of the unreinforced and
50% reinforced ballast layers. Interestingly,
the unreinforced ballast layer modulus
values of the first and final loading cycles
showed only a 2.8% difference, whereas all
the reinforced ballast samples showed sig-
nificant modulus gains of at least 40% (fully
reinforced expansion) to 72%, as in the case
of the fully reinforced ballast layer with
unlimited foam expansion.
Ballast breakdown
The ballast breakage index was calculated
for the unreinforced and partially reinforced
tests. In Figure 12 it can be seen that the
grading of the partially reinforced ballast
sample falls outside the theoretical maxi-
mum ballast breakage line. This could be as
a result of the significantly thinner ballast
layer, which had a depth of 150 mm. The
fact that the ballast below this thin layer was
reinforced with a rigid polyurethane foam
could also have played a role in the ballast
breakage behaviour.
The partially reinforced ballast layer had
a BBI of 3.3%, and the unreinforced ballast
layer had a BBI of 3.8%, indicating that
there was no significant difference in ballast
breakage between the partially reinforced
and unreinforced ballast tests.
Testing frequency comparison
A test was conducted in which the 260 kN
cyclic load was applied across a range of fre-
quencies from 1 Hz to 15 Hz at 1 Hz incre-
ments. At each frequency, a constant testing
time (60 seconds) was applied, resulting in a
variation in the number of cycles from 60 to
900 for the 1 Hz to 15 Hz frequency range.
4 6 8 10 12 14
Frequency (Hz)
Resilient sleeper deflection Stiffness
The effect of the change in frequency was
examined by comparing the sleeper deflec-
tion and settlement per cycle. An increase in
the resilient modulus of the ballast samples
of 22% was observed over the frequency
range (1 Hz to 15 Hz). The resilient deforma-
tion and resultant ballast resilient modulus
trends as a function of testing frequency are
shown in Figure 13.
CONCLUSIONS
The following main conclusions were
reached during the course of this study:
■■ The use of rigid polyurethane foam
as ballast reinforcement is suitable for
reducing settlement in track structures.
The significant reduction in total set-
tlement of ballast layers reinforced with
rigid polyurethane foam could result in
better long-term track geometry stability.
As a result, a longer track life cycle can be
expected when using rigid polyurethane
foam as ballast reinforcement.
■■ Rigid polyurethane foam-reinforced bal-
last settled 60% less than conventional
unreinforced ballast, and reinforcing
only 50% of a ballast layer led to a reduc-
tion in settlement of 42% compared
to that of conventional unreinforced
ballast.
■■ In the case of rigid polyurethane foam-
reinforced ballast, the ballast resilient
modulus (Er) increased as the number of
cycles increased (25% increase in modulus
for the 50% reinforced ballast layer) and
20% increase for the layer that was fully
reinforced with foam expansion limited.
This increase in ballast resilient modulus
Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
120 with time could be used to aid engineers
in the design of track transitions or as
a solution to problem track transition
110 sections.
■■ The power and logarithmic function
settlement prediction equations can be
Resiliant modulus (MPa)
100 used for predicting the settlement of
reinforced ballast samples. Theoretical
logarithmic prediction functions were
90 more suitable for predicting settlements
in tests with large initial settlement.
In contrast to this, power functions
80
provided a more suitable method for pre-
dicting settlement tests with small initial
settlement values.
70
■■ The effect of different test frequencies
on the behaviour of the ballast with
60
regard to sleeper deflection per loading
16 cycle, and the resulting ballast resilient
modulus in the range of 1 Hz to 15 Hz
were investigated. An increase in the
resilient modulus of the ballast samples of
22% was observed over the 1 Hz to 15 Hz
frequency range.
In conclusion, the use of rigid polyurethane
foam as a means of reinforcing railway
ballast has numerous benefits. A reduction
in total settlement was achieved. The com-
posite rigid polyurethane foam and ballast
material produced a layer that resulted in a
structure that decreased the resilient sleeper
deflection and increased the ballast resilient
modulus values, with an increase in load
applications. The behaviour of the foam and
foam/ballast composite material was quanti-
fied, and the settlement and load deflection
behaviour of unreinforced and reinforced
ballast was characterised.
Acknowledgements
Transnet Freight Rail is gratefully acknowl-
edged for financial support to the Chair in
Railway Engineering at the University of
Pretoria. The authors also wish to thank
Mr Johan Scholtz and Mr Derek Mostert of
the Department of Civil Engineering at the
University of Pretoria for their assistance
with the laboratory tests, specialised equip-
ment and writing of this paper.
REFERENCES
AREMA (American Railway Engineering and
Maintenance-of-Way Association) 2010. AREMA
Manual of Railway Engineering. Lanham, MD:
AREMA.
Esveld, C 2001. Modern Railway Track. 2nd ed.
Zaltbommel, Netherlands: MRT Publications.
Indraratna, B, Ionescu, D, Christie, D & Chowdhury,
R 1997. Compression and degradation of railway
ballast under one-dimensional loading. Australian
Geomechanics, 32(December): 48–61.
9
Indraratna, B, Lackenby, J & Christie, J 2005. Effect
of confining pressure on the degradation of ballast
under cyclic loading. Geotechnique, 55(4): 325–328.
Keene, A K, Edil, T B, Tinjum, J M & Brown, R W
2012a. Characterization of polyurethane-stabilized
ballast. Paper presented at the 3rd International
Conference on New Developments in Soil Mechanics
and Geotechnical Engineering, Nicosia, North
Cyprus.
Keene, A, Tinjum, J, Tuncer, E & Brown, R 2012b.
Railway substructure stabilization with polyurethane
injections. Paper presented at the Mid-Continent
Transportation Research Forum 2012, 6–7
September 2012, Madison, WI.
10 Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
Keene, A, Tuncer, E, Fratta, D & Tinjum, J 2013.
Modeling the effect of polyurethane stabilization on
rail track response. Proceedings, Geo-Congress 2013,
3–7 March 2013, San Diego, CA.
Nicks, J 2009. The bump at the end of the railway
bridge. PhD thesis, College Station, TX: Texas A&M
University.
Read, D & Li, D 2006. Design of track transitions.
Transportation Research Cooperative Program:
Research Results Digest, 79(October): 1–36.
Sasaoka, C & Davis, D 2005. Implementing track
transition solutions for heavy axle load service.
Proceedings, AREMA Annual Conference, 25–28
September 2005, Chicago, IL.
Selig, E T & Waters, J M 1994. Track Geotechnology and
Substructure Management, 1st ed. London: Thomas
Telford Publications.
Shenton, M J 1974. Deformation of railway ballast
under repeated loading conditions. In: Kerr, A D
(Ed.), Railroad Track Mechanics and Technology, 1st
ed. Proceedings of a Symposium held at Princeton
University, NJ, 21–23 April 1975.
Woodward, P, Kennedy, J & Medero, G 2009. Three-
dimensional track reinforcement using polymer
geocomposites. Proceedings, Annual Convention
of the American Railway Engineering and
Maintenance-of-Way Association (AREMA),
Chicago, IL.
Fatigue behaviour in TECHNICAL PAPER
full‑scale laboratory tests Journal of the South African
Institution of Civil Engineering
of a composite deck slab ISSN 1021-2019
Vol 59 No 2, June 2017, Pages 11–18, Paper 1298

with PBL reinforcement PROF PENG-ZHEN LU is an Associate Professor in

the Faculty of Civil Engineering and
Architecture, Zhejiang University of Technology,
P Lu, X Zhan, R Zhao Hangzhou, China. He obtained his PhD from
Southwest Jiaotong University. He has a broad
interest in structural analysis and design,
Studies on the fatigue behaviour of composite deck slabs are relatively few. To assess the fatigue encompassing numerical modelling and
experimental work. His major current interest is
performance of a composite deck slab at specific design loads, and to provide a reference for
the modelling of the linear and nonlinear behaviour of composite bridges. A
its design in fatigue, two full-scale models A and B of a composite deck slab were developed, parallel development is composite structural application, including
comprising steel plates and steel-fibre-reinforced concrete slabs. These models will be a useful composite beams and pre-stressed composite beams, especially the
reference for experiments and design, and for developing codes. In this study we carried calculation theory and analytical method for composite box beams. He also
out fatigue experiments and focused on the fatigue performance of a composite deck slab has an interest in specially shaped bridges, bridge evaluation and bridge
strengthening, particularly as there are many old bridges in China.
in a column area, and the positive and negative bending moments. For the entire fatigue
loading cycle, the overall performance of Models A and B was good, the overall stiffness of the Contact details:
composite deck slab was rarely attenuated, the stress levels in the steel members in relation Faculty of Civil Engineering and Architecture
Zhejiang University of Technology
to the fracture strength were not significant, and the steel member was in the stage of flexible
Hangzhou 310014
work. Through comprehensive tests of Models A and B it was found that the original design PR China
exhibits good fatigue performance and meets the design requirements. The research results T: +86 0571 88320153
provide a basis for the design of a composite bridge deck slab in fatigue. E: pengzhenlu@zjut.edu.cn

PROF XIAOLI ZHAN is an Associate Professor in

INTRODUCTION
Perforated-stiffened-plate composite bridge
deck slabs are a novel type of composite
bridge deck slab. At the bottom sides of the
slab, steel plates and concrete are connected
through a perforated, stiffened plate that is
a stud shear connector, a PBL shear connector the Faculty of Civil Engineering and
Architecture, Zhejiang University of Technology,
is more prone to fatigue damage and failure.
Hangzhou, China. She obtained her PhD from
Composite deck slab structures are still in the Harbin Institute of Technology. She has a
their early stages, as they have not reached the broad interest in road and bridge structural
ends of their life cycles yet; thus, they have analysis and design, encompassing numerical
not yet exhibited the effects of fatigue loading. modelling and experimental work.

known as a perfobond (PBL) shear connector.
A perforated-stiffened-plate composite bridge
deck slab combines the advantages of steel
and concrete in a general composite deck slab.
For bridge construction, a steel plate is used
as permanent formwork. This design saves
time during the installation and removal of
the scaffolding setup formwork, and reduces
the costs in relation to the construction cost
of the whole steel beam structure. The steel
plate and concrete slab are connected by PBL
shear connectors to improve the composite
effect of the slab, enhance the slab’s stiffness
and mechanical performance, and reduce
deformation and slippage. The composite
bridge deck slab – in conjunction with PBL
shear connectors – is widely used in the
construction of new bridges, maintenance
activities and the reconstruction of existing
bridges. Moreover, the composite slab has
technical and economic advantages, and
improves material performance and fast
construction compared to other promising
materials. However, despite the increasing use
of PBL-shear-connected steel and concrete
composite slabs, the bridges in which they are
used are not immune to the long-term effects
of moving loads. In particular, compared with
Since the more serious effects of fatigue load- Contact details:
ing have yet to appear, the fatigue problem has Faculty of Civil Engineering and Architecture
Zhejiang University of Technology
not been sufficiently addressed, nor systemati-
Hangzhou 310014
cally researched. Fatigue damage and failure PR China
decrease the reinforcement provided by the T: +86 0571 88320153
steel in a new bridge deck, which reduces the E: zhanxl@zjut.edu.cn
superstructure stiffness while increasing slip-
ping. These effects seriously affect the vehicle PROF RENDA ZHAO is at the School of Civil
capacity of the bridge structure. Therefore, Engineering, Southwest Jiaotong University,
China. He is also a vice-president of the China
an assessment of the fatigue performance of a
Institute of Civil Engineering. He obtained his
PBL-shear-connected steel and concrete com- BSc degree from Chongqin University, and his
posite slab at highway loads has important MSc from Southwest Jiaotong University. He has
academic and design potential, and will serve a broad interest in structural behaviour and
as a basis for future strategies for preventing experimental work.
or reducing damage due to fatigue loading. Contact details:
Few studies of perforated-stiffened-plate School of Civil Engineering
deck slabs have been published. Ryu et al Southwest Jiaotong University
Chengdou 610011
(2007) conducted a full-scale model test of
PR China
a two-span continuous bridge deck slab to T: +86 028 87601333
study the crack development of a fatigue- E: rendazhao@163.com
loaded composite deck slab. The results
indicated that the cracks in a region with a Keywords: steel truss arch bridge, steel-concrete composite deck slab,
negative bending moment were controlled fatigue performance
within an allowable crack width for a specific
fatigue load. The steel shuttering described
in their paper was a profiled sheet; compared
with steel-concrete bridge deck slabs, the
perforated stiffened plates had a different

11
Lu P, Zhan X, Zhao R. Fatigue behaviour in full-scale laboratory tests of a composite deck slab with PBL reinforcement.
J. S. Afr. Inst. Civ. Eng. 2017:59(2), Art. #1298, 8 pages. http://dx.doi.org/10.17159/2309-8775/2017/v59n2a2
arrangement. To verify the composite effect
of the steel shuttering and the perforated
steel plate combined with concrete, a test of a
full-scale model of a perforated-stiffened plate
deck slab was conducted by Kim and Jeong
(2006). Their results showed that the perfo-
rated and stiffened plate effectively enhanced
the composite effect. Their study primarily
analysed the effect of the perforated steel
plate for the entire composite bridge deck slab.
Leitão (2011) developed a numerical model for
the dynamic analysis of composite highway
bridges using a finite-element method with
mesh-refining technology, and achieved
satisfactory results when assessing fatigue
behaviour. To evaluate cracking in the vertical
and horizontal joints of a composite bridge
deck, Chang and Shim (2001) performed
fatigue behaviour tests of the composite
connection points of a continuous composite
bridge. Their study also discussed methods
of longitudinal pre-stressing. Although they
did not conduct a thorough analysis of the
effects of fatigue at the connection points on
the entire composite bridge deck slab, their
results provide a reference for the study of the
fatigue behaviour of perforated steel plates,
including the junctions of the underside
and side steel plates. Allahyari et al (2014)
investigated the behaviour of bridge decks
for a static load applied to the centre of the
deck. To evaluate the dynamic properties of
the decks, they experimentally investigated
the dynamic properties of exodermic bridge
decks with alternative PBL shear connectors.
Millanes et al (2014) investigated the design
of a composite steel-concrete deck for a long
railway bridge. Gara et al (2013) investigated
the effectiveness of various casting techniques
used to control the tensile stresses in the slab
during the construction of continuous steel
and concrete composite bridge decks. Leitão
et al (2013) carried out a fatigue analysis and
predicted the lifetime of composite highway
bridge decks under traffic loading conditions.
Wang and Jiang (2007) reviewed the
fatigue problem of composite structures and
the direction of future studies on composite
structures. They argued that future studies
should continue to focus on the mechanism
of fatigue damage accumulation, damage
identification, and fatigue reliability. Yang et
al (2012) performed mechanical performance
tests of a fatigue-loaded composite beam of a
composite bridge deck. Their results indicated
that the fatigue failure mode of the composite
beam specimens with a positive bending
moment was the crushing of the concrete
in the compression zone resulting from
fatigue damage to the lowest steel beam. The
fatigue life was directly related to the range
of fatigue stresses; however, the upper and
lower limits of fatigue loading had little effect
12 Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
on the fatigue life. Liu et al (2012) proposed
a new composite bridge deck system and
performed a theoretical analysis including a
full-scale model test of a bridge deck. Their
results showed that after 2 million cycles of
fatigue loading at a load frequency of 5 Hz,
no new cracks appeared in the composite
deck, nor had the initial cracks expanded.
Additionally, the stiffness of the test beam had
not decreased, indicating that this composite
bridge deck system had satisfactory fatigue
properties. Zong and Che (2000) performed
a fatigue test on a simply-supported continu-
ous pre-stressed composite beam. They also
ana­lysed pre-stressed composite beams
constructed from different types of concrete
at different magnitudes and orders of pre-
stressing. They drew some valuable conclu-
sions by summarising the fatigue test results
of the pre-stressed composite beams, discuss-
ing the maintenance of the steel members and
stud connectors, and proposing a principle for
the corresponding fatigue strengths.
For the fatigue of steel-concrete composite
bridge decks, the theoretical analyses and
experimental results provided by the above-
mentioned studies are in agreement regarding
fatigue-loaded steel-concrete composite
bridge decks. However, the individual behav-
iour of various combinations of steel-concrete
composite bridge decks is quite different, i.e.
different combinations of materials and forms
lead to different forms of fatigue behaviour.
Additionally, fatigue test results are discrete.
Therefore, many tests are needed to explore
the mechanical behaviour of perforated, stiff-
ened composite slabs. There are relatively few
studies of fatigue performance and the cor-
responding design for novel composite bridge
deck slabs. Many theoretical and technical
problems still have not been solved, such as
the failure mechanism, mechanical behaviour,
load-bearing capacity, and deformation of
novel composite bridge deck slabs under
fatigue-loading conditions. Therefore, con-
sidering the variety of steel-concrete bridge
decks and fatigue problems, the present study
aims to evaluate the fatigue performance of
composite bridge decks under highway load-
ing conditions using the Dongping Bridge in
Guangdong as a case study.
In this study, we conducted a model test
and simulation analysis to evaluate the fatigue
performance of a perforated stiffened deck
slab at a highway load. We also propose some
important indicators of fatigue performance,
provide a reference for composite materials
and forms, and assess their superior mechani-
cal behaviour to validate the effectiveness
of the proposed deck system for bridge
applications. Our study contributes to the
understanding of fatigue-induced damage or
failure of bridge deck structures. The results
Region of model experiment
Figure 1 E levation of bridge, bridge floor
system and location of model test
of this study have important theoretical value
and practical value for the optimisation,
design and theoretical analysis of composite
deck slabs.
TEST MODEL
Engineering background
for the fatigue tests
The main span of the Dongping Bridge
(43.5 m + 95.5 m + 300 m + 95.5 m + 43.5 m)
in Guangdong is a half-through steel truss
arch bridge with a full length of 1 322.2 m.
The steel boxes and concrete slabs are con-
nected by PBL shear connectors. The grid
beams consist of three main longitudinal
girders, secondary longitudinal girders, main
beams and secondary beams. Perforated-
stiffened-plate composite bridge deck slabs
were erected on the grid beams. Each slab
has a minimum thickness of 12 cm and a
maximum thickness of 20 cm. The overall
layout of the Dongping Bridge and the test-
ing zone of the composite bridge deck are
shown in Figure 1.
The Dongping Bridge uses a composite
deck-binding system with a space grillage
design. Under the sustained action of moving
loads, the longitudinal shear performance
of the perforated and stiffened composite
bridge deck slabs will be significantly
degraded, and the degradation in the per-
formance of the PBL shear connectors will
reduce the composite effect of the composite
bridge deck slabs. As a consequence, the
load-bearing capacity and stiffness of the
composite bridge deck may be reduced,
which will affect the mechanical behaviour.
Under such conditions, fatigue failure of the
composite bridge decks will occur once the
fatigue damage has escalated to a certain
point. To assess the fatigue performance of
the composite decks of the Dongping Bridge,
we performed fatigue experiments and a
numerical simulation in an area comprising
positive and negative bending moments.
Actual bridge model selection
Domestic and international research indi-
cates that the stress amplitude Δσ and the
 bending moments. The elevation and plan
Steel Headed stud PBL shear Concrete
bottom shear connector connector slab of the model structure described above are
shown in Figures 2(a) and 2(b).

Design of the experimental model

According to the study topics, two types of
Stiffened rib
models – denoted as Models A and B – were
constructed. Model A simulated the region
of the bridge structure with a negative bend-
ing moment, which includes a secondary
(a) Elevation beam and concrete slabs. Model B simulated
the region of the bridge structure with a
positive bending moment, which includes
B region of model experiment the area between two beams. Models A and
(positive region) B were constructed to perform full-scale-
model tests of the composite decks of the
Dongping Bridge.
The two models were constructed using
A region of model experiment
structural dimensions that were consistent
(negative region) with the actual bridge structure, except
for the plate length L and plate width B.
The geometry of Model B includes L and
B as the objects of study. According to the
experimental requirements and conditions,
L = 5 000 mm and 6 000 mm for Models
A and B respectively. The plate widths for
(b) Plan Models A and B were selected according to a
formula for the effective width of a concrete
composite beam slab (JTT 1986; BS 1980).
2 400
5 000 The calculated span of the actual bridge was
230 1 940 230
8 000 mm, the centre-to-centre distance of
the adjacent secondary beams was 3 325 mm,
230

p p
p and the top flange of the secondary beams
Loading position
was 706 mm.
2 400
1 940

p The calculated width of the concrete

slab was determined in accordance with the
230

comparative results of the following four

principles:
2 500 2 500
■■ The first principle is based on the cal-
(c) Cross-section and loading position of Model A (mm) culated width of the concrete slab, the
sum of the width of the top flange and 12
times the thickness of the flange, and the
2 400 6 000 minimum value of one of the three fol-
lowing items: a third of the span, the clear
1 200 1 200 Loading position distance between two subbeam plate
p p p brackets, or 12 times the thickness of the
2 400

top flange of the subbeam.

2 350 1 300 2 350
(d) Cross-section and loading position of Model B (mm)
Figure 2 Design drawing of model test of composite bridge deck
number of cycles N are the predominant fac- Saint-Venant’s principle to relax the bound-
tors affecting fatigue strength. Considering ary condition, the longitudinal range of
that the bridge deck span in the column the column area and the transverse range
area is larger, the most unfavourable girder between two main girders were selected as
beam in the bridge deck in the column area the objects of study for a simplified analysis
was selected for the analytical model of an problem. The focus was the fatigue perfor-
actual bridge. Owing to the large span of mance of the composite deck slab in a col-
the column area of the bridge, according to umn area comprising positive and negative
■■ The second principle is based on the mini-
mum value of one of the three following
items: a third of the span, the centre dis-
tance of two subbeams, or the sum of the
width of the top flange and 12 times the
thickness of the top flange of the subbeam.
■■ The third principle is in accordance with
the CP117 specification based on the
minimum value of the following three
values: a third of the span, the centre
distance of two subbeams, or the sum of
the top width of the plate bracket and 12
times the flange thickness.
■■ The fourth principle is in accordance with
the BS 5400 specifications, with which
the size of the model design is calculated.

Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017 13
 According to comparative results of the
four principles, the size of Model B was
determined to be 2 146 mm. The size of
Model A is 2 400 mm × 5 000 mm, and the
size of Model B is 2 400 mm × 6 000 mm.
The cross-sections of Models A and B
and the loading positions are shown in
Figures 2(c) and 2(d).
DESIGN OF THE FATIGUE
MODEL TEST
Determination and theoretical
analysis of fatigue loads
The stress conditions of the actual concrete
plate were the focus of the present study. First,
we ensured that the stress conditions of the
top concrete plate were consistent with those
of an actual situation when we determined
the loading scheme of the test model by
theoretical calculations. The stress loading
was determined by theoretical calculation,
hence the stress conditions of the concrete
were verified for consistency with the actual
situation. To align the stress conditions of the
model with the actual structure, the upper
limit load P and its loading location were
determined. Vehicle loads of 20 t, 30 t and
55 t were used to calculate the fatigue vehicle
loading with an influence line. Considering
the effect of the deadweight of the bridge deck
pavement, the magnitude of the stress at the
lower limit fatigue load should be consistent
with the bridge deck pavement of the actual
bridge structure.
The theoretical results of the model
test were obtained using an ANSYS finite-
element model comprising spatial deck
elements and solid elements. The model
was based on 180 455 nodes and 207 65
elements. According to Table 1, the applied
fatigue vehicle load was determined and the
impact coefficient a = 0.295 was used. In the
transverse direction of the bridge, the most
unfavourable loading position was used, and
the position of the vertical loading was based
on the influence line of the control section
of the four-span continuous beams. The
finite-element model of Model B consisted of
31 250 nodes and 32 585 elements. The load-
ing scheme of Model B was not determined
until the stress conditions of the concrete slab
were consistent with the actual bridge condi-
tions. According to the comparative analysis
results for each fatigue vehicle load in Table
1, the vehicle load of 30 t was selected as the
control vehicle load for the actual bridge cal-
culation. The analysis was carried out by allo-
cating loads in a single lane. In the analytical
model, the impact coefficient was determined
to be 0.295, and the single wheel weight F on
the actual bridge rear axle was determined
14 Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
Table 1 Equivalent moment range (kN.m)
China’s
bridge design
Fatigue load BS5400 AASHTO
specifications
200 kN
The equivalent
bending moment 88.2 124.45 86.09
value
3
2
1
0
Stress (MPa)
–1
–2
–3
–4
–5
0 1
Stress of model test
Figure 3 E xperimental or numerical stress distribution of middle cross-section under upper limit
fatigue loading
to be 77.7 kN. According to the results of the
analysis above, for the same stress distribution
for Model B and an actual bridge under test
loading, through an extensive space model
analysis, the upper limit for the fatigue-
test design load Pp–max for Model B was
determined to be 84.10 kN, and the distance
between the loading points was 1 940 mm.
In addition, from the results of the analysis
above, the stress in the actual bridge structure
for a dead load is greater than that for Model
B under its own weight. During the model
test, the stress in the model at the lower limit
of the fatigue load was set to be equal to that
of the actual bridge at a constant load. Using
the same analysis method with the upper
limit of the fatigue-test design load, the lower
limit of Pp–min for Model B was determined
to be 2.74 kN, and the distance between the
loading points was 1 940 mm.
The stress distributions of the cor-
responding areas of the actual bridge and
Model B at the upper limit of the fatigue
vehicle load are shown in Figure 3. The lon-
gitudinal bending stress of Model B agreed
well with that of the actual bridge. According
to Figure 3, the compressive stress of the
actual bridge model was larger than that of
China’s China’s
bridge design bridge design BS5400
specifications specifications Axle load
300 kN 550 kN
120.30 154.50 68.36
2 3
(m)
Stress of actual bridge test
Model B within the longitudinal range of
0 mm to 350 mm. The longitudinal compres-
sive stress of the actual bridge was smaller
than that of Model B within the longitudinal
range of 2 000 mm to 2 400 mm. Further,
the longitudinal bending stresses of Model B
and the actual bridge were consistent, except
for the positions of loading and support.
Additionally, they met the requirements of
the model selection.
The stress distributions of the actual
bridge and Model B at the fatigue limit loads
are shown in Figure 4. According to Figure 4,
the stress distributions laws and peaks of
the longitudinal bending stress of Model B
and the actual bridge are consistent, except
for the locations of support and loading.
Additionally, they met the requirements of the
model selection. From a theoretical analysis,
the test design loads of Model B are listed in
Table 2, which indicates that the amplitude of
the fatigue load of Model B is 5.48–168.2 kN,
and the number of loading cycles is 2 million.
The size of Model B and its loading positions
are shown in Figure 2(d). The design idea of
Model A is similar to that of Model B. The
sizes of Models A and B and the number of
loading cycles are listed in Table 3.
 0.10 model was simplified as a four-span continu-
ous beam. The amplitude of the equivalent
bending moment of the middle cross-section
for the second span is listed in Table 1.
0.05
According to a comparison of the results for
different fatigue loads in Table 1, the 30 t
fatigue vehicle load specified by China was
Stress (MPa)

0 selected as the standard fatigue vehicle load

for the bridge model.

–0.05 Production and programme

design of the test model
The test model components consisted of
steel plates and steel-fibre-reinforced con-
–0.10
crete slabs. Model A had a length and width
of 5 000 mm and 2 400 mm respectively.
The thickness of the steel plate and the steel-
–0.15 fibre-reinforced concrete slab were 8 mm
and 120 mm respectively. PBL strips were
used in the shear connectors of the compos-
–0.20
ite bridge deck slabs. These connectors have
0 1 2 3 been widely used in composite structures to
(m) connect steel and concrete slabs. In this test
Stress of model test Stress of actual bridge test model, the PBL connectors were embedded
400 mm along the transverse direction of the
Figure 4 Stress distribution middle cross-section under lower limit fatigue loading concrete slab, and 15 PBL connectors were
used for the entire model. To strengthen the
Determination of real bridge Dongping Bridge, six fatigue loads based on joint between the steel and concrete slabs,
fatigue loads the BS 5400 fatigue vehicle load (BS 1980); the hole in each of the PBL strips contained
China has not yet selected a standardised the United States AASHTO fatigue vehicle steel rebar with a diameter of 12 mm, and
fatigue load spectrum for highway bridges, load (AASHTO 2005); the 20 t, 30 t, and the weld joints of the steel-plate structure
or established norms for selecting a vehicle 55 t fatigue vehicle loads specified by China were strictly in accordance with the relevant
for including fatigue loads. When deter­ (JTT 1986); and the BS 5400 axle loads standards. Furthermore, one half of the span
mining the load spectrum of the decks of the were compared. Moreover, the actual bridge of the weld joint of the PBL connectors and
the steel plate was intermittently welded,
Table 2 Design load of Model B whereas the other half was continuously
welded. A large tonnage jack and an MTS
Load of experiment Spacing between
Loading force servo system were used for the model test.
Load of actual bridge design loading points
(kN)
(kN) (mm) A data accumulator was used to collect and
Fatigue load (300 kN) + Fatigue ceiling Pp-max store data on a computer for analysis.
1 940 168.20
pavement’s self-weight (84.1) Owing to the complexity of the stress
distribution, a majority of the plates were
Fatigue ceiling Pp-max
Pavement’s self-weight
(2.74)
1 940 5.48 in a biaxial state of stress. Therefore, for
the strain gauges of the measuring points,
strain rosettes were used. A number of the
Table 3 Size and loading number for Models A and B strain gauges were fixed before the concrete
was poured, while the other strain gauges
Test model Load category Loading (kN) Number
were fixed after the model had been com-
Model B Fatigue load 5.48~168.2 2 million pleted. For the entire model, a total of 514
strain gauges were used, of which 370 were
Model A Fatigue load 3.33~63.52 2 million
attached to the steel structure surface, 46
Model A Fatigue load 3.33~78.28 100 000 were attached to the steel rebar surface, and
98 were attached to the concrete surface.
Model A Fatigue load 3.33~83.66 900 000
Displacements were measured by dial
gauges, with a total of 35 dial gauges arranged
Table 4 Measuring points for Models A and B around the test model. A constant-amplitude
sinewave load was used for the fatigue test.
Measure point
Measure point of strain The load frequency of the fatigue tester was
of displacement
Test model
7.0 Hz, the number of loading cycles was
Steel structure Reinforcement Concrete Dial gauge
2 million, and the amplitude of the fatigue load
Model A 370 46 98 35 was 5.48–168.20 kN. To monitor the stress
and deformation of the measuring points dur-
Model B 234 51 141 35
ing the fatigue test, data was obtained after the

Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017 15
following numbers of fatigue loading cycles: 16
10 000, 50 000, 250 000, 500 000, 1 million,
1.5 million and 2 million. The test programme
14
of Model B was similar to that of Model A.
The measuring points and numbers for
Models A and B are listed in Table 4. 12

10
FATIGUE TEST RESULTS

Stress (MPa)
Fatigue test results for Model A 8
The basis for the test number was the proxim-
ity to the maximum deflection of 0.645 mm 6
for a load of 83.66 kN during the static load-
ing stage. A deflection analysis shows that the
4
maximum deflection occurred at measuring
point #15 under all working conditions.
Measuring point #15 is located immediately 2
below a loading point. The maximum deflec-
tion of the composite deck was 0.515 mm 0
for a load of 63.52 kN after 2 million cycles. 0 20 40 60 80 100
It should be noted that after 2 million cycles (kN)
of fatigue loading, Model A had yet to show 0 2 million times 3 million times
evidence of fatigue cracking. Additionally, the 10 000 times 2.1 million times theoretical values
stiffness was not significantly reduced. To
further investigate the fatigue performance Figure 5 Curve of load stress for PBL
of Model A, a load of 78.28 kN was applied to
the model up to 100 000 fatigue cycles. After
3.5
100 000 cycles, the maximum deflection of
the composite bridge deck plate was 0.63 mm,
which is close to the maximum deflection 3.0
after 2 million cycles. Therefore, we increased
the fatigue limit load to 83.66 kN, and
again applied it up to 900 000 cycles. After 2.5

900 000 cycles, the maximum deflection was

0.65 mm, which is close to the maximum
Stress (MPa)

deflection (0.645 mm) at the static loading
stage (83.66 kN). The results showed that the
maximum deflection of the composite deck
was 0.63 mm for a load of 78.28 kN after
100 000 cycles. The maximum deflection of
the composite deck was 0.65 mm under a
load of 83.66 kN after 900 000 cycles loading.
Hence, the stiffness of the composite deck did
not decrease significantly.
A stress analysis showed that for a load of
83.66 kN after 900 000 cycles, the steel plate
surface, PBL connector and top concrete sur-
face had maximum principal tensile stresses
of 8.15 MPa, 14.28 MPa and 3.18 MPa, with
maximum principal compressive stresses of
11.01 MPa, 6.78 MPa and 0.81 MPa, respec-
tively. Each plate remained in an elastic
stress state, and none of the stress levels of
the plates were high, except for the concrete
slabs. Figure 5 shows the change in the maxi-
mum longitudinal stress in the shear con-
nectors with PBL strips based on a test using
different numbers of fatigue loading cycles.
After 2 million cycles, the measured stress
increased somewhat; however, the results
were mostly unchanged when compared with
those after 10 000 cycles.
2.0
1.5
1.0
0.5
0
0 20
0 2 million times
10 000 times
Figure 6 Curve of load stress for concrete slab top
Figure 6 shows the change in the maxi-
mum longitudinal stress of the top concrete
surface for different numbers of fatigue load-
ing cycles. Figure 6 shows that, after 10 000
cycles, the stress was greater than that before
fatigue loading; additionally, after 2 million
cycles, the measured stress was basically
the same as that after 10 000 cycles. Since
the effect of the upper steel rebar was not
considered in the theoretical calculation, the
40 60 80 100
(kN)
3 million times
2.1 million times theoretical values
ANSYS theoretical results were greater than
those of the static load test.
Testing indicated that after 10 000
cycles with an upper limit of 63.52 kN, four
longitudinal micro-cracks appeared in the
maximum transverse tensile stress zone of the
top concrete surface. Meanwhile, a transverse
crack appeared along the direction of the PBL
connector for 100 mm from the mid-span
of the model beam; the maximum width of

16 Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
 the longitudinal cracks was 0.04 mm. As
A–6 A–5 A–4 A–3 A–2 A–1 B–2 B–3 B–4 B–5 B–6
the number of fatigue load cycles increased,
new cracks continuously emerged, and the
existing cracks continuously expanded. The
19
widths of the longitudinal and transverse
22 25 cracks away from the mid-span section also
8 10
11 23 increased, whereas the transverse cracks near
24 26
2 7 14
the mid-span section did not increase in size.
13 12 16
15
28
29
17 After 2 million cycles, the maximum width
21
5 1 4 of the longitudinal cracks was 0.05 mm, and
3 20
27 the maximum width of the transverse cracks
was 0.10 mm. After 3 million cycles of fatigue
loading, the maximum width of the longitudi-
nal cracks was 0.055 mm, and the maximum
A’–6 A’–5 A’–4 A’–3 A’–2 A’–1 B’–2 B’–3 B’–4 B’–5 B’–6 width of the transverse cracks was 0.09 mm.
Details of the crack pattern of Model A are
Figure 7 Crack pattern detail for Model A shown in Figure 7.

Fatigue test results of Model B

0.9
The amplitude of the fatigue load for
Model B was 5.48–168.2 kN, and the number
0.8 of fatigue loading cycles was 2 million. The
deflection of the cross-section for 10 000,
0.7 1 million and 2 million cycles is shown in
Figure 8. The results indicate that the overall
0.6 stiffness did not change. The stress in the
PBL shear connector for different fatigue
Deflection (mm)

loads increased as the number of fatigue

0.5
loading cycles increased, and the stress
was lower. The changes in the stress of two
0.4
response points of the PBL shear connector
are shown in Figure 9. All slabs were in elas-
0.3 tic stress stage under fatigue loading, despite
the increase in the number of fatigue load-
0.2 ing cycles. In the fatigue loading stage, the
deflection at the maximum deflection point
0.1 for the model structure did not change, and
the stresses in the steel plate and concrete
0
plate were also lower. The strain data of the
0 500 1 000 1 500 2 000 2 500 fatigue test show that the cracks in tension
Longitudinal x direction (mm) are in a stable state and do not have any
10 000 times 1 million times 2 million times more widespread diffusion.

Figure 8 Deflection distribution of longitudinal x direction Models A and B: Comparative analysis

of the similarities and differences
The test programmes for Models A and B
14
were different owing to differences between
24 the structural characteristics of the two
12
66 model experiment regions. In Model A, the
10 steel I-beam mainly carried a negative bend-
ing moment, while in Model B, the region
Stress (MPa)

8 between steel I-beams primarily carried a

positive bending moment. Two million cycles
6 of fatigue loading were imposed onto the
Model A region with a load of 20 t. After
4 fatigue loading, 10 000 cycles of fatigue load-
ing at 30 t, 900 000 cycles of fatigue loading
2
at 30 t, and fatigue loading at 55 t were car-
0
ried out. In addition, for Model B, 2 million
0 50 100 150 200 250 cycles of fatigue loading at 30 t were loaded
Cycles (ten thousand times) directly in the positive bending moment
region. The results of the fatigue loading test
Figure 9 PBL stress load circle numbers for Models A and B show that after 2 million

Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017 17
cycles, the maximum deflection of Model A
was 0.515 mm, and the maximum deflection
of Model B was 0.8 mm. The deflection of
Model B was greater than that of Model A
after 2 million cycles. The stress level for all
types of loads for Model B was not greater
than when considering the stress at each
element in the model structure. For Model
A, none of the stress levels at the elements
were significant, with the exception of the
concrete slabs. However, the stress levels
were slightly greater than those for Model B.
Concerning the cracks, before carrying out
500 000 cycles of fatigue loading, the number
of cracks for Models A and B increased
linearly. After 2 million cycles, the number
of cracks for Model A increased slightly,
but gently. The cracks in the bottom of the
concrete slabs of Model B indicated that new
cracks continuously emerged. From the above
analysis, the test results for Model A were
different than those for Model B with regard
to the deflection, stress and cracks, because
of the different types of model study and the
fatigue loading, which led to differences in
the mechanical behaviour of the positive and
negative bending moment regions.
CONCLUSIONS
The mechanical behaviour of the Dongping
Bridge deck structure was studied using
experiments and finite-element analyses.
The following conclusions were reached after
evaluation of the experimental and finite-
element analyses of the positive and negative
bending moment regions of the composite
bridge decks.
■■ The test results for Model A revealed a
maximum deflection of 0.65 mm, a maxi-
mum principal tensile stress in the bottom
surface of the T-steel beam of 8.15 MPa,
a maximum principal tensile stress in
the PBL connector of 14.28 MPa, and a
maximum tensile stress in the longitudinal
rebar of 13.53 MPa. Additionally, the stress
level was not high in the steel structure,
with the entire structure remaining in the
elastic stress state. At less than 500 000
cycles, the number of cracks increased
linearly. However, after 500 000 cycles,
there was only a small increase in the
number of cracks. The overall stiffness of
the composite deck exhibited no apparent
attenuation after 2 million cycles of fatigue
loading, and the maximum width of the
longitudinal cracks in the top concrete
surface was 0.05 mm.
■■ The test results for Model B revealed a
maximum deflection of 0.80 mm, which
18 Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
was 19% greater than before loading.
However, the growth rate was only 2.6%
from 10 000 to 2 million cycles. This
indicated that the plate had good overall
performance, and its stiffness was not
attenuated after some initial loss of stiff-
ness. The stress levels across all slabs
were not high, and in the same sections
the horizontal and longitudinal strain
curves of the components appeared to be
consistent for different numbers of cycles.
During cyclic fatigue loading, the strain
in the steel increased with the number of
loading cycles, and cracks in the concrete
slab formed or expanded.
■■ The fatigue tests for Models A and B
indicated that the steel-concrete compos-
ite bridge deck of the Dongping Bridge
performed well overall, and the fatigue
performance met the Chinese Code for
the Design of Steel (GB 2003). Under
normal conditions, the composite bridge
deck of the Dongping Bridge will not
experience fatigue damage.
■■ The results of the fatigue tests of Models
A and B indicate that the state of the
loading levels are different for Models A
and B. Model A was located in a nega-
tive bending moment area. For a load of
83.66 kN and after 3 million cycles, the
maximum deflection of Model A was
0.65 mm, which is close to the deflec-
tion of 0.645 mm before cyclic loading.
Model B was located in a positive bending
moment area. For a load of 168.2 kN
and after 2 million cycles, the maximum
deflection was 0.80 mm, resulting in an
increase of 19% compared to the deflec-
tion of 0.67 mm before cyclic loading.
ACKNOWLEDGEMENTS
The authors gratefully acknowledge financial
support provided by the Science Foundation
of China (Grant No 2016M60035), the
Science Foundation of the Ministry of
Housing and Urban-Rural Development of
the People’s Republic of China (Grant No
2012-K2-6), the Science and Technology
Agency of Zhejiang Province (Grant No
2015C33222), the Science Foundation of
Shanghai (Grant No 13R21421100) and
Wenzhou City Science And Technology
Projects (Grant No G20140017).
REFERENCES
AASHTO (American Association of State Highway and
Transportation Officials) 2005. LRFD Bridge Design
Specifications, 3rd ed. Washington, DC: AASHTO.
Allahyari, H, Dehestani, M, Beygi, Morteza, H A, Neya,
B N & Rahmani, E 2014. Mechanical behavior of
steel–concrete composite decks with perfobond
shear connectors. Steel and Composite Structures,
17(3): 339–358.
BS (British Standard) 1980. BS 5400 Part 10: Code
of Practice for Fatigue. London: British Standards
Institution.
Chang, S P & Shim, C-S 2001. Continuous composite
bridges with precast decks. Steel Structures, 1:
123–132.
Gara, F, Leoni, G & Dezi, L 2013. Slab cracking control
in continuous steel–concrete bridge decks. Journal of
Bridge Engineering, 18(12): 1319–1327.
GB (Chinese Standard) 2003. GB 50017-2003. Code
for Design of Steel Structures. Beijing, China: China
National Institute of Standardization.
JTT (Chinese Industry Standard) 1986. JTT 025-86
Specifications for Design of Steel Structure and
Timber Structure Highway Bridges and Culverts.
Beijing, China: China Communications Press [in
Chinese].
Kim, H-Y & Jeong, Y-J 2006. Experimental investigation
on behaviour of steel–concrete composite bridge
decks with perfobond ribs. Journal of Constructional
Steel Research, 62(5): 463–471.
Leitão, F N, da Silva, J G S, da S. Vellasco, P C G,
de Andrade, S A L & de Lima, L R O 2011. Composite
(steel–concrete) highway bridge fatigue assessment.
Journal of Constructional Steel Research, 67(1): 14–24.
Leitão, F N, da Silva, J G S & de Andrade, S A L 2013.
Fatigue analysis and life prediction of composite
highway bridge decks under traffic loading. Latin
American Journal of Solids and Structures, 10(3):
505–522.
Liu, M, Shao, X, Zhang, Z & Hu J 2012. Experiment
on flexural fatigue performance of composite
deck system composed of orthotropic steel deck
and ultra-thin RPC layer. Journal of Highway and
Transportation Research and Development, 29(10):
46–53.
Millanes, F, Bordo Bujalance, E, Mansilla Dominguez,
J L & Martin Suarez, J 2014. Archidona Viaduct:
Composite steel–concrete deck for a long railway
bridge. Structural Engineering International, 24(1):
122–126.
Ryu, H-K, Kim, Y-J & Chang, S-P 2007. Crack control
of a continuous composite two-girder bridge with
prefabricated slabs under static and fatigue loads.
Engineering Structures, 29: 851–864.
Wang, P & Jiang, S 2007. Overview of fatigue damage
for composite structure (bridge deck). Journal of
Harbin Institute of Technology, 39(2): 694–698.
Yang, Y, Zhou, X & Xue, J 2012. Experimental study
on fatigue behavior of composite girders with steel
plate–concrete composite bridge decks. China Civil
Engineering Journal, 45(6): 123–131.
Zong, Z & Che, H 2000. Fatigue behavior of pre-
stressed composite steel–concrete beams. Journal of
the China Railway Society, 22(3): 92–95.
The influence of TECHNICAL PAPER
foundation stiffness on Journal of the South African
Institution of Civil Engineering
the behaviour of surface ISSN 1021-2019
Vol 59 No 2, June 2017, Pages 19–27, Paper 1423

strip foundations on sand EDUARD LEMMEN, who is a member of the

South African Institution of Civil Engineering
(SAICE), received his BEng in Civil Engineering in
H E Lemmen, S W Jacobsz, E P Kearsley 2012 and an MEng (Structural Engineering) in
2015, both from the University of Pretoria. This
paper is based on research conducted for his
The objective of this study was to determine whether the contact stress distribution underneath Master’s degree. He is currently working as
structural engineer at EDS Engineering.
a strip footing on dense cohesionless sand can be predicted as a function of the properties of
the footing and that of the underlying sand. Together, the footing and the underlying sand form Contact details:
a foundation system which can be classified in terms of relative stiffness. Centrifuge model tests EDS Structural, Civil & Transportation Engineers
PO Box 33920
were conducted on seven aluminium footings of various depths and hence stiffness, and one
Glenstantia
reinforced concrete footing, during which the qualitative variation in contact stress distribution Pretoria
underneath the footing was captured with the use of Tekscan™ pressure sensors. 0010
It was found that, as long as the relative stiffness of the foundation classifies as “stiff”, the South Africa
contact stress distribution underneath the footing is approximately uniform. As the relative T: +27 12 991 1205
E: eduardlemmen@icloud.com
foundation stiffness is reduced, the contact stress underneath the footing edge as a percentage
of the maximum stress occurring underneath the column, reduces. In the case of a semi-flexible
PROF SW JACOBSZ is a member of the South
footing tested, zero pressure was measured underneath the footing edge in response to loading
African Institution of Civil Engineering (SAICE).
of the footing. An expression is presented to allow the contact stress under the edge to be He graduated with an MEng (Geotechnical
estimated as a function of the relative stiffness of the foundation. This allows the contact stress Engineering) from the University of Pretoria in
distribution underneath semi-stiff footings to be estimated. 1996 and a PhD from the University of
The stiffness of the model reinforced concrete footing tested reduced considerably due to Cambridge in 2003. He is Associate Professor in
the Department of Civil Engineering at the
cracking during loading. It is recommended that footings be provided with sufficient stiffness
University of Pretoria, with 20 years of design
to classify as stiff. A modification of the benchmark value for the minimum relative stiffness and research experience in industry and academia.
classifying as stiff is proposed, based on the results of the centrifuge tests.
Contact details:
Department of Civil Engineering
University of Pretoria
INTRODUCTION and is difficult to use for modelling problems Pretoria
A foundation system consists of the combina- which approach the ultimate bearing capac- 0002
South Africa
tion of the soil supporting a structure and a ity of a foundation system. Also, it cannot
T: +27 12 420 3124
footing that spreads the imposed load of the take strain compatibility below and adjacent E: sw.jacobsz@up.ac.za
structure over a volume of soil large enough to the foundation into account, as it does not
to form a stable system. Aspects typically model lateral soil response. PROF ELSABÉ KEARSLEY is a Fellow of the South
considered during the design of a founda- Structural engineers often increase the African Institution of Civil Engineering (SAICE),
tion system are ultimate bearing capacity depth (thickness) of a footing to improve and she is currently Professor in Civil
and serviceability (i.e. an upper limit on the shear performance. However, there is no Engineering at the University of Pretoria. She
holds a PhD from the University of Leeds. She
acceptable settlement) (Aiban & Znidarčic limit to the extent to which the depth can be
worked as a structural design engineer in both
1995), with the settlement performance of the increased. To obtain the most economical South Africa and the United Kingdom before
system generally being the governing factor. foundation design, an optimum footing depth becoming a staff member at the University of
When a foundation system is designed, it is for a given soil stiffness can be determined. Pretoria. She was the 2009 President of the South African Institution of Civil
difficult to accurately determine the material The extent to which an increase in the Engineering (SAICE). For the last 23 years she has been involved with cement
and concrete materials research.
properties of the soil in the system, whereas footing depth would improve the perfor-
the properties of the materials used to con- mance of a foundation system was inves- Contact details:
struct the footing in a foundation system can tigated, as well as the effect of the overall Department of Civil Engineering
University of Pretoria
be quantified fairly accurately. foundation stiffness on the deformation of
Pretoria
To simplify the design of a foundation and stress distribution beneath the footing. 0002
system, structural engineers regularly The focus of this investigation was limited South Africa
model the material behaviour of the soil as to strip footings on dense cohesionless sand T: +27 12 420 2176
a system of linear springs. The linear spring underlain by a rigid base. E: elsabe.kearsley@up.ac.za

model delivers fairly accurate settlement

and deflection results when relatively small
strains are present within the soil mass
(Conniff & Kiousis 2007). This system can,
however, not accurately predict inelastic
material behaviour in response to loading,
BACKGROUND
The basis on which reinforced concrete
footings or slabs are designed was first pio- Keywords: strip footing, centrifuge modelling, sand,
neered by Westergaard (1925). Through the relative foundation stiffness

19
Lemmen HE, Jacobsz SW, Kearsley EP. The influence of foundation stiffness on the behaviour of surface strip foundations on sand.
J. S. Afr. Inst. Civ. Eng. 2017:59(2), Art. #1423, 9 pages. http://dx.doi.org/10.17159/2309-8775/2017/v59n2a3
study of concrete pavements he presented
a mathematical model to compute the
stresses in a concrete slab by assuming that
the slab acts as an isotropic, homogeneous,
elastic solid, and that the subgrade reaction
is perfectly vertical and proportional to
the size of the slab. He assumed that the
subgrade was an elastic medium, where
the stiffness of the subgrade could be
considered as the force which would cause
unit deflection if it were spread over a unit
area. Westergaard (1937) also proposed a
measure of the stiffness of the subgrade
compared to the stiffness of the founda-
tion, termed the radius of relative stiffness.
The work done by Westergaard has been
refined over the years through analytical
models and experimental procedures, and it
still remains the basis of a large number of
design procedures (Hemsley 2000).
For the design of footings from a struc-
tural engineer’s point of view, the footing
is often considered as a rigid body resting
on an elastic medium. This leads to the
assumption that the vertical settlement of
the footing on the elastic medium must
have a planar distribution, because a rigid
structure remains planar when it settles.
The planar distribution theory is based on
the assumption that the ratio of pressure to
settlement remains constant, and that the
stress distribution below the rigid footing
is uniform. However, these assumptions
are not necessarily true, as experimental
studies have shown that soil exhibits plastic
behaviour, footings have a finite stiffness and
the distribution of pressures below a footing
varies with time. The pressure distribution
in the soil is a function of the type of soil
(i.e. sand or clay), as well as the stiffness of
the footing and superstructure (Algin 2007).
Currently footing design is based on an
iterative process where the required size of
the footing often relies on the engineer’s
educated guess. The size of the footing is
then checked to see whether it satisfies
the serviceability conditions and that the
bearing capacity of the soil is not exceeded
(usually not critical). As soon as the size of
the footing has been confirmed as adequate,
the depth of the footing is determined to
ensure that it has adequate flexural, shear
and punching shear strength to resist the
applied loading conditions (Algin 2007). It
is assumed that sufficient structural depth
at the column will provide the footing with
adequate shear, diagonal tension, and bend-
ing moment capacity (Pisanty & Gellert
1972). The current design procedure for
surface footings does not take the stiff-
ness of either the footing or the soil into
account, but rather focuses on the strength
of the reinforced concrete and the allowable
20 Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
bearing capacity of the soil. The strength
and behaviour of the footing are also con-
sidered to remain constant regardless of the
applied load.
Footings are often analysed and designed
using the Winkler hypothesis, where the
soil is modelled as a continuous system of
non-connected discrete springs. Due to the
simplicity of the Winkler model, it is used
extensively to solve soil-structure interac-
tion problems and has produced satisfactory
results in many cases (Morfidis & Avramidis
2002). The Winkler approach yields relatively
accurate results if the structure does not
yield at any point. However, it becomes less
applicable if the load keeps on increasing
beyond yielding (Baumann & Weisgerber
1983). The Winkler method is a very basic
simplification of the behaviour of the soil
under loading conditions (Morfidis &
Avramidis 2002). As load transfer occurs
between the footing and the soil, the foun-
dation deforms, causing internal forces to
change within the structure, thus creating
a complex soil-structure interaction. The
significance of this interaction is important
for both dynamic and static loading condi-
tions. The most important aspect of the soil-
structure interaction is the contact stress
distribution beneath the footing (Conniff &
Kiousis 2007). The relationship between the
stiffness of the footing and the soil controls
the contact stress distribution beneath the
footing (Aiban & Znidarčic 1995; Arnold
et al 2010).
The structural behaviour of a footing can
largely be attributed to the stiffness thereof.
It is important to consider the stiffness of
the entire foundation system, and not just
the stiffness of the footing. Arnold et al
(2010) conducted a series of centrifuge model
experiments to determine the effect of the
system stiffness on the stress distribution
below the footing. From their analysis they
found that the stress distribution below the
footing is dependent on the stiffness of the
footing, as well as the footing settlement for
similar soil conditions. Whether or not a
foundation behaves in a flexible or rigid fash-
ion can be assessed using a dimensionless
parameter Ks that incorporates a representa-
tive Young’s modulus for the soil, combined
with terms representing the stiffness of the
footing in order to determine an equivalent
system stiffness (Canadian Foundation
Engineering Manual 2006). The stiffness
parameter Ks is calculated using Equation 1
(Arnold et al 2010). The stiffness of the sys-
tem can be classified based on the ranges for
Ks given in Table 1.
1 Eb d 3
Ks = (1)
12 Es l the behaviour of shallow foundation in the
Where:
Ks = System stiffness
Eb = Young’s modulus of the footing
Es = Secant modulus of the soil
d = Depth of the footing
l = Width of the footing
Table 1 S tiffness limits for foundation systems
(Arnold et al 2010)
Ks range System stiffness
0 Absolutely flexible
0–0.01 Semi-flexible
0.01–0.1 Semi-stiff
0.1–infinity Stiff
Arnold et al (2010) concluded that the
stress distribution below a footing on sand
is dependent on the magnitude of the load
applied to the footing. Initially the behaviour
of a footing represents that of a stiff system,
and changes to flexible with an increase in
the applied load. The use of a single system
stiffness parameter is therefore not adequate,
as it does not take the applied load into
account (Arnold et al 2010). The constant
stiffness typically assumed for a reinforced
concrete structure is also not valid, as its
stiffness is significantly influenced by non-
linearity in the concrete’s material properties
associated with cracking. The stiffness of a
footing changes when the concrete starts to
crack (Skorpen & Dekker 2014).
The aim of the current investigation was
to determine how changes in the stiffness
of strip footings affect the deformation of
such footings and the resulting contact stress
distribution underneath during loading. This
was investigated by testing a range of foot-
ings of various depths in a physical model
study. The study modelled strip footings
on dense sand. In further discussion in this
paper the term “footing” refers to the con-
crete slab bearing on the soil and the term
“foundation” refers to the combination of the
footing and the underlying soil.
EXPERIMENTAL SETUP
Description of centrifuge model
A physical model was developed in which
footings of various depths on dense sand
could be loaded under plane-strain condi-
tions. Due to the non-linear stress-strain
properties of soils, the model was tested in
a geotechnical centrifuge at the appropriate
acceleration to correct the stress distribution
to that of the full-scale prototype following
Schofield (1980). Physical models examining
 are similar (Knappett et al 2010). The only
Applied load
physical property of the aluminium other
IDTs than density necessary for this work is the
Young’s modulus which was measured
at 70 GPa.
Because the majority of structural foot-
ings comprise reinforced concrete, the varia-
tion in the pre- and post-cracking behaviour
of reinforced concrete complicates behav-
iour. It was therefore decided to carry out
Model footing one additional test on a reinforced concrete
150 mm model footing to compare behaviour with
that of the much more ductile aluminium
footings. The compressive strength of the
300 mm

concrete used for this footing was measured

as 30 MPa, the modulus of elasticity 25 GPa
and the measured flexural strength approxi-
mately 4.5 MPa. The footing was reinforced
with 19 steel wires of 0.71 mm diameter
Dense sand
spaced at 7.5 mm, spanning the width of the
footing. Their yield stress was measured at
400 MPa and the measured Young’s modulus
600 mm was 165 GPa. The reinforcement was placed
(a) Elevation A-A with 2.5 mm cover to the base of the footing.
The footing was 30 mm deep.

Instrumentation and actuation

A tactile pressure sensor mat by Tekscan™
(2014) was placed underneath each model
Strongbox side walls footing to record the contact stress during
tests. The model was observed from the side
through the glass window in the strongbox
Rigid divider
using a high-resolution digital camera. The
150 mm images recorded enabled displacement of the
sand and the footing to be tracked by means
of Particle Image Velocimetry (PIV) (White
et al 2003). In addition, the settlement of the
A IDTs A
Model footing footing and the adjacent sand surface were
150 mm

monitored using inductive displacement

Load application point
(b) Plan
Figure 1 Schematic elevation A-A (a) and plan (b) views of the centrifuge model
geotechnical centrifuge have been reported
by many researchers, e.g. Kusakabe (1995),
Brown et al (2004), Haigh et al (2010)
and others.
A model scale of 1:30 was selected, based
on practical considerations, and all models
were therefore tested at an acceleration of
30 g. It was intended to study the behaviour
of strip foundations. The model was set up in
a 150 mm wide compartment in a strongbox
equipped with a glass window through
which the foundation could be observed.
The length and height of the model compart-
ment were 600 mm x 400 mm respectively.
The depth of sand in the model was 300 mm.
The model is illustrated in Figure 1.
Model tests were conducted on seven
aluminium footings. All footings measured
150 mm wide, representing a 4.5 m proto-
type width, and 150 mm long to fit within
the model space shown in Figure 1. The
geometry of the model footings is illustrated
in Figure 2. The thickness of the aluminium
footing was changed during tests to change
the bending stiffness of the footings over a
range of relative foundation stiffnesses rang-
ing from semi-flexible to stiff (Arnold et al
2010). In addition, one test was carried out
on a scaled concrete footing.
Under elastic conditions aluminium is a
convenient substitute for concrete in certain
classes of physical models, as their densities
transducers (IDTs), as illustrated in Figure 1.
The footing was loaded by means of a
purpose-built actuator driven with a stepper
motor. The applied load was measured using
a load cell and the imposed displacement
by means of a displacement transducer
incorporated into the actuator. Load appli-
cation occurred in displacement control
during which the actuator moved at a rate of
approximately 0.01 mm per second. A maxi-
mum load of 30 kN was applied, exerting an
average contact stress of 1 333 kPa under the
footing, considerably exceeding the typical
allowable bearing capacity of sand applicable
to this situation of 600 kPa (see BSI 2015).
Sand properties
The experimental work described in this
paper was carried out using a fine silica from
a commercial source near Cullinan, east of
Pretoria. The properties were reported by
Archer (2014). The median particle size (D 50)
was 0.135 mm. The particle size distribu-
tion of the sand is presented in Figure 3.

Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017 21
The maximum and minimum relative
Plan view Side view
densities were determined using ASTM
D4253 (ASTM 1996a) and ASTM D4254
(ASTM 1996b) respectively at 1 669 kg/m3

75 mm
and 1 392 kg/m3. The specific gravity of the 16 mm
grains was 2.67.

150 mm
As discussed earlier, the structural behav- 14 mm
iour of a footing is dependent on the proper-
ties of the soil on which it is constructed. In
order to characterise the relative stiffness of
a foundation, a representative secant Young’s H
modulus is required for the soil (in this case,
the sand). Representative Young’s moduli for
each experiment were back-calculated from 150 mm 67 mm
elastic solutions by Milovic et al (1970) based
on the load-settlement response observed Figure 2 Model footing geometry
during the tests. Milovic’s solution applies
to rigid strip footings underlain by an elastic
100
layer of finite thickness and is given below:
90
P
E= ω 0(2)
ρz 80

70
Where: E is the Young’s modulus, P the load
Percentage passing (%)

intensity applied to the strip in kN/m, ρz the 60

observed settlement, and ω 0 a factor that
depends on Poisson’s ratio (assumed to be 50
0.2) and the ratio between the strip width
40
and the layer thickness.
The load-displacement curves observed 30
during loading of the model footings were
remarkably linear over the load ranges 20
applied (see below), which made the choice
10
of applied load and corresponding settlement
increment for the calculation of modulus 0
values straightforward. Table 2 summarises 0.01 0.1 1
various foundation properties, calculated Particle size (μm)
moduli, relative foundation stiffnesses
calculated from Equation 1 and the stiffness Figure 3 Particle size distribution of Cullinan sand used in centrifuge tests
classifications from Table 1.
Table 2 Model foundation properties
Back-
RESULTS Sand calculated Relative
Relative
Sand density relative secant Young’s foundation
Model footing stiffness
(kg/m3) density modulus for stiffness
Load-settlement response (%) sand
Ks
classification
of aluminium footings (MPa)
The footing settlements as measured at the
3 mm (aluminium) 1 578 67 18.4 0.003 Semi-flexible
column edge, normalised by the footing
width, are presented in Figure 4 for the vari- 6.7 mm (aluminium) 1 587 70 35.6 0.015 Semi-stiff
ous footing depths tested.
The settlement of all footings initially 10 mm (aluminium) 1 560 61 40.5 0.043 Semi-stiff
increased approximately linearly, with an
increase in the applied load over the load 16 mm (aluminium) 1 587 70 47.4 0.149 Stiff
range tested. However, in the case of the
3 mm deep footing, settlement measured 25 mm (aluminium) 1 587 70 47.4 0.569 Stiff
at the column edge began to accelerate and
became non-linear after approximately 8 kN 35 mm (aluminium) 1 577 67 47.4 1.562 Stiff

as the footing began to bend, suffering plastic

50 mm (aluminium) 1 587 71 47.4 4.555 Stiff
deformation. The load displacement results
show that, with increasing footing depth,
30 mm (concrete) 1 548 56 32.2 0.5181* Stiff*
the settlement under a given applied load
reduced, but only up to a point. The results * uncracked
show that, once the footing classified as stiff

22 Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
 6 (i.e. for footing depths of 16 mm and higher),
the load-settlement response could not be
improved by making the footing any stiffer,
5

le
Normalised settlements δ/B (%)

given the properties of the sand as tested.

x ib
Settlement rate

e
Careful examination of the 16 mm deep

-f l
decreases with

mi
4 iff
increasing stiffness i- s t footing showed, however, that it did not

Se
S em St i f f behave entirely rigidly. Figure 5 presents the
3 settlement measured across the width of
the footing at third points, showing slight
2 n variation and hence bending of the footing at
uc t i o f ne s
s
r a t e re d t i n g s t i f larger loads. This suggests that the classifica-
e n t f o o
em ng
s e t t l i nc r e a s i tion system by the Canadian Foundation
1 it o f
L i m v a ble by
ac h i
e Engineering Manual (2006) could perhaps be
0
refined somewhat at the boundary separating
0 5 10 15 20 25 30 35 semi-stiff and stiff footings.
Applied load (kN)
Model footing depth: 3 mm 10 mm 25 mm 50 mm Load-settlement response of the
6.7 mm 16 mm 35 mm reinforced concrete footing
In addition to the investigation of the influence
Figure 4 Normalised load-settlement response for aluminium footings of varying depth of footing rigidity using aluminium footings,
one test was carried out using a footing made
from scaled-down reinforced concrete. The
3.5 depth of the reinforced concrete footing was
chosen so that, for an uncracked section, the
3.0 foundation system classified as stiff (Ks value of
Normalised settlements δ/B (%)

0.52), as concrete foundations in reality would

2.5 be expected to be stiff. The uncracked bending
stiffness (EI) of the reinforced concrete footing
2.0 was similar to that of an aluminium footing
with a depth of 21.3 mm.
1.5 The normalised settlements measured at
third points along the width of the reinforced
1.0 concrete footing are presented in Figure 6.
The onset of cracking seems to occur at an
0.5 applied load of approximately 8 kN, i.e. a con-
tact stress of 356 kPa. This stress is well below
0 the allowable stress of approximately 600 kPa
0 5 10 15 20 25 30 35
for a dense sand calculated using the safety
Applied load (kN)
factor of 3 with the factors by Meyerhof (1963)
Column edge 25 mm from centre 50 mm from centre Footing edge
or Brinch Hansen (1968) and recommended
by BSI (2015). This indicates that cracking can
Figure 5 Normalised settlement of 16 mm thick aluminium footing
affect the performance of a foundation in its
working stress range.
4.0 A clear difference in the response of the
footing is evident prior to and after cracking.
3.5 The applied load was increased to 18.7 kN
(average contact stress of 832 kPa), beyond
Normalised settlements d/B (%)

3.0 Broken line shows settlement rate for which the footing was unable to accept
semi-flexible footing from Figure 4 additional load and just continued to deform.
2.5
Cracking of the footing resulted in the foot-
2.0 First onset of cracking ing effectively changing from stiff to much
more flexible. The settlement rate for the
1.5 3 mm deep footing, which classified as semi-
flexible, is presented against the settlement
1.0 record of the concrete footing measured
Load settlement response of 20 mm at the column edge for comparison. This
0.5
deep aluminium footing with similar illustrates that, after cracking, the load-
Ks value as concrete footing
0
settlement response of the concrete footing
0 5 10 15 20 effectively became semi-flexible.
Applied load (kN)
Column edge 25 mm from centre 50 mm from centre Footing edge Contact stress distribution
The contact stress distribution underneath
Figure 6 Normalised settlement of reinforced concrete footing each model footing was measured using a

Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017 23
 Front of footing
Active part of pressure sensor
Figure 7 Position of pressure sensors under footings
0 15
0
Qualitative contact stress
(increasing downward)
Footing depth:
Figure 8 Qualitative contact stress distribution under aluminium footings at 10 kN applied load
Tekscan™ tactile pressure sensor mat placed
underneath the footings. A top view of the
model footing superimposed over the pres-
sure mat is illustrated in Figure 7, together
with a 3D view of a measured contact stress
distribution. The active area of the pressure
sensor was smaller than the footing foot-
print, as shown in Figure 7.
Figure 8 illustrates qualitative measure-
ments of the pressure distribution across half
the width of the aluminium footings tested
at an applied load of 10 kN (contact stress
444 kPa). In general, despite some scatter,
three classes of behaviour can be identified:
Stiff footings showed a slightly concave pres-
sure distribution, increasing from the centre
towards the footing edge, but for practical
purposes this pressure distribution can be
taken as approximately constant. Semi-stiff
footings showed a slightly convex pressure
distribution, reducing slightly from the cen-
treline towards the edge. The semi-flexible
(3 mm deep) footing showed a pressure
distribution reducing approximately linearly
towards the edge. Due to excessive bending,
the edges actually lifted upwards, losing
24
Fooring
centreline
Edge of footing
Distance from centreline (mm)
30 45 60
3 mm 10 mm 25 mm 50 mm
6.7 mm 16 mm 35 mm
contact with the sand. The area of maximum
pressure near the footing centreline coin-
cided with the width of the column.
With increasing load, the shape of the
pressure distribution under the more flex-
ible footings evolved somewhat, as reported
by Arnold et al (2010). Upon very first load
application, the contact stress distribution
was approximately uniform. With further
load the stress distribution became approxi-
mately trapezoidal, reducing from the centre
towards the edges (see Figure 8). As load
increased further, the shape of the contact
stress distribution became approximately
triangular (see Figure 8, 3 mm thick foot-
ing), as bending of the footing occurred and
the sides of the footing lifted off the sand
surface. Contact stress distribution with
increasing load for the more flexible footings
is presented conceptually in Figure 9. The
reduction in contact stress under the footing
edges during loading became less as the foot-
ing depth increased, as discussed below.
The measured stress distributions
illustrate that the assumption of a uniform
pressure distribution is only approximately
Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
(a) Uniform distribution
(b) Trapezoidal distribution
75
(c) Triangular distribution
Figure 9 I dealised pressure distributions
under footings
valid for stiff foundation systems, and that
the pressure distributions vary somewhat
depending on the stiffness of the foundation
and the applied load.
DISCUSSION OF RESULTS
Settlement and contact stress
distribution as a function
of footing stiffness
The normalised settlements of the footings
measured at the edge of the column at vari-
ous loads are illustrated as functions of foot-
ing bending stiffness (EI) in Figure 10. The
bending stiffness values used in the graphs
are those of the model footing and not the
prototype footing (the scale factor from an
aluminium model to concrete prototype
stiffness is N4 E Al/Econ or approximately
290 000, where N is the model scale factor
of 30 and E Al and Econ the Young’s moduli of
respectively aluminium and concrete). The
stiffness values are presented on a logarith-
mic scale due to their large range. Data from
the 3 mm footing is not presented beyond an
applied load of 15 kN, as the footing could
not be loaded beyond this value due to exces-
sive bending. The density of the sand in the
 rapidly with increasing foundation stiffness
Footing depth
and then stabilises, becoming practically

6.7 mm

50 mm
25 mm

35 mm
10 mm

16 mm
independent of foundation stiffness once the

3 mm
foundation classifies as stiff. This implies
6 that, as soon as the stiffness of the footings
was sufficient for the foundation to classify
5 as stiff, the behaviour of the footings was
Normalised settlement δ/B (%)

similar and independent of actual relative

foundation stiffness, with settlement con-
4
trolled by the properties of the sand. The
model tests indicate that this occurred at
3 Ks values exceeding 0.3, as discussed below,
and this value represents the benchmark
2 that a designer would want the foundation
to exceed. In the case of semi-flexible and
1 semi-stiff foundations, the settlement was
dependent on both the deformation of the
0
footing and the sand.
10 100 1 000 10 000 100 000 1 000 000
Footing bending stiffness EI (N.m2) Cracking of concrete
Applied load: 5 kN 10 kN 15 kN 20 kN 25 kN 30 kN The response of the concrete footing dur-
ing loading is also illustrated in Figure 11.
Figure 10 S ettlement at the column edge as a function of footing bending stiffness The settlement of the uncracked reinforced
(aluminium footings) concrete footing, measured at the column,
was similar to that of the stiff aluminium
footings, but once the concrete had cracked,
Footing depth the column settlement increased to that
of a semi-flexible aluminium footing. This
6.7 mm

50 mm
25 mm

35 mm
10 mm

16 mm
3 mm

occurred well within the working contact

stress range as discussed previously.
0.35

Contact stress distribution

0.30 Settlement occurs due to
The effect of footing stiffness on the meas-
Normalised settlement rate (1/kN)

deformation of the soil Settlement occurs due to

and some deformation deformation of the soil. ured contact stress distributions underneath
0.25 of the footing.
the footing is illustrated in Figure 12. The
Ef f
ec t figure presents the contact stress measured
0.20 of
cr a at the edge of the footing as a percentage of
ck
i ng
of the contact stress measured under the centre
0.15 co
Settlement occurs nc of the footing plotted against the relative
re t
due to significant e
foundation stiffness. The 3 mm deep foot-
0.10 deformation of
both the footing ing was sufficiently flexible that stress was
and the soil. not measured under its edges during load
0.05
Semi-flexible Semi-stiff Stiff application. The 25 mm, 35 mm and 50 mm
0
deep footings experienced an approximately
0.001 0.01 0.1 1 10 uniform pressure distribution, with the
Relative foundation stiffness (Ks) pressure under the edge similar to the pres-
Aluminium Concrete sure under the centre of the footing. These
footings classified as stiff-based on their
Figure 11 Settlement rate as a function of the relative foundation stiffness Ks relative foundation stiffness. The 16 mm
deep footing also classified as stiff according
test on the 10 mm deep footing was less than Table 1). The implication for design is that a to Table 1 with a Ks value exceeding 0.1,
in the other tests, resulting in these results footing just needs to be made deep enough but experienced a reduced pressure below
falling somewhat outside the general trend to meet the relative stiffness classification of the edge, indicating that it did not behave
observable in Figure 10. “stiff” (Table 1). as entirely stiff. With further reduction in
Consider any applied load in Figure 10. Figure 11 illustrates the slope of the footing depth, the pressure registered under
Increasing the footing’s bending stiffness load-settlement curve for the various the footing edges reduced further. This illus-
resulted in a reduction in the settlement footings from Figure 4 plotted against the trates that the assumption of a uniform con-
measured at the column, but only up to a relative foundation stiffness Ks. (The slope tact stress distribution (in effect the Winkler
point. The results in Figure 10 suggest that of the load-settlement curve is represented model) is only valid for stiff footings which
there is no benefit in using a footing stiffer by the settlement measured at the column do not deform appreciably, which correlates
than the 16 mm footing. For this footing expressed as a percentage of the footing with the findings of Morfidis and Avramidis
the relative foundation stiffness classified width divided by load in kN). Starting with (2002). Fitting tentative bounds around the
as stiff (i.e. a Ks value exceeding 0.1, refer to a low Ks value, settlement with load reduces contact stresses recorded at the footing edges

Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017 25

3 mm
100
Edge contact stress / centre contact stress (%)
90
80
70
60
50
40
30
20
10
0
0.001
Applied load:
Figure 12 Effect of footing bending stiffness on stress distribution under footings
in Figure 12 and extrapolating suggest that,
for perfectly stiff conditions, the relative
stiffness Ks should exceed 0.3.
Predicting bending moment
in the footing
Stress measurement underneath the model
footings showed that, for stiff foundation
systems (Ks > 0.3), the contact stress under-
neath the footings can be taken as approxi-
mately uniform. As the relative stiffness
reduces into the semi-stiff range, the contact
stress underneath the footing edge as a
percentage of the maximum reduces. When
the stiffness reduces further to semi-flexible,
the contact stress at the edge approaches
zero and the edge may even lift up. It would
be beneficial to be able to predict the contact
stress distribution as a function of the rela-
tive foundation stiffness, as the contact stress
distribution affects the bending moment that
the footing has to be designed for.
Figure 12 shows that the contact stress
at the edge of the footing, as a percentage of
the maximum stress, does not change much
during loading. It also shows that, across a
range of relative stiffness values, the contact
stress at the footing edge as percentage of the
maximum varies approximately linearly with
the logarithm of the relative foundation stiff-
ness Ks. Let the ratio between the contact
stress under the edge of the footing versus
the maximum value occurring under the
centre be represented by c. It is proposed that
c varies linearly with the logarithm of the
relative foundation stiffness from zero at a
value of 0.003 to 1 at a value of 0.3 as shown
26
Footing depth
6.7 mm
50 mm
25 mm
35 mm
10 mm
16 mm
Ks =0.3
log Ks
c= – 1.26
2 RECOMMENDATIONS
0.01 0.1 1 10
Relative foundation stiffness Ks
5 kN 10 kN 15 kN 20 kN 25 kN
in Figure 12. This relationship is described
by Equation 3:
log Ks
c= – 1.26
2 behaviour of the footing, but it affects the
Let a represent the maximum contact stress
under the footing centreline and b the foot-
ing width. The maximum bending moment
per unit length occurring under the centre-
line of the footing, assuming a trapezoidal
distribution (see Figure 9b), is then:
ab2
M= (1 + 2c)(4)
24
For a stiff footing the pressure distribution
is uniform and c = 1 so that the bending
moment M = ab2/8. With a reduction in
relative stiffness to semi-flexible, c reduces
to zero, but under the same vertical load a
increases by a factor of 2 so that the bend-
ing moment becomes M = ab2/12. The
bending moment in a footing forming part
of a semi-flexible foundation can therefore
be as little as two-thirds of the moment in
footing which is part of a stiff foundation. In
addition, using the correct stress distribu-
tion, the deflected shape of the footing can
be calculated more accurately than a result
simply based on the assumption of a uniform
stress distribution.
Relative foundation stiffness
“stiff” versus “semi-stiff”
Although the boundary between stiff foot-
ings and semi-stiff footings is given by the
Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
Canadian Foundation Engineering Manual
(2006) as a Ks value of 0.1, the 16 mm footing
(Ks value of 0.15) did not exhibit the same
behaviour as the stiff footings. Therefore, it
is suggested that the proposed lower bound-
ary to which a uniform pressure distribution
is valid is adjusted upward from a Ks of 0.1
to a value of 0.3. Even at conventional loads
(equivalent to 5 kN (222 kPa) or 10 kN
(444 kPa)) semi-stiff footings still showed
reduced stress under the edges, not behaving
as stiff as shown in Figure 12.
CONCLUSIONS AND
The following conclusions from the model
study are presented:
■■ The relative stiffness of a foundation sys-
tem affects the deflection of and contact
stress distribution beneath the footing.
The contact stress distribution beneath
30 kN footings in a stiff foundation on dense
sand is approximately uniform. A reduc-
tion in stiffness results in the footing
bending during loading, which reduces
the contact stress at the footing edges.
■■ The contact stress distribution beneath
a footing plays an important role in
not only the settlement and deflection
(3)
bending moments in the footing that the
structural engineer needs to design for.
■■ The results presented suggest that the
optimum depth of a footing is that depth
resulting in the foundation just classify-
ing as stiff, i.e. corresponding to a rela-
tive foundation stiffness Ks of 0.3. For
this, and higher stiffnesses, the contact
stress distribution under a strip footing
can be assumed to be approximately
uniform.
■■ The behaviour of an initially stiff con-
crete footing will become more flexible
after the onset of cracking. This can
occur well below the allowable bearing
capacity of the foundation. The settle-
ment undergone by a column bearing
on such a footing will then depend on
both the deformation of the underlying
soil and the deformation of the footing.
The contact stress at the footing edges
will reduce so that the maximum stress
under the foundation will increase if
the footing still has to support the same
load. This situation appears less optimal
than the behaviour of a stiff foundation.
The recommendation is therefore that
foundations be designed to have a relative
foundation stiffness classifying as “stiff”,
i.e. a Ks value of 0.3. It does not appear
that increasing the stiffness beyond this
value is beneficial.
REFERENCES
Aiban, SA & Znidarčic, D 1995. Centrifugal modeling
of bearing capacity of shallow foundations on sands.
ASCE Journal of Geotechnical Engineering, 121(10):
704–712.
Algin, H M 2007. Practical formula for dimensioning
a rectangular footing. Engineering Structures, 29(1):
1128–1134.
Archer, A 2014. Using small-strain stiffness to predict
the settlement of shallow foundation on sand. MEng
dissertation, University of Pretoria.
Arnold, A, Laue, J, Espinosa, T & Springman, S M 2010.
Centrifuge modelling of the behaviour of flexible raft
foundations on clay and sand. In: Springman, S M,
Laue, J & Seward, L (Eds.), Physical Modelling in
Geotechnics. Proceedings of the 7th International
Conference on Physical Modelling in Geotechnics
(ICPMG 2010) held in Zurich. London: Taylor and
Francis, 679–684.
ASTM (American Society for Testing and Materials)
1996a. ASTM D4253 –93: Test Methods for
Maximum Index Density and Unit Weight of Soils
Using a Vibratory Table. West Conshohocken, PA:
ASTM International.
ASTM (American Society for Testing and Materials)
1996b. ASTM D4254 –91: Standard Test Method
for Minimum Index Density and Unit Weight of
Soils and Calculation of Relative Density. West
Conshohocken, PA: ASTM International.
Baumann, R A & Weisgerber, F E 1983. Yield-line
analysis of slabs-on-grade. Structural Engineering
Journal, 109(11): 1553–1567.
Brinch Hansen, J 1968. A revised extended formula
for bearing capacity. Danish Geotechnical Institute
Bulletin, No. 28. Brabrand, Denmark.
Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
Brown, R, Valsangkar, A J & Schriver, A B 2004.
Centrifuge modeling of surface footings on a sand
layer underlain by a rigid base. Geotechnical and
Geological Engineering 22: 187–198.
BSI (British Standard) 2015. BS 8004:2015: Code of
Practice for Foundations. London: British Standards
Institution.
Canadian Foundation Engineering Manual, 4th
ed. 2006. Richmond, BC, Canada: Canadian
Geotechnical Society.
Conniff, D E & Kiousis, P D 2007. Elastoplastic medium
for foundation settlements and monotonic soil–
structure interaction under combined loadings.
International Journal for Numerical and Analytical
Methods in Geomechanics, 31(1): 789–807.
Haigh, S K, Houghton, N E, Lam, S Y, Li, Z &
Wallbridge, P J 2010. Development of a 2D servo-
actuator for novel centrifuge modelling. In:
Springman, S M, Laue, J & Seward, L (Eds.), Physical
Modelling in Geotechnics. Proceedings of the 7th
International Conference on Physical Modelling in
Geotechnics (ICPMG 2010) held in Zurich. London:
Taylor and Francis, 239–244.
Hemsley, J A 2000. Design Application of Raft
Foundations. London: Thomas Telford.
Knappett, J A, O’Reilly, K, Gilhooley, P, Reid, C &
Skeffington, K 2010. In: Springman, S M, Laue, J &
Seward, L (Eds.), Physical Modelling in Geotechnics.
Proceedings of the 7th International Conference
on Physical Modelling in Geotechnics (ICPMG
2010) held in Zurich. London: Taylor and Francis,
141–146.
Kusakabe, O 1995. Foundations. In: Taylor, R N (Ed.),
Geotechnical Centrifuge Technology, London: Blackie
Academic and Professional, 118–165.
Meyerhof, G G 1963. Some recent research on
the bearing capacity of foundations. Canadian
Geotechnical Journal, 1(1): 16–26.
Milovic, D M, Touzot, G & Tournier, J P 1970.
Stresses and displacements in an elastic layer due
to an inclined eccentric load over a rigid strip.
Geotechnique, 20(3): 231–252.
Morfidis, K & Avramidis, I E 2002. Formulation of
a generalised beam element on a two-parameter
elastic foundation with semi-rigid connections
and rigid offsets. Computers and Structures, 80(1):
1919–1934.
Pisanty, A & Gellert, M 1972. Automatic design of
sloped spread footings. Building Science, 7(1): 53–59.
Schofield, A N 1980. Cambridge geotechnical
centrifuge operations. Geotechnique, 30(3): 227–268.
Skorpen, S & Dekker, N 2014. The application and
interpretation of linear finite element analysis results
in the design and detailing of hogging moment
regions in reinforced concrete flat plates. Journal of
the South African Institution of Civil Engineering,
56(1): 77–92.
TekscanTM 2014. Sensor Model 5105. Available at:
http://www.tekscan.com/5101-pressure-sensor.
(Accessed on 29 October 2014).
Westergaard, H M 1925. Stresses in concrete pavements
computed by theoretical analysis. Public Roads, 7: 25.
Westergaard, H M 1937. What is known of stresses.
Engineering News Record, 118(1): 26–29.
White, D J, Take, W A & Bolton, D 2003. Soil
deformation measurement using particle
image velocimetry (PIV) and photogrammetry.
Geotechnique, 53(7): 619–631.
27
TECHNICAL PAPER Revised regional sediment
Journal of the South African
Institution of Civil Engineering
yield prediction methodology
ISSN 1021-2019
Vol 59 No 2, June 2017, Pages 28–36, Paper 1326 for ungauged catchments
DR VINCENT MSADALA holds PhD and
in South Africa
MSc degrees in Hydraulic Engineering from
Stellenbosch University. He has worked as a
lecturer at the University of Malawi in the V C Msadala, G R Basson
Department of Civil Engineering. His
research interests are in hydraulics, hydrology
and water resources development. Vincent This paper presents the research on the revision of the regional probabilistic method
is a member of the South African Institution
(Rooseboom et al 1992) for the prediction of catchment sediment yields in South Africa. The
of Civil Engineering (SAICE).
determination of sediment yields using probabilistic or empirical methods is suitable for
Contact details: ungauged catchments in the absence of observed data. The prediction of sediment loads is a
Department of Civil Engineering
key component in the quest to deal with reservoir and river sedimentation which is a potential
Stellenbosch University
Private Bag X1 threat to the sustainability of water resources in southern Africa.
Matieland The revision was necessitated by increased sediment data availability and improved
7602 data analysis tools. Ten new sediment yield regions were demarcated in South Africa and
South Africa Lesotho. Two analytical approaches were developed, namely probabilistic and empirical.
T: +265 888 817 470 / +265 882 750 550
The probabilistic approach is applicable in sediment yield Regions 3, 6 and 9. The empirical
E: vmsadala@gmail.com
approach is applicable in sediment yield Regions 1, 2, 4, 5, 7 and 8. The estimation of
sediment yields in Region 10 (Lesotho Highlands) needs to be based on direct measurements
PROF GERRIT BASSON is Head of the Water
Division, Department of Civil Engineering, and locally observed data since no adequate analysis of sediment loads was possible due to
Stellenbosch University. He specialised in river limited data. GIS and electronic portable document file (pdf) copies of maps were produced
hydraulics, sedimentation and design of for the retrieval of catchment data.
hydraulic structures. He worked for ten years at
consulting engineers in South Africa, then
joined the University of Pretoria in 1997,
and since 2000 he is based at Stellenbosch INTRODUCTION OBJECTIVES OF THIS PAPER
University. He has worked on projects in 21 countries and has 30 years’ Sediment affects the water quantity and This paper discusses the work carried out
experience as hydraulic engineer. He recently served the dam engineering quality in rivers, lakes and reservoirs. The recently to revise the sediment yield predic-
industry as Vice-President of the International Commission on Large Dams analysis of reservoir sedimentation rates in tion method by Rooseboom et al (1992).
(ICOLD), from 2012 to 2015. Prof Basson is also a Member of the South
South Africa indicated an estimated mean The detailed information on the research
African Institution of Civil Engineering (SAICE).
annual loss in original storage capacity of project is contained in a report to the Water
Contact details:
0.4% per year. Although the estimated global Research Commission of South Africa
Department of Civil Engineering
Water Division
reservoir sedimentation rate is 0.8% per year (Msadala et al 2012). The objectives of the
Stellenbosch University (ICOLD 2009), the sedimentation rate in research project on the revision of the 1992
Private Bag X1 South Africa is quite significant considering methodology were to use the most recent
Matieland the extent of increased water demand across reservoir sedimentation and river sediment
7602
southern Africa. Almost 25% of the total transport data, and to evaluate a regional
South Africa
T: +27 21 808 4355
number of reservoirs that were analysed in empirical stream power approach in addition
E: grbasson@sun.ac.za South Africa had lost between 10% and 30% to the original probabilistic sediment yield
of their original storage capacity. This calls prediction methodology, in order to improve
for increased attention towards reservoir the prediction accuracy.
sedimentation problems.
There have been extensive studies with
regard to sediment transport in rivers and SEDIMENT YIELD PREDICTION
reservoirs in South Africa dating back some APPROACHES
50 years. Some of the studies resulted in In practice, the sediment yield at any refer-
the development of the sediment yield map ence point is found by applying the following
of southern Africa and a treatise on sedi- four major approaches:
ment transport in rivers and reservoirs by ■■ Direct measurements from reservoir sedi-
Rooseboom et al (1992). The latter resulted ment deposit surveys
in what was considered as a basic handbook ■■ River suspended sediment sampling
for sediment yield prediction. With time it ■■ Catchment sediment yield mathematical
became apparent that the methods contained modelling
in the book needed to be updated to take ■■ Analytical methods such as sediment
into account the additional observed sedi- yield maps.
Keywords: sediment yield, analytical, probabilistic method, empirical method, ment data and technologically advanced data Two analytical approaches were evaluated,
ungauged catchments analysis tools. namely the probabilistic method (Rooseboom

28
Msadala VC, Basson GR. Revised regional sediment yield prediction methodology for ungauged catchments in South Africa.
J. S. Afr. Inst. Civ. Eng. 2017:59(2), Art. #1326, 9 pages. http://dx.doi.org/10.17159/2309-8775/2017/v59n2a4
et al 1992) and the empirical method. The 78
probabilistic method was based on statisti-
cal analysis of observed data, particularly 77
newly calculated sediment yield data, and the
revised erosion hazard classes. The empirical 76
method was based on the total input stream

Storage capacity (10 6 m3)

power concept in terms of the direct relation- 75
ship between the observed sediment yields
or loads and the selected predictor variables 74
within a region. South Africa, including
Lesotho, was demarcated into ten relatively 73
homogeneous sediment yield regions. The
new empirical method was observed to 72
have better predictive accuracy levels (when
compared to the probabilistic method) in 71
sediment yield regions with relatively large
sample sizes. Based on the guidelines by 70
Knofczynski and Mundfrom (2008) on the 1950 1960 1970 1980 1990 2000 2010
minimum sample size and the squared Period (years)
population multiple correlation coefficients
needed for better predictions, it was observed Figure 1 S urveyed reservoir storage capacity change over time at Rustfontein Dam
that the empirical method could only be
considered more accurate in sediment yield
7
Regions 1, 2, 4, 5, 7 and 8. Since the predic-
tion levels of the multiple regression models
6
for the empirical method in Regions 3, 6
Sediment volume
and 9 were not good, a revised probabilistic after 50 years
Sediment volume (10 6 m3)

5
method (Rooseboom et al 1992) was imple-
mented. As such, the probabilistic method is
4
applicable to sediment yield Regions 3, 6 and
9, while the empirical method is applicable
3
to sediment yield Regions 1, 2, 4, 5, 7 and
8. The estimation of sediment yields in
2
Region 10 (Lesotho Highlands) should be
based on direct measurements and locally
1 y = 3.4678ln(x) – 7.0333
observed data since no adequate probabilistic R2 = 0.9807
or deterministic analysis of sediment loads 0
and predictor variables was possible due to 1 10 At 50 years 100
limited data. Period (year)

Figure 2 T emporal changes in reservoir sediment deposit volume at Rustfontein Dam

DATA COMPILATION AND
ANALYSIS PROCEDURES
The observed sediment yields were obtained
from the latest reservoir survey data and
river suspended sediment sampling data.
The Department of Water and Sanitation
(DWS 2015) dam list of 2015 provided data
on reservoir sediment deposition. Where
reliable information was available, the sedi-
ment yields were obtained from technical
or research reports of individual dams. The
criteria for the validation of sediment deposit
data for quality control purposes included
evaluating the following aspects: records
or period between surveys had to be longer
than ten years, a reliably high reservoir sedi-
ment trap efficiency (at least 97%) and the
effects of raising or lowering of a dam (if any
was done).
The mean annual runoff (MAR) was
used to compute the trap efficiency indicator
based on the storage capacity of a particular
dam. The trap efficiency is indicated by the
ratio of the storage volume (V W) against the
MAR of the reservoir. The trap efficiency
indicator was then used to quantify the
estimated amount of sediment that was
trapped out of the total sediment inflow. In
semi-arid regions, the storage capacity of a
reservoir for irrigation or potable water use
is usually in the order of the MAR of the
catchment, and the reservoirs therefore trap
approximately 97% of the sediment yield,
with only colloidal sediment transported
through the reservoir (Basson & Di Silvio
2008). The MAR data was obtained from the
Surface Water Resources of South Africa 1990
(WR90) (Midgley et al 1994), which gives
the MAR for each quaternary catchment in
South Africa.
The effective catchment area was applied
in the calculation of the sediment yield
where reliably high reservoir sediment trap
efficiency was computed for an upstream
dam (if any). Furthermore, the trap efficiency
was taken into consideration only if there
was a common period of sampling or survey-
ing between cascaded dams.
The difference between the initial storage
capacity and the subsequent storage capaci-
ties, as illustrated in Figure 1, is related to
the surveyed sediment that is deposited in
the reservoir. A graph was used to predict
the volume of sediment in the reservoir
after or at 50 years, as shown in Figure 2.
According to Rooseboom et al (1992), the
sediment volume after 50 years is considered
to be a true reflection of accumulated sedi-
ment in the reservoir, because by that time
sufficient consolidation is expected to have
taken place. After 50 years of deposition,
the average sediment density can be taken
as 1.35 t/m3. This average density of sedi-
ment is considered in this case, due to the

Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017 29
 20°E 25°E 30°E 35°E
potential consolidation characteristics of WATER RESEARCH COMMISSION (2010): REVISED SEDIMENT YIELD MAPS
SOUTH AFRICA AND LESOTHO
fine sediment (clay and silt). A logarithmic SEDIMENT YIELD BY RIVER GAUGE AND DAM CATCHMENT
trend line was fitted through the data points
to determine the 50-year sediment volume
from the equation of the trend line. This was
based on the assumption that a logarithmic
relationship exists between the sediment

25°S

25°S
Legend
deposited in a reservoir and the time period
Rivers
(Rooseboom et al 1992). The 50-year sedi-
Regions
ment volume and the effective catchment Sediment yield (t/km2.a)
area were used to compute the sediment 1–50
50.000001–100
yield or load at the dam.
100.000001–150
A map of observed sediment yields is 150.000001–200

30°S
given in Figure 3. It should be noted that it 200.000001–250

30°S
250.000001–300
was not an objective of this study (Msadala 300.000001–350
et al 2012) to develop a sediment yield map 350.000001–400

by extrapolating the map in Figure 3 to N

400.000001–450
450.000001–500
ungauged catchments, but to develop a cal- W E 500.000001–600
culation methodology for the prediction of S
600.000001–700
700.000001–800
sediment yields or loads in ungauged catch- Map information:
800.000001–1 000
35°S

Datum: WGS 84
ments in South Africa.

35°S
Projection: Albers Equal Area 1 000.000001–2 000
0 100 200 300 400 500 600 700 800 900 1 000
Central Meridian: 24°E Kilometres
The revised sediment yield values were Standard Parallels: 18°S and 32°S 2 000.000001–2 500

sometimes higher for similar dams when 20°E 25°E 30°E 35°E

compared to those of the 1992 sediment Figure 3 O

yield map of Rooseboom et al (1992) and
vice versa. Nevertheless, a statistical analysis
of the general trends in the sediment yields
showed that the newly calculated sediment
yields were comparatively higher. Based on
 bserved sediment yield map of South Africa and Lesotho (Msadala et al 2012) for
gauged catchments
REVISED SEDIMENT YIELD MAP OF SOUTHERN AFRICA (1992)
Sediment Yield Regions

this finding, the sediment yields in South

Prepared for Sigma Beta by the Institute for Cartographic Analysis
Africa are generally increasing and this could University of Stellenbosch
Erodibility Index 0 100 200 300 400 500
be attributed to land use change. 1–8 High km

9–15
16–20 Low
Yield region
THE PROBABILISTIC APPROACH Catchment
boundaries

Background
The probabilistic approach was an updated
methodology that was based on the previous
report’s (Rooseboom et al 1992) fundamental
assumptions. The main underlying assump-
tion in the probabilistic approach is that the
fine cohesive sediment transport is depend-
ent on sediment availability, and in turn
sediment availability is influenced by the
soil erosion hazard. This was a rather crude
assumption, considering the dominant role
of other significant sediment-controlling fac-
tors, such as river network density and floods
that could also affect the sediment yield Based on classification by Prof E Verster, UNISA

of a catchment. The probabilistic analysis

approach relates the erosion hazard classes
within known catchment areas to the cor-
responding observed sediment yields within
a homogeneous sediment yield region.
Demarcation of the new
sediment yield regions
The identification and demarcation of
regional boundaries were based on the lat-
est calculated sediment yields, hydrological
parameters such as watershed quaternary
boundaries, flood regions and soil erosion
Figure 4 P
 revious sediment yield regions by Rooseboom et al (1992) showing erodibility indices
and sediment yield regions
hazard classes. The previous sediment yield Soil erosion hazard classes
regions of Rooseboom et al (1992) are shown The erosion hazard classes were generated
in Figure 4, and the revised sediment yield after a re-assessment of the soil erosion risk
regions (Msadala et al 2012) are shown in in South Africa and Lesotho. Lesotho was
Figure 5. Figures 4 and 5 give erodibility included due to its importance in the Senqu-
indices and erosion hazard classes respec- Orange River catchment. The soil erosion
tively, which are basically different quantita- hazard classes were based on the original
tive classification criteria of erosion risk at erosion map by Morgenthal et al (2006) and
national scale. the water erosion prediction map by Le Roux

30 Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
 20°E 25°E 30°E
Erosion Index
area that was covered by each individual
1 – Very low erosion hazard class. Each observed sedi-
2
3
ment yield value was associated with its
4 – Moderate corresponding single dominant erosion
5 1
6
hazard class (Erosion Index – EI) within the
2
25°S

catchment area. The single dominant erosion

25°S
7 – High
8
9
hazard class per catchment was based on the
10 – Extremely high weighted average erosion hazard class that
Regions was converted to the nearest integer in the
4
5 range of 1 to 10. The sediment yield values
for dam and river catchments with similar
10 dominant erosion hazard classes were
3
30°S

30°S
6
grouped together. The median values of the
sediment yield values with similar dominant
erosion hazard class were computed.
7
Probabilistic analysis was done to obtain
9 sediment yields for all the required prob-
8
abilities of exceedance within a region. The
0 200 400 600 800 Kilometres
plotting of sediment yields data on a graph
35°S

15°E
Figure 5 N
 ew sediment yield regions showing erosion hazard classes (Msadala et al 2012)
et al (2008), whereby newly improved cover
and topography factor maps were applied.
Ten erosion hazard classes were identified
using Geographical Information Systems
(GIS) spatial data analysis tools. According
to Figure 5, very low erosion hazard was
taken as Class 1, and Class 10 represented
extremely high erosion hazard. The deci-
sion to use ten erosion hazard classes was
necessitated by the high variability in erosion
hazard potential across southern Africa. The
use of three groups of erodibility indices
35°S
20°E 25°E 30°E
in the previous methodology (Rooseboom
et al 1992) was found to be not representa-
tive enough with respect to the observed
erosion potential in the characteristically
diverse homogeneous regions. Hence, the
current approach proved to be a significant
improvement in the prediction accuracy of
sediment yields.
The probabilistic methodology
The probabilistic analysis applied the pro-
portion of area out of the total catchment
at the required exceedance probability was
done manually using Microsoft Excel. Three
distribution para­meters were employed to
statistically analyse the data, namely the
Log Normal (LN), the Log Pearson III (LP3)
and General Extreme Value using Mean
Moments (GEVmm). The LN distribution was
found to be best suited to data for Regions 3
and 6, while data for Region 9 was based on
the average between the LN and Log GEVmm
distributions’ values, which gave a better fit.
The main assumption in the probabilistic
analysis was that the standardised average
yield (SYsty) for each region, could be taken
as the sediment yield at 50% exceedance

10 000
Region 3 plot
Region 6 plot
Region 9 plot
Region 3 data points
Region 6 data points
1 000 Region 9 data points
Sediment yield (t/km 2 .a)

100

10

1
0.998 0.995 0.99 0.98 0.95 0.9 0.8 0.5 0.2 0.1 0.05 0.02 0.01 0.001 0.0001
Probability of exceedance

Figure 6 S tatistical analysis graphs for standardised average yield determination

Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017 31
probability from the statistical analysis plots
in Figure 6.
The probabilistic approach was designed
to use sediment potential factors to relate the
standardised average yield (SYsty) per region
to the estimated sediment yield SYest for an
ungauged catchment within a homogeneous
sediment yield region. The sediment potential
factor was taken as the ratio of the median of
each group of observed sediment yield with
similar dominant erosion hazard class and the
standardised average sediment yield.
For any ungauged catchment within these
three regions, Equation 1 is used to predict
or estimate the sediment yield SYest.

A1 A2 A3
SYest = SYsty F1 + F2 + F3
AT AT AT
A4 A5 A6 A7 Figure 7 I llustration of data query per quaternary catchment (Msadala et al 2012)
+ F4 + F5 + F6 + F7
AT AT AT AT
Acrobat Reader. As shown in Figure 7, the factor of 1 is applied at 50% probability of
A8 A9 A10 user is provided with sediment-related qua- exceedance, such as when Equation 1 is used
+ F8 + F9 + F10 (1)
AT AT AT ternary data (as shown on the left pane) upon without considering confidence limits. The
clicking within any quaternary. multiplication factors for selected confidence
Where: Equation 1 only calculates the estimated limits for Regions 3, 6 and 9 can be obtained
SYest = estimated median sediment yield median sediment yield value (SYest). In order from the appendices in the Water Research
value (t/km2.a) to get the predicted sediment yield (termed the Commission Report (Msadala et al 2012).
SYsty = standardised average sediment yield ‘factored estimated sediment yield’ (SYfest)),
value (t/km2.a) for a specific region. the estimated median sediment yield (SYest) Verification of results
value should be multiplied by a special mul- The predictive accuracy of the probabilistic
The standardised average yield for each sedi- tiplication factor. The multiplication factor, method was evaluated by using the discrep-
ment yield region is shown in Table 1. F1 to which is dependent on the preferred confi- ancy ratio test at 50% exceedance probability.
F10 are sediment potential factors, of which dence limit, catchment area size and sediment The technique compares all the predicted
the values are also given in Table 1. A1 to yield region is obtained from sediment yield sediment yields against all the observed sedi-
A10 in Equation 1 are catchment areas that confidence limits graphs, which were specifi- ment yields in terms of the discrepancy ratio
are covered by erosion hazard classes 1 to 10 cally prepared for each of the three sediment xi whereby each predicted value is divided by
respectively within the total catchment AT. yield regions. The relationship is shown below: the corresponding actual observed value. In
The determination of the area covered mathematical terms the discrepancy ratio is
by each of the erosion hazard classes within SYfest = Multiplication Factor × SYest(2) given by the following relationship:
the ungauged catchment to be applied in
Equation 1 is illustrated in Figure 7, which Where: SYsim
= xi(3)
shows a sample of an electronic portable SYfest = Factored estimated median sediment SYobs
document file (pdf) copy for sediment yield yield value (t/km2.a).
Region 1. The quaternary catchments are Where:
labelled based on the WR90 nomenclature, Multiplication factors and SYsim = predicted sediment yield
and in this particular case it relates to A24H. confidence limits SYobs = observed sediment yield.
The area covered by each of the erosion The confidence limits are required to act as
hazard classes within a quaternary catch- envelope values on the regional standardised The predicted sediment yields were calcu-
ment can be displayed upon querying each average yield upon application of Equation 2. lated using Equation 2 at 50% exceedance
of the individual quaternaries within the A 50% exceedance probability within a given probability. The results were as follows:
catchment boundary. The querying is done region indicates that 50% of the predicted val-
using the ‘Object Data Tool’ functionality ues could be lower, or 50% of the predicted val- 0.33 ≤ xi ≤ 3 ( 81% of the data was in this
in the ‘Model Tree’ viewing mode in Adobe ues could be higher. By default a multiplication range)

Table 1 Sediment potential factors and regional standardised average yield

Standardised Sediment potential factors
Region average yield
(t/km 2 .a) F1 F2 F3 F4 F5 F6 F7 F8 F9 F10

3 40 0.027 0.417 1.202 1.110 0.575 0.000 0.000 0.000 0.000 0.000

6 622 0.000 0.000 0.000 0.000 0.000 1.428 1.042 1.002 0.998 0.000

9 50 0.196 1.824 1.871 0.000 0.000 0.000 0.000 0.000 0.000 0.000

32 Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
0.5 ≤ xi ≤ 2 ( 68% of the data was in this
range)
0.67 ≤ xi ≤ 1.5 (43% of the data was in this
range)
The predictive accuracy was also compared
with the previous probabilistic approach
(Rooseboom et al 1992) results, which were
as follows:
0.33 ≤ xi ≤ 3 ( 70% of the data was in this
range)
0.5 ≤ xi ≤ 2 (47% of the data was in this
range)
0.67 ≤ xi ≤ 1.5 (32% of the data was in this
range)
The results for each of the three regions
are shown in Table 2 for the probabilistic
approach.
These ranges were within the limits of
acceptable predictive accuracy, considering
the complex nature and spatial variability
in sediment yields. The small number
of observations in Table 2 had sufficient
statistical significance within the objectives
of the study, since the combined effect of
the catchment areas represented a large
significant proportionate area with respect
to the total area in each of the homoge-
neous sediment yield regions that were
under consideration.
Special considerations in using
the probabilistic approach
The estimation of the sediment yield was
developed from the average of the observed
data series which was taken as the 50th
percentile. For higher or lower percentiles,
multiplication factors were proposed. These
factors are supposed to be applied with cau-
tion in order to avoid over-prediction. The
probabilistic approach predicts a sediment
yield value that statistically masks all possi-
ble values with respect to the selected prob-
ability of exceedance for a specific region.
Specifically the results showed that for some
low observed sediment yields, particularly
less than 100 t/km2.a, the method gives rela-
tively higher predicted sediment yields for
ungauged catchments, and vice versa.
Two cautionary measures are recom-
mended. Firstly, it is necessary to check the
predicted sediment value at 50% confidence
limit against the nearest observed yield
values within the region. Secondly, the
graphs for the statistical distribution shown
in Figure 6 (probability of exceedance) of
the observed sediment yields for the each
of the regions can be used to compare the
predicted value against the expected sedi-
ment yield value on the graph at any specific
probability of exceedance. These graphs
Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
Table 2 Discrepancy ratio results for the probabilistic method (Msadala et al 2012)
Region Obs n
0.67 ≤ xi ≤ 1.5
3 7
6 10
9 9
give an estimate of the general variation of
the expected sediment yields within a given
homogeneous region.
THE EMPIRICAL APPROACH
Theoretical basis – the concept
of total input stream power
The empirical approach was aimed at
establishing the relationships between
sediment yield and any data that pertains to
significant variables which are involved in
sediment transport and yield processes. The
total input stream power was the basis on
which the empirical equations were derived
through regression analysis. The regression
analysis of the variables was done per sedi-
ment yield region.
The rate of energy dissipation that would
be required to transport material is related
to the rate of material to be transported
according to the general concept of physics.
According to Yang (1996), the sediment
transport rate is directly related to the unit
stream power. Therefore, sediment transport
can be described by the following total input
stream power relationship:
ρgQS
Qs ∝
w (4)
Where:
Qs = sediment load (t/a)
Q = discharge (m3/s)
s = energy slope (m/m)
w = settling velocity of sediment (m/s)
ρgQS = total input stream power (ρg is
assumed constant).
Equation 4 assumes that there is sediment
transport capacity based on the local hydrau-
lic conditions and sediment characteristics.
This is generally true for coarse sediment
(sand and gravel), but in the South African
conditions where more than 75% of the sedi-
ment transported consists of clay and silt
fractions, the sediment transport capacity
is high, but the sediment availability from
the catchment could be limited. Therefore,
additional variables had to be considered
to account for the sediment availability. In
Percentage of data in this range
0.5 ≤ xi ≤ 2 0.33 ≤ xi ≤ 3
71 71 86
56 67 89
44 78 89
other words, it was assumed that there is a
combined effect of both sediment production
and transport capacity controlling factors
related to hydraulic conditions and sedi-
ment characteristics. The settling velocity
variable was not included in the regression
model, because of the difficulty with which
it could be universally determined at a
catchment scale.
A weighted erosion hazard class (EIw) was
added to the regression model to account
for sediment production. River network
density (Rnd) is a parameter that affects
sediment transport through the catchment.
The catchment area (A) was also added to
describe the sediment source spatial extent
and characteristics. The discharge (Q) was
based on a recurrence interval flood pro-
posed to be the 10-year annual recurrence
interval (ARI) flood. This was established
through regression analysis accuracy checks
of all available recurrence interval floods to
determine the ARI that best fitted the data.
The energy slope (S) for a catchment was
taken as the average river slope of the longest
water course.
Derivation of the empirical
equations
According to Rooseboom et al (1992), sedi-
ment transport is a hydrological process,
and therefore is a function of the same para­
meters that influence all hydrological pro-
cesses. Hydrological data is usually strongly
skewed, but the logarithms of the data have
a near symmetrical distribution (Hazen
1914). The dependent variable of sediment
load and the independent variables were
logarithmically transformed and applied
in multiple regression analysis in order to
derive a better regression model. A column
of the sediment load as a dependent variable,
and five columns comprising the recurrence
interval flood, average river slope, river net-
work density, catchment area and weighted
erosion hazard class (index) were prepared.
The regression analysis was performed in
Microsoft Excel 2007.
The proposed equations showing the
results of the derived coefficients after
regression analysis and correlation against
observed data are shown in Table 3.
33
Table 3 E mpirical equations based on GIS maps using a similar procedure that was Significance of the variables in
regression analysis explained under Figure 7. The electronic the empirical equations
version of these maps were prepared as part
Region Proposed equation
of the WRC project (Msadala et al 2012) and Weighted erosion hazard class
1 0.98 S –0.38 R 0.13 A 0.10 EI 1.22
Qs = 15Q10 0 nd e w were included as part of the report appen- The weighted erosion hazard class basi-
2 0.32 S 0.19 R 0.63 A 0.95 EI –1.44
Qs = 6.4Q10 dices of maps. The weighting of the erosion cally provides a quantitative measure of
0 nd e w
hazard classes is with respect to the catch- the following parameters: climate, soil
4 0.53 S 0.43 R 1.08 A 0.88 EI –1.04
Qs = 0.2Q10 0 nd e w ment area covered by each erosion hazard profile, relief, vegetation, land use and land
5 1.18 S 0.58 R –1.27 A 0.54 EI –0.30
Qs = 1 208Q10 0 nd e w
class. The 10-year ARI flood and effective management practices based on the Revised
catchment area can be calculated using Universal Soil Loss Equation (RUSLE) model
7 0.43 S 0.6 R 0.17 A 0.5 EI 0.76
Qs = 40Q10 0 nd e w standard hydrological methods. Similarly, of Renard et al (1991).
8 –0.24 S 1.27 R 1.52 A 1.26 EI –0.61
Qs = 0.004Q10 0 nd e w the river network density can be obtained
from the electronic pdf maps in km/km2 per River network density
Where:
quaternary based on the data, such as that Drainage density is a measure of the length
Qs = sediment load (t/a)
Q10 = a flood of a recurrence interval of 10 years
given in Figure 7. It should be noted that of stream channel per unit area of basin.
(m3/s) electronic pdf maps give river network densi- River network drainage density can be clas-
Rnd = river network density (m/km 2) ties in km/km2 units, and these values need sified as one of the factors that determine
Ae = effective catchment area (km 2) to be converted to m/km2 prior to applying the catchment’s sediment yield according
EIW = weighted erosion hazard class according to in any of the equations in Table 3. Using the to Strand and Pemberton (1982). It can be
sub-catchment areas
area for each quaternary, the river length per assumed that, with all factors equal for the
So = average river slope (%)
quaternary can be computed by multiplying same catchment area, longer lengths of
the total area and the corresponding river river channel per unit area must be able to
The equations in Table 3 can be used to network density, and then summing up all transport more sediment than shorter river
predict the sediment yield or load in an river lengths for all quaternaries to give the channels within the catchment.
ungauged catchment as long as accurate total river length in the whole catchment.
data with respect to the specified variables is Finally, the river network density for the Recurrence interval flood
applied. The weighted erosion hazard class whole catchment can be computed by divi­ In the original application of the unit stream
can be determined using data obtained from ding the total river length against the total power, the instantaneous discharge was used
electronic portable document files (pdf) or catchment area under consideration. in the relationship to describe the sediment

Table 4(a) Split-sample analysis for Region 1

1:10-year
River network Erodibility Calculated Observed
recurrence Average slope Effective area
ID Station name density (Rnd) index/hazard sediment sediment
interval flood (So) (river) (%) (Ae) (km 2)
(m/km 2) class (EIw) load (t/a) load (t/a)
(Q10) (m3/s)

1 Albasini Dam 195 2.33 0.0448 500 4.72

2 Bospoort Dam 210 1.78 0.1195 580 3.71

3 Buffelspoort Dam 80 3.29 0.2186 116 4.95

4 Cross Dam 155 4.00 0.1322 301 6.89

5 Doorndraai Dam 150 2.05 0.1283 386 5.44

6 Hans Strijdom Dam 480 2.63 0.1506 4 328 6.16

7 Hartebeespoort Dam 615 2.76 0.1318 3 473 6.20

8 Klein-Maricopoort Dam 200 1.68 0.1527 827 4.03

9 Klipvoor Dam 220 1.33 0.1138 4 708 3.51

10 Koster Dam 110 1.72 0.0901 289 3.15

11 Kromellenboog Dam 235 2.09 0.1173 607 3.69 37 079 39 472

12 Lehujwane Dam 115 1.22 0.1502 200 3.23 31 554 21 102

13 Madikwe Dam 150 1.79 0.1533 313 3.52 26 770 19 746

14 Marico-Bosveld Dam 170 3.16 0.0938 948 4.73 17 962 59 334

15 Mzhelele Dam 530 5.00 1.1160 832 7.00 101 315 230 847

16 Olifantsnek Dam 510 2.36 0.1026 499 3.93 119 249 50 513

17 Roodeplaat Dam 229 2.05 0.1531 689 4.94 65 737 65 418

18 Vaalkop Dam 770 1.39 0.1244 3 917 3.33 183 235 231 148

34 Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
Table 4(b) Split-sample analysis for Region 2
1:10-year
River network Erodibility Calculated Observed
recurrence Average slope Effective area
ID Station name density (Rnd) index/hazard sediment sediment
interval flood (So) (river) (%) (Ae) (km 2)
(m/km 2) class (EIw) load (t/a) load (t/a)
(Q10) (m3/s)

1 Blyderivierspoort Dam 580 7.78 0.1342 1 235 7.78

2 Bronkhorstspruit Dam 380 1.44 0.1385 1 244 3.53

3 Buffelkloof Dam 160 4.92 0.1768 279 7.00

4 Corumana Dam 2 042 2.60 0.1578 6 271 3.30

5 Da Gama Dam 115 4.87 0.3262 44 2.55

6 Ebenezer Dam 65 4.36 0.2152 126 7.60

7 Hans Merensky Dam 258 8.11 0.1202 89 5.06

8 Klaserie Dam 195 3.82 0.1725 168 5.39

9 Klipkoppie Dam 195 4.11 0.1823 77 2.51

10 Kwena Dam 270 4.49 0.1215 950 5.90

11 Langmere Dam 119 3.65 0.2263 32 3.00

12 Loskop Dam 1 180 3.30 0.1305 3 973 5.91

13 Magoebaskloof Dam 110 10.13 0.1189 81 6.20

14 Massingir Dam 5 309 2.62 0.1223 63 350 4.81 4 786 006 15 520 750

15 Middel Letaba Dam 715 2.92 0.1331 1 051 5.84 48 752 547 048

16 Nooitgedacht Dam 245 1.46 0.1220 1 583 4.07 140 983 193 161

17 Ohrigstad Dam 35 10.38 0.0948 85 6.57 2 526 3 370

18 Rietfontein Dam 102 1.11 0.2147 86 3.40 16 360 4 608

19 Rietspruit Dam 100 0.86 0.1467 392 4.13 41 062 24 444

20 Rust de Winter Dam 125 1.83 0.1128 1 127 4.84 76 287 27 955

21 Trichardsfontein Dam 25 1.00 0.1000 11 3.00 688 779

22 Tzaneen Dam 310 5.33 0.1196 284 5.03 13 429 70 552

23 Vlugkraal Dam 40 4.87 0.2200 14 6.00 1 153 650

24 Vygeboom Dam 460 5.06 0.1477 1 541 5.59 109 221 84 761

25 Witklip Dam 65 6.81 0.1634 60 2.00 18 468 24 293

transport. However, when considering Table 5 Results of split-sample predictive accuracy analysis
sediment load over a long period of time,
Obs Percentage of the data in this range
an effective discharge passing through a Region
point along a river or a reservoir would best N 0.67 < xi < 1.5 0.5 < xi < 2.0 0.33 < xi < 3.0

be represented by a recurrence interval 1 8 63 75 88

flood, since more sediment is transported
during floods than the mean runoff from
the catchment.
Homogeneous regions and
catchment areas
The sediment load must be related to the
catchment area in order to account for
homogeneity. It was therefore decided that
the regression analysis should include the
parameter of the catchment size as one of
the variables.
Confidence intervals
The prediction of the sediment load using
the empirical method also requires con-
fidence limits. The equations in Table 3
2 12
calculate the value of the predicted sediment
load around an assumed regional mean,
which is at 50% confidence limit. The
multiplication factors for the other required
confidence limits for Regions 1, 2, 4, 5, 7 and
8 can also be obtained from the appendices
in the Water Research Commission Report
(Msadala et al 2012).
Verification and analysis of results
The sample was split, and one portion of
the sediment load was used for independent
verification of the results of the empirical
model. The verification exercise using the
42 58 75
split-sample method in the two regions
(Regions 1 and 2) that had relatively larger
sample sizes showed that the predictive accu-
racy of the empirical method was relatively
good. Tables 4a and 4b show the data that
was used for split-sample analysis in Regions
1 and 2. Ten observations were applied to
derive a regression equation for Region 1. The
results were used to predict the sediment
loads for the remaining eight independent
observations within the same region. A simi-
lar procedure was done for Region 2 and the
summary of split-sample analysis results for
both Regions 1 and 2 is shown in Table 5.

Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017 35
 The empirical equations in Table 3
were derived from all the observations in
the given region. The predictive accuracy
checks per region were done by way of
inspection of the discrepancy ratio. Table 6
shows the results of the discrepancy ratio
test for each region.
The discrepancy ratio results in Table 6
were analysed and found to be within
acceptable predictive accuracy limits and
are satisfactorily good when one considers
the general erratic behaviour of fine sedi-
ment transport and the ranges of predictive
accuracy that were achieved in the previous
methodology (Rooseboom et al 1992). In
some regions, there were outliers, but overall
they were checked to have no significant
effect on the overall predictive accuracy of
the regression equations.
CONCLUSIONS AND
RECOMMENDATIONS
Methods for the prediction of sediment yields
in ungauged catchments were d ­ eveloped. The
accuracy of predictions using the revised sedi-
ment yield methods is higher than those of
Rooseboom et al (1992).
The proposed methods are based on long-
term average sediment yields, and therefore
regional confidence limits need to be applied
to scale up (or down) the predicted sediment
yields to typically expected values based
on expert knowledge and knowledge of the
ungauged catchment under investigation.
Realistic predictions are possible using both
the probabilistic and empirical methods in
their respective regions. However, where
observed data is available, it is always recom­
mended to use observed data, since the
regional probabilistic and empirical methods
have limited predictive capability due to the
range of calibration data and conditions from
which they were derived.
There is need for continued monitoring
and measurement of sediment load at exist-
ing and additional sampling stations in order
to accumulate sufficient long-term data for
calibration and validation during mathemati-
cal modelling in the future.
Based on the prediction accuracy of the
sediment yield prediction methods discussed
in this paper it is proposed that the 95%
confidence limit is considered in sediment
36 Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
Table 6 Discrepancy ratio results for the empirical method
Region Obs n
0.67 ≤ xi ≤ 1.5
1 18
2 25
4 30
5 12
7 14
8 9
yield prediction of ungauged catchments for
the current development scenario.
The proposed sediment yield prediction
methods of this paper are based on historical
data and therefore do not account for future
land use change with possible land degrada-
tion, or climate change impacts.
ACKNOWLEDGEMENTS
The authors would like to acknowledge
the the Water Research Commission of
South Africa for funding the research, and
the Department of Water and Sanitation
of South Africa for providing data. The
Institute for Water and Environmental
Engineering, in the Department of Civil
Engineering at Stellenbosch University,
carried out this research project in conjunc-
tion with the Institute for Soil, Climate
and Water of the SA Agricultural Research
Council who were appointed as part of the
project for specialist inputs on soil erosion
hazard maps and Geographical Information
Systems (GIS).
REFERENCES
Basson, G R & Di Silvio, G 2008. Erosion and sediment
dynamics from catchment to coast. UNESCO,
International Hydrological Programme, Technical
Documents in Hydrology, No. 82.
DWS (Department of Water and Sanitation) 2015. Dam
list. Pretoria: Ministry of Water and Environmental
Affairs.
Hazen, A 1914. Discussion on “ flood flows” by W E
Fuller. Transactions of the American Society of Civil
Engineering, 77: 626–632.
ICOLD (International Commission on Large Dams)
2009. Sedimentation and sustainable development of
Percentage of data in this range
0.5 ≤ xi ≤ 2 0.33 ≤ xi ≤ 3
56 83 94
48 68 88
40 53 80
42 67 92
57 84 95
57 71 93
dams in river systems. ICOLD Bulletin, Paris, France:
ICOLD.
Knofczynski, G T & Mundfrom, D 2008. Sample sizes
when using multiple linear regression for prediction.
Educational and Psychological Measurement, 68,
431–442.
Le Roux, J J, Morgenthal, T I, Malherbe, J, Smith, H J,
Weepener, H L & Newby, T S 2008. Water erosion
prediction at a national scale for South Africa. Water
SA, 34(3): 1–10.
Midgley, D C, Pitman, W V & Middleton, B J 1994.
Surface water resources of South Africa (WR90).
Volumes I–VI (Appendices) and Volume I–VI
(Maps), WRC Report Nos 298/1.1/94, 298/1.2/94,
298/2.1/94, 298/2.2/94, 298/3.1/94, 298/3.2/94,
298/4.1/94, 298/4.22/94, 298/5.1/94, 298/5.2/94,
298/6.1/94, 298/6.2/94, Pretoria: Water Research
Commission.
Morgenthal, T L, Malherbe, J, Van Zyl, D & Newby, T S
2006. Evaluation of MODIS satellite data as an
assessment tool to audit the state of agricultural
resources. ISCW Report No. GW/A/2006/72,
Pretoria: ARC Institute for Soil, Climate and Water.
Msadala, V P, Basson, G R, Rooseboom, A, Le Roux, J &
Gibson, L 2012. Sediment yield prediction for South
Africa, 2010 edition. WRC Report No. K5/1765.
Pretoria: Water Research Commission.
Renard, K G, Foster, G R, Weesies, G A & Porter, J P
1991. RUSLE: Revised Universal Soil Loss Equation.
Journal of Soil and Water Conservation, 46(1): 30–33.
Rooseboom, A, Verster, E, Zietsman, H L & Lotriet, H H
1992. The development of the new sediment yield
map of South Africa. WRC Report No. 297/2/92,
Pretoria: Water Research Commission.
Strand, R I & Pemberton E L 1982. Reservoir
sedimentation technical guidelines for Bureau
of Reclamation. Denver, CO: U.S. Bureau of
Reclamation, 48.
Yang, C T 1996. Sediment Transport: Theory and
Practice. New York: McGraw-Hill.
The two-points condensation TECHNICAL PAPER
technique (TPC) for Journal of the South African
Institution of Civil Engineering
detection of structural ISSN 1021-2019
Vol 59 No 2, June 2017, Pages 37–48, Paper 1400

damage due to vibration DR AHMED AL-QAYYİM is a lecturer at the

University of Babylon. He obtained his
Bachelor’s and Master’s degrees in Civil
A N Al-Qayyim, B Ö Çağlayan Engineering from the University of Babylon in
Iraq, and his PhD from Istanbul Technical
University. His field of work includes structural
In recent years, damage detection, as determined by variations in the dynamic characteristics or health monitoring, signal processing, damage
detection of structures, system identification,
response of structures, has received considerable attention in the literature. This paper proposes
and nonlinear analysis of structures. His experience includes assessments
a new damage identification technique that identifies damage location. A methodology termed for more than 40 railway bridges, as well as design, construction, and
the Two-Points Condensation Technique (TPC) is presented. It uses identification of stiffness management of large structures for water and wastewater treatment
matrix terms to assess damage, based on the incomplete measurement of captured vibration plants, bridges and multi-storey buildings.
test data. This study identifies damage using free vibration test data in the time domain. Most Contact details:
other techniques used at present are based on data in the frequency domain. The TPC method Engineering Affairs Department
uses a set of matrices by reducing the structural system to a two-degrees-of-freedom system University of Babylon
and then compares the identified coefficients of the stiffness matrices with the coefficients of Babylon
Iraq
the theoretically condensed stiffness matrices. The damage location is obtained by observing
T: +964 780 041 8990
the change in value of the stiffness coefficients of the two-degrees-of-freedom systems. For E: ahmed_alqayyim@yahoo.com
the computation, an optimisation uses a program written in MATLAB code. The code can be
executed both under the MATLAB and Octave environments. The TPC technique is applied to PROF BARLAS ÇAĞLAYAN is Assistant Professor
experimental data obtained from a steel beam model structure after introducing a thickness and senior lecturer and researcher at Istanbul
change in one element. Two case studies are considered. In both cases, the method accurately Technical University (ITU). He obtained his
detects the damage, and determines its location. In addition, the results illustrate that observing BEng, MSc and PhD degrees in Civil Enginering
from ITU. He gained experience in testing
changes in the stiffness matrix coefficients can be a useful tool for monitoring structural health.
structures, system identification and structural
As the procedure proposed here is in a time domain, to eliminate time-consuming calculations fatigue through the NATO Science for Stability
this procedure is suitable for structures that are not continuously monitored, but are monitored Programme research projects between 1990
within scheduled time periods. and 1996. Since 1992 he has tested more than 200 railway bridges and
building structures in Turkey and abroad. The focus areas of his work over
the past decade include structural health monitoring, damage detection,
earthquake engineering and post-earthquake structural
INTRODUCTION of engineering structures. The modification damage evaluation.
Changes in the material and/or geometric to material properties or structural physical
Contact details:
properties of a structural or mechanical parameters can also be considered as damage Department of Civil Engineering
system, including the changes in the bound- (Tadeusz et al 2013). Istanbul Technical University
ary conditions and system connectivity, are The main damage detection approaches Sariyer – Istanbul 34469
defined as ‘damage’ that adversely affects may have some limitations. For example, the Turkey
T: +90 212 285 6561
the current or future performance of that mode shape approach is only sensitive in the
E: caglayan@itu.edu.tr
system. Implicit in this definition is a com- case where the measurement point is close to
parison between two different states of the the node points (see Figure 4) for a particular
system (Farrar & Doebling 1999). mode (Liang & Chan 2009).
Structural health monitoring (SHM) is a Salawu and Williams (1995) conducted
procedure targeted at providing accurate and modal tests of a full-scale bridge before and
real-time information about the performance after rehabilitation, and determined that
and health state of a structure. SHM includes the variation in natural frequencies of the
an approach to the process of characteris- bridge was not only due to structural repairs,
ing and detecting damage of engineering although both modal assurance criterion
structures. The objective of SHM is to (MAC) and the coordinate modal assurance
monitor the in situ behaviour of a structure criterion (COMAC) performed well to indi-
accurately and efficiently, to evaluate its cate the location of the repairs.
performance under various service loads, to Doebling and Farrar (1996) pointed
discover areas subject to damage or weaken- out that the frequency shift method has
ing, and to determine the health or condition considerable practical limitations for civil
of the structure (Czichos 2013). In the area structures, because it is insensitive to dam-
of SHM, the term ‘damage’ in mechanical age unless when there is severe damage or
structures can be understood intuitively as an accurate measurement is applied. On the
denoting a defect or failing that impairs the other hand, the frequency response function Keywords: damage detection, two-points condensation,
functional behaviour and working conditions approach is insensitive to the damage unless structural health monitoring, signal processing, optimisation

37
Al-Qayyim AN, Çağlayan BÖ. The two-points condensation technique (TPC) for detection of structural damage due to vibration.
J. S. Afr. Inst. Civ. Eng. 2017:59(2), Art. #1400, 12 pages. http://dx.doi.org/10.17159/2309-8775/2017/v59n2a5
severe damage has occurred or accurate
measurement of vibration was done. The
Structural drawings
inverse method is at present still only being
investigated theoretically (Friswell 2008).
For the curvature/strain modes approach,
Pandey et al (1991) demonstrated that the Mass Mo Stiffness Ko
damage to beam structures can be identi-
fied using mode shape curvature. Chance et
al (1994) found the measured strain mode Reduction to 2DOF
shape to be much more feasible for damage
localisation. But recording strain over the Experimental test
whole structure is not yet feasible with cur-
rent technology; therefore, a new methodol- Mr Kr

ogy and equipment are needed.

Stubbs et al (1992) presented the pioneer-
ing work on using modal strain energy for
damage localisation. However, when the
damage is located in a structural element
that is not sensitive to the modal parameter
changes, the modal strain energy approach
cannot detect the damage in such an element
in a structure. On the other hand, for the
dynamic flexibility approach based on higher
vibration modes rather than lower modes, a
large number of dynamic modes are needed
to find a stiffness matrix estimation or its
changes, and mainly the higher modes need
to be measured. However, measuring the
higher frequency response is very difficult
to do because of practical limitations (Sinou
2009). From a testing standpoint, exciting
the higher frequency response of a structure
requires more energy to produce a measur-
able response at these higher frequencies
than at the lower frequencies. Koh et al
(2006) demonstrated a method called con-
densed model identification for identification
of full stiffness matrices for damage assess-
ment based on incomplete measurement.
They used three types of reduction methods,
namely static condensation, dynamic con-
densation, and System Equivalent Reduction
Expansion Process (SEREP) methods.
Pokharkar and Shrikhande (2010) used
the same Koh et al (2006) approach with the
mass-invariant constraint for the identifica-
tion of condensed models in different time
windows. Both studies used the input and
output data to develop mathematical models
to characterise the input-output behaviour
of an unknown system by means of experi-
mental data, which was acceleration within
their studies.
For all the above-mentioned methods,
input and output data are needed to identify
the location and severity of damage, while the
present study indicates the damage and its
location for incomplete measurement using
the output data only. The present method
requires fewer sensors than the above-­
mentioned methods. The method is also a
useful tool in that it evaluates the structure
using real-time acceleration measurements.
Acceleration
signal
üi
Velocity
u̇i
signal
ui
Displacement
signal
Measuring the frequencies
Figure 1 The main steps of the TPC technique
The two-points condensation (TPC)
technique reduces the whole structure
stiffness matrix to a set of two-degrees-
of-freedom matrices. The identified stiff-
ness matrices Kc are compared with the
undamaged reduced stiffness matrices Kr.
By observing the changes in the stiffness
matrix coefficients of the two-degree-of-
freedom systems, the damage location
can be obtained. In the TPC technique,
the identified stiffness matrix Kc can be
obtained by optimising the equation of
motion using the measured test data. The
stiffness matrix of the undamaged structure
Kr is obtained by reducing the theoretical
stiffness matrix of the structure.
In this study, the theoretical stiffness
matrix Ko is generated using data from the
design drawings of the structure. SEREP is
used to condense the matrices required in
the TPC technique. To verify the efficiency
of the technique, two cases are studied:
Damping
Optimisation
Kc
No
Are the
frequencies
OK?
Yes
Comparison
Damage detection
■■ For the first case, the influence of the
position of the vibration sensors relative
to the damage location is studied.
■■ For the second case, the sensitivity of the
technique to the size of the damage is
studied.
TWO-POINTs CONDENSATION
(TPC) TECHNIQUE
The TPC technique is an analytical method
that detects structural damage by observing
the value of changes in the stiffness matrix
coefficients. The technique compares the
reduced theoretical stiffness matrix Kr
with the identified stiffness matrix Kc. For
modelling the beam in the MATLAB/Octave
environment, the consistent mass matrix and
the Euler–Bernoulli beam element stiffness
matrix are used (Reddy 2006).
The theoretical stiffness matrix Ko is
obtained using the as-built drawings of a

38 Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
structure. The reduced theoretical stiffness
Kr is generated by condensing the theoretical
stiffness matrix Ko to a two-degrees-of-
freedom (2-DOF) stiffness matrix. The selec-
tion of these degrees of freedom corresponds
to the vibration acceleration measurement
sensor locations.
In this technique the SEREP method is
used to reduce the stiffness and mass matrix
to 2-DOFs. The SEREP condensation has the
best computational performance and leads
to smaller errors in the identification of stiff-
ness values (Koh et al 2006).
The TPC technique calculates the
identified stiffness matrix Kc by finding
the optimal solution of the equations of
motion, which should correspond to the
real system properties (the mode shapes and
frequencies).
Mr{ü}2×1 + [Cr]2×2{u̇}2×1 + [Kr]2×2 {u}2×1 = 0 (1)
Where:
Kr = The stiffness matrix of the reduced
system
Mr = The mass matrix of the reduced system
Cr = The damping matrix of the reduced
system
ü = The acceleration vector
u̇ = The velocity vector
u = The displacement vector.
In this technique, the input data includes the
reduced mass matrix Mr and initial stiffness
matrix Kr in addition to the corresponding
vectors of acceleration ü, velocity u̇, and dis-
placement u. The first and second measured
modal frequencies are part of the input data
and are used to calculate the damping coeffi-
cients and control the solution of finding the
identified (reduced) stiffness matrix.
The technique uses the theoretical mass
matrix of the structure because it does not
generally change (Pokharkar & Shrikhande
2010). The damping matrix C is calculated
according to Rayleigh damping. The damp-
ing matrix is optimised due to the stiffness
updating. The optimisation solver minimises
the function given in Equation 2, which is
based on the equation of motion.
q = sum [0 – [[Mr]{ü} + [α[Mr] + β[K]]{u̇}
+ [K]{u}]]2(2)
Where α and β are the damping coefficients
(see “Damping matrix” section below). This
technique uses the multi-objective function
solver to find the optimal solution of the
function given in Equation 2. MATLAB code
was developed to analyse the beam based on
the steps outlined in Figure 1.
The comparison of the identified stiffness
matrices Kc and reduced theoretical stiffness
Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
matrices Kr is used to locate the damage.
Large changes in the stiffness coefficients
of the 2-DOF matrices indicate the location
of the damage. The changes in the stiffness
matrix are shown in Equation 3.
∆Ki–j = ∆kii ∆kij (3)
∆k ji ∆k jj
Where:
∆Ki–j = the change matrix of the set of nodes
i and j
∆kii = the difference in the coefficients of
the Kc matrix and the Kr matrix at
position (i, i)
∆kij = the difference in the coefficients of
the Kc matrix and the Kr matrix at
position (i, j)
∆k ji = the difference in the coefficients of
the Kc matrix and the Kr matrix at
position (j, i)
∆k jj = the difference in the coefficients of
the Kc matrix and the Kr matrix at
position (j, j).
When large changes are observed by com-
paring ∆kii with ∆k jj, the location of the
damage can be determined. If ∆kii is greater
than ∆k jj, it means that the damage is located
near to the node i.
Analytical reduction of system
matrices using SEREP method
From an analytical approach standpoint, the
finite element method assumes that a contin-
uous structure can be discretised by describ-
ing it as an assembly of finite elements, each
with a number of boundary points that are
commonly referred to as nodes. The main
problem to overcome in SHM and damage
detection is the typical mismatch of the
selected number of degrees of freedom of an
analytical and an experimental representa-
tion of a structural dynamic system.
For damage detection, the concept of
model reduction (or, alternatively, model
expansion) plays an important role. Using
condensation or expansion, it is possible to
compare a large analytical set of DOFs to a
relatively small set of experimental DOFs.
Reduction and expansion also play a very
important role with regard to model updat-
ing. Consequently, the set of the tested DOFs
requires reducing the number of DOFs of a
large model without losing any information
or characteristics of the dynamic system in
the modelling process.
The SEREP condensation method parti-
tions the degrees of freedom into a set of
slave DOFs and master DOFs. The DOFs are
arranged to place the slave DOFs as the first
s coordinates, while the remaining master
DOFs are the last m coordinates (Friswell
& Motiershead 1995). Here the coordinates
represent the location of sub-matrices in the
original matrix.
The reduction of the stiffness matrix
is thus accomplished by identifying those
degrees of freedom to be condensed or
reduced as slave degrees of freedom, and to
express them in terms of remaining master
degrees-of-freedom.
The dynamic equations of equilibrium
for an undamped n degree-of-freedom model
may be written as:
M ü(t) + K U(t) = F(t)(4)
Where: ü(t) and U(t) are the acceleration and
displacement response vectors. The displace-
ment response vector U(t) in Equation 4 can
be expressed as shown in Equation 5 using
the mode superposition method:
U(t) = Φ q(t)(5)
in which Φ is the complete eigenvector matrix
of the full model, and q(t) is the modal coor-
dinate vector. It is well known that the com-
putation of the complete eigenvector matrix
is not required for a large model. Therefore,
modal truncation is usually used in the mode
superposition technique (Qu 2004). If p eigen-
vectors of the full model are used in the mode
superposition, Equatioin 5 is rewritten as:
U(t) = Φpq p(t)(6)
With the arrangement of the total degrees of
freedom, Equation 6 may be partitioned as:
U(t) =  Um(t)  = Φmp q p(t)(7)
 
U
 s (t)  Φsp
This is equivalent to two equations (8 and 9):
Um(t) = Φmpq p(t)(8)
Us(t) = Φspq p(t)(9)
Equation 8 provides a description of the
displacement responses at the master DOFs
in terms of the eigenvector matrix at these
DOFs. The sub-matrix Φmp is generally not a
square matrix. Since the number of knowns
in Equation 8 are greater than the number of
unknowns, Equation 8 can be put into a nor-
mal form by transforming this equation as:
Yp(t) = ΦTmpUm(t)(10)
Substituting Equation 8 into Equation 10
produces:
Yp(t) = ΦTmpΦmp q̌ p (t)(11)
39
in which q̌p(t) is an approximate solution of
qp(t) (O’Callahan et al 1989). Although the
square coefficient matrix of q̌p(t) will in gen- 1
eral be of full rank and possess an inverse, 2ζ
the determining of the inverse of this matrix 1
using standard methods may encounter 2√2ζ
some numerical difficulty, and singular-value

Amplification
decomposition solution is usually required.
Symbolically q̌ p(t) could be solved from
Equation 11 as:

q̌ p(t) = Yp(t)[ΦTmpΦmp]–1(12) 1

Substituting Equation 10 into Equation 12

produces the general form of the solution of
the modal coordinates in terms of physical
coordinates and modal matrix as:
0
ω1 ωn ω2
q̌ p(t) = Φ+mpUm(t)(13)
∆ω
Where Φ+mp is the generalised inverse of Frequency
matrix Φmp and is defined as:
Figure 2 The half-power bandwidth method for finding damping ratios
Φ+mp = ΦTmp [ΦmpΦTmp]–1(14)
Kr = TTKT(21) The coefficients α and β can be determined
Equation 13 represents the “best” solution from damping ratios of the ith and jth modes
of the p variables given in Equation 8. For respectively: ξi and ξj respectively. Expressing
convenience, the solution q̌ p(t) of Equation 8 Mr = TTMT(22) Equation 24 for these two modes in matrix
can be approximated by q p(t), (Qu 2004): form leads to:
Due to its significant numerical computa-
q p(t) = Φ+mpUm(t)(15) tional advantages, the SEREP condensation 1 1/ω ωi  α  =  ξi (25)
i
   
method is often used in commercial FEM 2 1/ωj ω j  β   ξj 
Substituting Equation 15 into Equation 9 packages.
leads to: These two algebraic equations can be
Damping matrix solved to determine the coefficients α
Us(t) = ΦspΦ+mpUm(t)(16) Udwadia (2009) discusses non-proportional and β. Seeing that the damping matrix
damping in linearly damped vibrating is proportional to the mass and stiffness
systems in which the stiffness and damping matrices, the change in the damping matrix
R = ΦspΦ+mp(17) matrices are not restricted to being symmet- due to the change in the stiffness is taken
ric and positive-definite in simple systems into account.
When the dynamic condensation matrix with two-degrees-of-freedom; they conclude The damping ratios ξi and ξj for the
is available (Equation 17), the coordinate that if the system has an active element, as first mode and second mode are calcu-
transformation matrix T may be given by the commonly arises in the active control of a lated using the half-power bandwidth
following (Kammer 1987): structure, the stability is more difficult to method using Equation 26 (Silva &
physically interpret, and their approximation Clarence 2000):
T= I (18) by damping matrices that commute with the
ΦspΦ+mp stiffness matrices needs to be carried out ω2 – ω1
2ξ = (26)
with considerable care and caution. In this ωn
The coordinate transformation matrix is study, as there is no active element, propor-
obtained by substituting Equation 15 into tional damping is taken into consideration. The angular frequency ωn for each
Equation 7: The damping matrix is calculated using mode is obtained by using a Fast Fourier
Rayleigh damping where the damping is Transform technique (FFT) (Monson
 
U(t) =  Um(t)  = T um(t)(19) defined as being proportional to the mass 1996). By plotting the Fourier amplitude
 Us(t)  and the stiffness of the structure (Chopra spectra of the signals recorded, the
2012): frequencies associated with the modes of
Where: vibration of the structure are located at
Cr = αMr + βKr(23) the corresponding vibration amplitude
g  ΦmpΦ+mp  peak values.
T= Φp Φ mp =  (20)
+  The damping ratio for the nth mode of such
 ΦspΦ mp  As shown in Figure 2, the angular fre-
a system is: quencies ω2 and ω1 are obtained by finding
Using the coordinate transformation in the corresponding angular frequencies for
Equation 20, the reduced system matrices are α 1 β the amplitude that is equal to the amplitude
ξn = + ωn(24)
given by: 2 ωn 2 at ωn divided by √2.

40 Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
 6. To determine the upper limit and lower
limit of the solution, it is assumed for this
study that the stiffness degradation is
x(n) between 140% and 80%, respectively.
7. A goal function value of zero is used in
order to get the optimal value of K in the
function given in Equation 2.
8. Weighting function values of unity are
used in this study.
The technique uses the reduced theoretical
stiffness matrix as a starting point where the
solver finds the best stiffness matrix values
x(n – 1) that satisfy the equation of motion and the
associated modal frequencies.
The output of the solution process is the
identified (optimised) stiffness matrix. Kc is
x(n – 2) then used for comparison purposes to locate
the damage.

EXPERIMENTAL STUDY
The structural system used in this study is
T T a simply-supported steel beam with a span
of 2.3 m. The cross-section of the beam is a
Figure 3 The bilinear transform of an integrator rectangular section of 100 mm by 21.9 mm.
The material properties of the beam are
as follows:
g
Mass density = 7 850 kg/m3
Young’s modulus E = 205.9 GPa

g 2.30 m The experimental setup is shown in Figure 4.

450 mm 500 mm
Figure 4 A-plate welded to a simply-supported steel beam for damage detection
Signal integration
To obtain the velocity and displacement
required for the equation of motion, the
trapezoidal rule for numerical integration is
applied to acceleration data, where:
New area = area so far + increment of the area
The increment of the area is the trapezoidal
approximation, i.e. the half sum of the paral-
lel sides times the perpendicular distance
(Leis 2011), as illustrated in Figure 3.
Optimisation technique
Multi-objective optimisation is a math-
ematical optimisation technique used to
optimise more than one objective function
simultaneously. In this study, a multi-
objective optimisation technique is used to
optimise the equation of motion used in the
TPC technique.
The problem of reducing a set of
nonlinear functions Fi(x) subject to a set
3.8 mm
21.9 mm
100 mm
The cross-section g-g
of goals Fi* is addressed by the MATLAB/
Octave code developed. The maximum
of Fi(x) – Fi* is minimised by the unscaled
goal attainment problem. The algorithm
coded is used to find x to minimise the
maximum of [(Fi(x) – Fi*)/‌wi], where the
weighting variables w ‌ i are a given positive
value. The solution is controlled by the
given lower and upper limits of the solution,
where lower limit ≤ x ≤ upper limit
(Edward et al 2011). The TPC technique
finds the best stiffness value to satisfy the
function given in Equation 2, as well as the
corresponding frequencies. The initial inputs
are as follows:
1. The reduced theoretical stiffness matrix.
2. The reduced mass matrix.
3. The acceleration, velocity and displace-
ment vectors.
4. The first mode and second modal
frequencies.
5. The damping ratios of the first two modal
frequencies.
Damage in the form of stiffness degrada-
tion is introduced into the beam by welding a
plate (100 mm × 500 mm) with a thickness of
3.8 mm at a distance of 450 mm from the left
beam support. The material properties of the
plates are the same as the material properties
of the beam. The change in thickness leads to
change in the stiffness of a part of the beam.
For modelling the beam in the MATLAB/
Octave environment, the consistent mass
matrix and the Euler–Bernoulli beam ele-
ment stiffness matrix are used (Reddy 2006).
In order to apply the technique, the beam
is divided into five segments, as shown
in Figure 5(a). The damaged segment is
located between points 2 and 3, as shown in
Figure 5(b).
The DOFs considered in the model are
vertical displacements and rotations at each
node. The DOFs (θ1, θ 2, θ 3, θ4, θ 5 and θ 6)
refer to the rotations, where the DOFs (v 2, v3,
v4 and v5) refer to the vertical displacements.
For the damaged beam prototype, four
accelerometers are placed at points 2, 3, 4
and 5 of the beam, as shown in Figure 5(c).
The mass of each accelerometer including
the clamping apparatus is 0.88 kg. These
masses are added to the global mass matrix
of the system as lumped masses at the nodes
where the accelerometers are located. It is
essential to collect sufficient data to establish

Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017 41
 θ1 θ2 v2 θ3 v3 θ4 v4 θ5 v5 θ6

A 1 2 3 4 5 6 B
1 2 3 4 5

Ra Rb

450 mm 500 mm 450 mm 450 mm 450 mm

 (a) The undamaged steel beam model A-B with five elements

θ1 θ2 v2 θ3 v3 θ4 v4 θ5 v5 θ6

A 1 2 3 4 5 6 B
1 2 3 4 5

Ra Rb

450 mm 500 mm 450 mm 450 mm 450 mm

(b) The damaged steel beam model A-B with five elements

(c) The simply-supported steel beam model specimen with accelerometers placed on the beam

v θ
Node points Vertical DOF Rotational DOF

Figure 5 ( a) An undamaged steel beam model A-B with five elements, (b) the damaged steel beam model A-B with five elements, (c) the simply-
supported steel beam model specimen with accelerometers placed on the beam

Table 1 The reduced stiffness and mass matrices using SEREP Table 2 D
 ifferences between the damaged and the theoretical
undamaged stiffness matrices for the sample steel beam
DOFs that K Theoretical reduced Theoretical reduced
is reduced to stiffness matrix (N/m) × 10 4 mass matrix (Kg) Identified changes in
DOFs that K Theoretical changes in the
stiffness matrix using
reduced to stiffness matrix (ΔK%)
6.700 –3.715 1.817 0.075 TPC technique (ΔK%)
[v 2 – v4]
–3.715 2.850 0.075 1.744
28.13 16.63 27.37 19.13
[v 2 – v4]
16.63 10.71 19.13 10.19
3.778 –2.694 2.297 1.056
[v 2 – v5]
–2.694 3.778 1.056 2.313 25.19 13.19 24.39 14.18
[v 2 – v5]
13.19 4.43 14.18 4.24
8.721 –8.479 2.366 –1.062
[v 3 – v4] 25.81 18.96 24.59 19.86
–8.472 9.021 –1.062 2.218 [v 3 – v4]
18.96 15.30 19.86 14.58
3.062 –4.137 1.861 –0.089 18.99 10.57 18.06 11.70
[v 3 – v5] [v 3 – v5]
–4.2137 7.447 –0.0897 1.831 10.57 5.39 11.70 5.17

the correct mass matrix by visual inspec-
tion of the model and to use the applicable
shop drawings.
The motion of the beam is initiated
by using a pullback-quick release method
by tying a rope to the middle of the beam
span and then cutting it after having pulled
it downwards.
Figures 6(a) and 6(b) show the accelera-
tion data captured during the test, as well
as the velocity and displacement signals
obtained by integrating the acceleration
signals obtained from the four channels
using the trapezoidal method. The accelera-
tion data is collected at a sampling rate of
500 Hz.
To apply the damage detection algorithm
developed, as described in this paper, the

42 Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
 (a) The recorded acceleration signals from four channels for five-element steel beam

1 000

500
Acceleration (mm/sec2)

–500

–1 000

–1 500

5 10 15 20 25 30
Time (sec)

(b) The velocity signals for five-element steel beam

15
10
Velocity (mm/sec)

5
0
–5
–10
–15
5 10 15 20 25 30
Time (sec)

(c) The displacement signals

0.2
Displacement (mm)

0.1

–0.1

–0.2

5 10 15 20 25 30
Time (sec)

Figure 6 ( a) The recorded acceleration signals from four channels for five-element steel beam, (b) the velocity signals for five-element steel beam,
(c) the displacement signals

theoretical stiffness matrix Ko and mass
matrix M of the system are reduced to
two DOFs using the SEREP method. The
DOFs are selected to cover all the pos-
sible combinations of vertical DOFs. The
selected DOFs are v 2 – v4 , v 2 – v5, v 3 – v4 ,
and v 3 – v5. The resultant matrices for the
condensed structural system are given in
Figure 7.
Using the SEREP condensation method,
the reduced stiffness and mass matrices are
produced for each set of two DOFs, as shown
in Table 1.
By plotting the Fourier amplitude spectra
and using Equations 23 through 26, the
damping matrix, frequencies and damping
ratios are calculated. Figure 8 shows the
Fourier amplitude spectra for the first three
frequencies for the measured values for the
four channels of the five-segment beam
model. The frequency of the first mode for
all nodes is 9.4 Hz, while the damping ratio is
0.0185. The second mode has a frequency of
38 Hz, and the damping ratio is 0.0681.
The identified stiffness matrix Kc , which
is obtained as output from the optimisation,
is compared with the reduced theoretical

Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017 43
 v2 v4
Kv v Kv v
1 2 3 4 5 6 Kv = 2 2 2 4
2 –v4 Kv v Kv v
1 2 3 4 5 4 2 4 4 2x2

v2 v5
Kv v Kv v
1 2 3 4 5 6 Kv = 2 2 2 5
2 –v 5 Kv v Kv v
1 2 3 4 5 5 2 5 5 2x2

v3 v4
Kv v Kv v
1 2 3 4 5 6 Kv = 3 3 3 4
3 –v4 Kv v Kv v
1 2 3 4 5 4 3 4 4 2x2

v3 v5
Kv3v3 Kv3v5
1 2 3 4 5 6 Kv3 –v5 =
Kv5v3 Kv5v5 2x2
1 2 3 4 5

Figure 7 Reduced stiffness and mass matrices of the beam into sets of 2 DOF systems

stiffness matrices Kr . The comparison is

14
based on Equation 3. The results of the
comparison between (Kc) and (Kr) are pro-
A ch-1 ch-2 ch-3 ch-4 B
vided in Table 2. The solution is checked by
comparing the frequencies computed from 12

the analytical model to that of the physical

structural system (see Figure 1).
10 ch-3
The results in Table 2 show that the
large percentage changes in coefficients of 14
the stiffness matrix occur at coefficients of ch-2
Amplitude (×10 4)

nodes that are near to the location of the 8

damage. ch-4
For the condensed stiffness matrix Kv 2–v4
using DOFs (v 2 – v4), the percentage change 6 ch-1
in coefficient k v 2v 2 is larger than the change
in coefficient k v4v4 . This indicates that the
damage is located near to node 2 rather than 4
near to node 4. The same conclusion can
be established from ΔK for DOFs v 2 – v4,
v 2 – v5, v3 – v4, and v3 – v5 ( see Table 2 2
column 1).

The effect of acceleration 0

sensor positioning
For this first case study, the effect of the
position of the sensors relative to the damage
location is investigated. To determine the
effect of sensor positioning, the same beam
described above is divided into six segments
of equal length (Figure 9).
Five accelerometers are placed at the
intermediate points 2, 3, 4, 5 and 6. The
damaged segment is located between points
2 and 4, as shown in Figure 9(a). The degrees
of freedom θ1, θ 2, θ 3, θ4, θ 5, θ 6 and θ 7 refer
to the rotations, while the degrees of free-
dom v 2, v3, v4, v5 and v6 refer to the vertical
displacements, as shown in Figure 9(b).
The acceleration data is collected using a
0 50 100 150
ω1 ω2 ω3
Frequency (Hz)
Figure 8 F ourier amplitude spectra for four channels of five-element steel beam where the first
three frequencies appear
sampling rate of 500 Hz. Figure 10 shows the matrix. The damping parameters are calcu-
recorded vertical acceleration signals, veloc- lated in the same way as that of the previous
ity signals, and displacement signals. These example. Table 3 shows a comparison of the
signals are for the degrees of freedom v 2, v3, results for the theoretical coefficients of the
v4, v5 and v6 respectively. condensed stiffness matrices and the coef-
In this test, the global stiffness matrix ficients calculated using the TPC technique.
is reduced to the following DOFs: ([v 2 – v4], For the first result of DOFs [v 2 – v4],
[v 2 – v5], [v3 – v4], [v3 – v5] and [v4 – v5]). both the nodes (2, 4) are located outside the
The initial value for the stiffness matrix damaged area. Node 2 is closer than node
used is the undamaged theoretical stiffness 4 to the damage area, which explains the

44 Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
 (a) The specimen with sensors

θ1 θ2 v2 θ3 v3 θ4 v4 θ5 v5 θ6 v6 θ7

A 1 2 3 4 5 6 7
1 2 3 4 5 6

383.3 mm 383.3 mm 383.3 mm 383.3 mm 383.3 mm 383.3 mm

(b) Geometrical details

Figure 9 A
 2.3 m steel beam specimen divided into six segments, where the damaged part is located between points 2 and 4: (a) the specimen with
sensors, (b) geometrical details

Table 3 P
 ercentage changes in the coefficients of the reduced stiffness matrices for the measured large change that occurs in the coefficient
and the theoretical cases for the steel beam with six segments k v 2v 2 . The result for [v3 – v4] shows the large
changes concentrated in k v3v3 rather than
Identified changes in stiffness Theoretical changes in the
DOFs that K reduced to k v4v4 because node 3 is located at the dam-
matrix using TPC technique (ΔK%) stiffness matrix (ΔK%)
aged segment, while node 4 is outside the
23.81 15.69 22.61 16.74 damaged segment.
[v 2 – v4]
15.69 11.99 16.74 11.07
The result of [v4 – v5] indicates the
[v 2 – v5]
24.04 13.01 22.63 14.21 damage at the left side of the segment; both
13.01 7.20 14.21 6.14
nodes are, however, outside the damaged
34.14 25.53 33.06 26.80 segment. The result of [v4 – v5] provides a
[v 3 – v4]
25.53 21.31 26.80 19.80 good indication of the location of damage.

31.96 21.73 30.56 22.52

[v 3 – v5]
21.73 13.88 22.52 12.92 Sensitivity of the TPC technique
for damage extent
12.42 9.92 11.26 8.77 The sensitivity of the technique to indicate
[v4 – v5]
9.92 7.38 8.77 6.30
the extent of damage is investigated by

Table 4 P
 ercentage changes in the coefficients of the reduced stiffness matrices for the measured and the theoretical cases for the steel beam with
five segments
ΔK using TPC technique (%)
DOFs that K reduced to
Plate size Plate size Plate size Plate size Plate size
100 mm × 430 mm 100 mm × 360 mm 100 mm × 290 mm 100 mm × 230 mm 100 mm × 170 mm

17.56 11.14 11.06 6.22 6.46 2.95 3.73 1.21 1.90 0.22
[v 2 – v4]
11.14 5.48 6.22 2.78 2.95 1.10 1.21 0.28 0.22 2 * 10–4

17.70 9.49 12.35 5.81 8.25 3.26 5.56 1.79 3.48 0.83
[v 2 – v5]
9.49 2.71 5.81 1.46 3.26 0.67 1.79 0.29 0.83 0.08

22.31 17.96 19.61 15.69 16.26 12.99 12.99 10.31 9.53 7.49
[v 3 – v4]
17.96 13.12 15.74 11.99 12.99 9.40 10.31 7.40 7.49 5.30

18.43 12.83 16.89 11.99 14.58 10.46 12.05 8.67 9.12 6.53
[v 3 – v5]
12.83 6.53 11.99 6.30 10.46 5.60 8.67 4.6 6.53 3.50

Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017 45
 3 000

2 000
Acceleration (mm/sec2)

1 000

–1 000

–2 000

5 10 15 20 25 30
Time (sec)
40
30
20
Velocity (mm/sec)

10
0
–10
–20
–30
–40
0 5 10 15 20 25 30 35 40 45
Time (sec)
0.8
0.6
0.4
Displacement (mm)

0.2
0
–0.2
–0.4
–0.6
–0.8
0 5 10 15 20 25 30 35 40 45
Time (sec)

Figure 10 Recorded acceleration signals, velocity signals and displacement signals captured at points 2 to 5 for the sample steel beam

reducing the size of the damaged area. The
same beam shown in Figure 5 is considered
here. For this case, the damage size is reduced
by cutting a piece off the plate. The width of
the plate remains the same, but the length
of the plate representing the damage is now
reduced to 430 mm, 360 mm, 290 mm,
210 mm and 170 mm, as shown in Figure 11.
A 500 Hz sampling rate is used for all
tests, and the global stiffness matrices are
reduced to the following DOFs: ([v 2 – v4],
[v3 – v4] and [v3 – v5]. The recorded signals
are pre-processed, and the results of five
tests using the TPC technique are shown in
Table 4.
The percentage differences of the values
of the coefficients of the stiffness matrices
indicate the stiffness degradation caused by
the structural damage. The bigger difference
between Kii and Kjj for any given i and j indi-
cates the damage location.
The results shown in Table 4 demonstrate
that the technique is successful to find the
location of the damage in the beam, even if
the damage is small. Investigating the results
of all tests for one set of points can indicate
the ability of the technique to reflect the
degradation in stiffness.
CONCLUSIONS
This study investigated the changes in the
stiffness due to the presence of structural
damage. Experimental studies using the
two-points condensation technique (TPC)
were described. The approach using TPC
represents a non-destructive test method
that uses vibration signal records to identify
the structural damage according to the mea-
sured changes in dynamic characteristics of
the structure.
The experimental results obtained from
a steel beam model structure demonstrate
the usefulness of the TPC technique. This
method has several advantages:
■■ The concept is simple and easy to apply
without using complex calculations.
■■ Using the TPC technique, the structure
is reduced to a 2-DOF structure; this
decreases the volume of data to be
captured and the volume of data to be

46 Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017

(a) 430 mm
(b) 360 mm (c) 290 mm
(d) 210 mm (e) 170 mm
Figure 11 S imply-supported five-segment beam configured for each case by reducing the length of
plate, where the lengths are: (a) 430 mm, (b) 360 mm, (c) 290 mm, (d) 210 mm, (e) 170 mm
dealt with at a time. Using more tests will Chopra, A K 2012. Dynamics of Structures: Theory and
help to obtain more precise values for Applications to Earthquake Engineering, 4th ed.
the change in stiffness coefficients and Harlow, UK: Pearson Education.
improve damage location. Czichos, H 2013. Handbook of Technical Diagnostics
■■ Accuracy is achieved in detecting the Fundamentals and Application to Structures and
location of the damage where the TPC Systems. Berlin, Heidelberg, Germany: Springer Verlag.
technique results have demonstrated Doebling, S W & Farrar, C R 1996. Computation of
good agreement with actual results. structural flexibility for bridge health monitoring
The results indicate that the TPC technique using ambient modal data. In: Lin, Y K & and Su,
is capable of indicating the damage location T C (Eds.), Proceedings, 11th ASCE Engineering
for different sizes of damage. According to the Mechanics Conference held at Fort Lauderdale, FL,
damage size sensitivity tests, the technique 20–22 May 1996. New York: ASCE Press, 1114–1117.
can show stiffness degradation. Additionally, Edward, B M, Shapour, A & Balakumar, B 2011. An
experimental results demonstrate that the Engineer’s Guide to MATLAB, 3rd ed. Upper Saddle
SEREP reduction method is suitable for use River, NJ: Pearson Education.
with the TPC technique. The damping effect Farrar, C R & Doebling, S W 1999. Damage detection
has to be accounted for to ensure accurate and evaluation. II: Field applications to large
results when applying this solution technique. structures. In: Silva, I M M & Maia, N M M (Eds.),
Finally, the TPC technique shows the Modal Analysis and Testing, NATO Science Series
location of damage regardless of the place E, Vol. 363, London: Kluwer Academic Publishers,
of the acceleration sensors or the size of the 345–378.
damage. The technique finds the location of Friswell, M I & Motiershead, J E 1995. Finite Element
damage when the acceleration sensors are Model Updating in Structural Dynamics. Dordrecht,
near to the damage, as well as when they are Netherlands: Springer Science & Business Media.
far from the damage. Friswell, M I 2008. Damage Identification using Inverse
Methods. Vol. 499 of the series CISM International
Centre for Mechanical Sciences. Vienna, New York:
REFERENCES Springer Verlag, 13–66.
Chance, J E, Worden, K & Tomlinson G R 1994. Kammer, D C 1987. Test-analysis model development
Processing signals for damage detection in using an exact modal reduction. International
structures using neural networks. Proceedings, SPIE Journal of Analytical and Experimental Modal
2191, Smart Structures and Materials, 187. Analysis, 2(4): 174–179.
Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
Koh, C G, Tee, K F & Quek, S T 2006. Condensed model
identification and recovery for structural damage
assessment. Journal of Structural Engineering, ASCE,
132: 2018–2026.
Leis, J W 2011. Digital Signal Processing Using MATLAB
for Student and Researchers, 1st ed. New York: Wiley.
Liang, W & Chan, T H T 2009. Review of vibration-
based damage detection and condition assessment of
bridge structures using structural health monitoring.
Paper presented at the 2nd Infrastructure Theme
Postgraduate Conference: Rethinking Sustainable
Development: Planning, Engineering, Design and
Managing Urban Infrastructure, 26 March 2009,
Brisbane, Australia.
Monson, H H 1996. Statistical Digital Signal Processing
and Modeling, 1st ed. New York: Wiley.
O’Callahan, J, Avitabile, P & Riemer, R 1989. System
equivalent reduction expansion process (SEREP).
Proceedings, 7th International Modal Analysis
Conference, Las Vegas, NV, 30 January – 2 February
1989, 29–37.
Pandey, A K, Biswas, M & Samman, M M 1991. Damage
detection from changes in curvature mode shapes.
Journal of Sound and Vibration, 145(2): 321–332.
Pokharkar, P V & Shrikhande, M 2010. Structural
health monitoring via stiffness update. ISET Journal
of Earthquake Technology, 47(1): 47–60.
Qu, Z-Q 2004. Model Order Reduction Techniques
with Applications in Finite Element Analysis, 1st ed.
London: Springer Verlag.
Reddy, J N 2006. An Introduction to the Finite Element
Method, 3rd ed. New York McGraw-Hill.
Salawu, O S & Williams, C 1995. Bridge assessment
using forced-vibration testing. Journal of Structural
Engineering, 121(2): 161–173.
Silva, D & Clarence, W 2000. Vibration Fundamentals
and Practice. Boca Raton, FL: CRC Press.
Sinou, J J 2009. A Review of Damage Detection and
Health Monitoring Mechanical Systems from
Changes in the Measurement of Linear and Non-
linear Vibrations. Hauppauge, NY: Nova Science
Publishers, 643–702.
Stubbs, N, Kim, J T & Topole, K 1992. An efficient and
robust algorithm for damage location in offshore
platforms. Proceedings, ASCE 10th Structures
Congress, 13–15 April 1992, San Antonio, TX,
543–546.
Tadeusz, S, Tadeusz, U & Wieslaw, S 2013. Advanced
Structural Damage Detection. New York: Wiley.
Udwadia, F E 2009. A note on nonproportional
damping. Journal of Engineering Mechanics, 135(11):
1248–1256.
LIST OF ACRONYMS
COMAC Coordinate Modal Assurance
Criterion
DOF Degrees Of Freedom
FFT Fast Fourier Transform
FEM Finite Element Method
MAC Modal Assurance Criterion
SEREP System Equivalent Reduction
Expansion Process
47
 SHM Structural Health Monitoring
TPC Two-Points Condensation
Technique
LIST OF SYMBOLS
α, β Damping coefficients
θi The rotational degrees of freedom at
node i
vi The degrees of freedom of vertical
displacements at node i
ξn The damping ratio for the nth mode
ξi Damping ration of mode ith
ξj: Damping ration of mode jth
Φ The complete eigenvector matrix of
the full model
48 Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
Φ+mp The generalised inverse of matrix Φmp ∆k ji The difference in the coefficients of
ωn The angular frequency the Kc matrix and the Kr matrix at
C Damping matrix position (j, i)
Cr Damping matrix of the reduced system ∆k jj The difference in the coefficients of
K Globel stiffness matrix the Kc matrix and the Kr matrix at
Ko Theoretical stiffness matrix position (j, j)
Kc Identified stiffness matrix M Globel mass matrix
Kr Undamaged reduced stiffness matrix Mr Mass matrix of the reduced system
∆Ki–j The change matrix of the set of nodes q(t) The modal coordinate vector
i and j q̌p(t) Approximate solution of δp(t)
∆kii The difference in the coefficients of u Displacement vector
the Kc matrix and the Kr matrix at u̇ Velocity vector
position (i, i) ü Acceleration vector
∆kij The difference in the coefficients of ü(t) Acceleration response vectors
the Kc matrix and the Kr matrix at u(t) Displacement response vectors
position (i, j) T The coordinate transformation matrix
The application of continuous TECHNICAL PAPER
surface wave testing for Journal of the South African
Institution of Civil Engineering
settlement analysis with ISSN 1021-2019
Vol 59 No 2, June 2017, Pages 49–58, Paper 1245

reference to a full-scale PROF GERHARD HEYMANN (Pr Eng, MSAICE) is a

load test for a bridge at professor in the Department of Civil Engineering

at the University of Pretoria. He holds BEng, BEng
(Hons) and MEng degrees from the University of

Pont Melin, Wales, UK Pretoria, and a PhD from the University of Surrey.
He has been involved with teaching and research
in geotechnical engineering for many years. His
field of interest includes the characterisation of
soil behaviour and its application in geotechnical engineering.
G Heymann, J Rigby-Jones, C A Milne
Contact details:
Department of Civil Engineering
University of Pretoria
The presence of a significant thickness of soft silts beneath the site of a proposed new Pretoria 0002
road bridge caused concerns over the suitability of shallow foundations. Despite extensive South Africa
conventional ground investigations, it was not possible to obtain reliable stiffness data, and so a T: +27 12 420 3627
E: gerhard.heymann@up.ac.za
full-scale, fully instrumented load test was undertaken to better determine ground stiffness. The
data obtained was used in conjunction with sophisticated numerical analysis to demonstrate that
JOHN RIGBY-JONES (CEng, EurIng, MICE) holds
the settlement of shallow foundations would be acceptable. Following completion of the bridge BEng and MSc degrees, and has more than 24
construction, Continuous Surface Wave testing was undertaken and a range of simple non-linear years of geotechnical engineering, geophysical
elastic settlement analyses conducted to model the behaviour of the load test. Through these and research experience gained from
analyses this paper demonstrates the benefit of using continuous surface wave ground stiffness universities, large consultancies, and lately RJM
Ground Solutions Ltd, a specialist consultancy
testing in conjunction with simple analysis methods. It underlines the importance of obtaining
set up with co-author Chris Milne. At RJM, John
high-quality stiffness data to accurately predict settlement of shallow foundations. led the development of a commercial CSW
(continuous surface wave) system, which was recognised in the 2013 Ground
Engineering Awards where RJM was a finalist in the Ground Investigation
INTRODUCTION the new bridge on spread foundations. Specialist of the Year category. In 2016 John and Chris set up Ground Stiffness
Surveys Ltd in order to focus on the delivery of a CSW testing service.
The existing 18th century masonry arch The thickness of the silt deposits limited
bridge which crosses the river Afon Alwen options for deep foundations (Solera & Contact details:
Ground Stiffness Surveys Ltd
at Pont Melin, 2 km west of Corwen in Milne 2009). Accurate settlement prediction
35 Church Hill
North Wales, was replaced in 2008 with a was required given the tight tolerance Arnside
new 45° skew bridge with 60 m single span of the bridge bearings to foundation Cumbria
between bearings (Figure 1). The presence of movements. Consequently, extensive ground LA5 0DJ
a 50 m thick sequence of saturated, soft silts investigations and a large-scale load test United Kingdom
T: +44 7879 457210
of glacio-lacustrine origin (the Rûg Silts) were undertaken at considerable cost to
E: john@groundstiffnesssurveys.com
caused concern that excessive settlement provide data for settlement analysis of
would be generated by construction of the foundations. CHRIS MILNE (CEng, FICE, EurIng, CGeol, FGS,
EurGeol) holds BSc (Hons) and MSc degrees and
a Postgraduate Diploma in Geotechnical
Engineering Design & Management. He is
currently a director of Ground Stiffness Surveys, a
specialist CSW ground stiffness profiling
developer and contractor established with fellow
director John Rigby-Jones. Chris has over 25 years
of geotechnical engineering, geophysical and commercial experience,
including for large consultancies working at senior level on major infrastructure
projects. His current role focuses on promoting the commercial application of
CSW testing in UK practice.
Contact details:
Ground Stiffness Surveys Ltd
35 Church Hill
Arnside
Cumbria
LA5 0DJ
United Kingdom
T: +44 7921 492853
E: chris@groundstiffnesssurveys.com

Keywords: continuous surface wave test, CSW, soil stiffness, non‑linear analysis,
Figure 1 S ite location foundation settlement

49
Heymann G, Rigby-Jones J, Milne CA. The application of continuous surface wave testing for settlement analysis with reference to a full-scale load test for a
bridge at Pont Melin, Wales, UK. J. S. Afr. Inst. Civ. Eng. 2017:59(2), Art. #1245, 10 pages. http://dx.doi.org/10.17159/2309-8775/2017/v59n2a6
 Accurate measurement of stiffness in
SPT N-value (blows/300 mm)
the poorly understood and sensitive Rûg
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75
Silts was challenging, with a wide range 0
of values and significant scatter identified
Alluvium
from data from different sources. Solera and 5
Milne (2009) give details of the settlement Rûg Silts
analysis, which adopted a conservative 10
approach and resulted in predictions of
settlement in the range of 75–85 mm.

Depth (m)
15
This level of settlement provided
significant constraints on the construction 20
methodology, with the installation of
bearings and landing of deck beams 25
requiring programme hold periods and
careful monitoring. The actual settlements 30
Excludes refusals
suffered by the bridge foundations were on boulders
observed to be less than half the predicted 35
values. Whilst conservatism was justified SI 1986 SI 2002 SI 2003 BH201 SI 2003 BH201a
in terms of the scale and risk of the project,
the availability of accurate stiffness values Figure 2 S tandard penetration test (SPT) results
would have permitted a less conservative
and simplified analysis approach, and
provided greater confidence in the results Cone resistance (MPa)
and potentially removed the constraints 0 4 8 12 16 20 24
0
on construction associated with the
high-predicted settlements.
Seismic testing techniques have been Alluvium
used successfully for the design of many
projects around the world (Stokoe et al
2004), but its wider adoption has been
limited to date by the familiarity of design-
ers with more conventional, often intrusive
Rûg Silts
and empirically based methods. Continuous 10
Surface Wave (CSW) testing is a rapid and
cost-effective technique for measuring
seismic surface wave velocities at a range
of frequencies allowing an in situ ground
stiffness profile to be determined. Unlike
the most frequently used techniques for
assessing ground stiffness, CSW is non-
intrusive and measures the undisturbed
Depth (m)

stiffness of the ground at the current in 20

situ stress conditions. Furthermore, the
technique measures stiffness at small-strain
levels (Go), which can then be adjusted
for the desired design strains using well
understood and simple to apply stiffness
reduction models which are remarkably
independent of soil type. The intimate fun-
damental relationship between shear wave
velocity and soil stiffness, combined with 30

the CSW technique’s direct measurement of Becomes

more sandy
in situ soil properties, makes the technique
a highly attractive method for the meas-
urement of ground stiffness profiles. The
extensive data set available for Pont Melin
Rûg, including a wide range of intrusive
testing and a fully instrumented load test,
afforded an excellent opportunity to evalu-
40
ate the accuracy of settlement prediction
using the CSW technique on this problem-
atic ground not well suited to other types of
stiffness measurement. Figure 3 C
 one penetration test (CPT) results

50 Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017

Figure 4 F ull-scale load test
SITE INVESTIGATION
The Rûg Silts deposits were formed by sea-
sonal deposition of glacially derived rock flour
within a glacial lake (Nichol and Reynolds
2002). The behaviour of the Rûg Silts was
fundamental to foundation design at the site
and there were concerns that the properties
of these deposits were not fully understood.
Low SPT N-values recorded during an early
investigation, ranging from less than 1 to 4,
caused particular concern that the deposits
may be inhomogeneous, with the potential for
large or differential settlements. Furthermore,
experience during an early investigation of
casing loss whilst drilling, together with the
very low blow counts from SPT testing, sug-
gested that the deposits could be sensitive to
disturbance and hence difficult to evaluate
through intrusive testing.
Figure 2 shows the SPT N-values measured
at the site. The data set clearly shows the
interface of the alluvial gravel and the Rûg
Silts, with significantly higher SPT N-values
recorded for the gravel than the Rûg Silts.
Very large scatter of the SPT N-values may
be seen, particularly for the alluvial gravels.
SPT testing was done for each of the three site
investigations conducted for the bridge foun-
dations. The very low blow counts measured
during the first investigation in 1986 were sub-
sequently discounted, as it was concluded that
the material was disturbed due to inadequate
water balance during drilling. During the later
site investigations in 2002 and 2003, full-time
supervision was conducted on site to ensure
good quality control during the drilling,
and in particular to monitor the water bal-
ance. Figure 2 shows that in general the SPT
N-values were higher for these investigations.
Notwithstanding the higher N-values, the data
was still scattered, indicating the difficulties
in reliably quantifying the in situ properties
of these soils, which are sensitive to distur-
bance, by a process which inherently causes
disturbance. Stroud (1989) recognised the
importance of strain level on soil stiffness and
suggested relationships between the drained
Young’s modulus (E’) for foundations on sand
and energy-corrected SPT N-value (N60) as a
function of the factor of safety. Using this data,
Clayton (1995) argued that most foundations
Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
Figure 5 F ull-scale load test layout
have a factor of safety considerably in excess of
3, and suggested E’ = 2N60 (MPa) for normally
consolidated sands. The N-values of between
6 and 15 measured in the upper part of the
Rûg Silts therefore indicate a drained Young’s
modulus (E’) of between 12 and 30 MPa. The
scatter in the SPT N-values for the alluvial
gravel was too large to make a meaningful
estimation of the Young’s modulus.
Figure 3 shows a typical cone penetration
test (CPT) result. The interface of the alluvial
gravel and Rûg Silts is also clearly evident,
but the data scatter again demonstrates the
difficulties of quantifying suitable stiffness
values for design. For qc less than 10 MPa,
Lunne and Christoffersen (1985) suggested
the relationship between cone resistance
(qc) and the drained Young’s modulus for
normally consolidated sands as E’ = 4qc. For
qc ranging from 2 to 6 MPa this relates to
a drained Young’s modulus of 8 to 24 MPa.
The cone resistances in the alluvial gravel
again exhibited large scatter, making the
estimation of design stiffness difficult.
Laboratory stiffness measurement of the
alluvial gravel and the underlying soft silt
material also proved challenging because of
the difficulty in obtaining high-quality sam-
ples, due to the susceptibility of the material
to disturbance during borehole drilling. The
Rûg Silts were particularly vulnerable to
drilling disturbance, and during the first site
investigation the drillers recorded numerous
cases of “blowing” where the base of the
borehole raised several metres inside the
casing overnight, probably exacerbated by
inadequate water balance during the drilling
process. In another instance the borehole
casing settled 2.7 m under its own weight.
Given the difficulty experienced in
accurately determining stiffness moduli
from the SPT and CPT testing, as well as the
laboratory testing, a full-scale load test was
Excavation
Concrete footing (5 m × 5 m)
Settlement marker
Piezometer
Inclinometer
Extenometers
undertaken to provide more reliable data for
settlement prediction.
FULL-SCALE LOAD TEST
A 5 m × 5 m × 0.5 m thick concrete footing
was constructed 2.5 m below the natural
ground level, and loaded with 442 tonne steel
kentledge, as shown in Figure 4. The load test
was undertaken approximately 8 m north of
the northern edge of the east abutment. The
boreholes close to the footing indicated that
the footing was between 2.1 m and 3.0 m
above the base of the alluvial gravel, with an
average depth of 2.6 m. Numerous instru-
ments were installed to monitor the behaviour
of the soil and foundation during the load test.
Eight magnetic extensometers were installed
in a borehole near the centre of the concrete
footing. Four vibrating wire piezometers
were installed in the Rûg Silts to monitor the
pore pressure regime below the foundation
and to detect possible excess pore pressure
during loading. Two inclinometer tubes were
installed adjacent to the footing to measure
lateral movement of the soil. Precise levelling
with an accuracy of 0.1 mm was conducted
of the footing and adjacent ground surface at
the positions shown in Figure 5. This included
eight ground settlement markers adjacent
to the excavated area and eight settlement
markers at the edge of the concrete footing.
The settlement markers situated around the
periphery of the test slab were domed-head
Feno-type steel pins set in concrete, and the
markers fixed to the concrete test slab sides
were Hilti-type bolts. One benchmark was
used a distance of 48 m from the footing.
Loading was applied in five increments of
48 T, 152 T, 248 T, 344 T and 442 T over a
period of six days, and all the instrumenta-
tion was read before and after each load
increment. The load settlement behaviour
51
of the footing is shown in Figure 6. At 200
the maximum applied load of 442 T, cor-
responding to a contact stress of 176.8 kPa, 180
the average settlement of the levelling points
was 18.3 mm. The maximum settlement 160
observed was 21.8 mm and the maximum
140
differential settlement was 5.7 mm.

Contact stress (kPa)

Settlement took place almost immediately
120
after load application. The vibrating wire
piezometers registered no excess pore 100
pressure, confirming that consolidation
occurred almost immediately, and confirm- 80
ing the relatively free draining nature of
the gravelly alluvium and sandy upper 3 to 60
4 m of the Rûg Silts. For the first three load
increments the load was held constant over- 40

night, and during this time settlement creep

20
of between 0.2 and 0.6 mm occurred. Before
the last load increment was applied, a rest 0
period of three days was observed during 0 5 10 15 20
which creep of 3.3 mm took place. The foot- Average settlement (mm)
ing exhibited typical non-linear settlement Observed average settlement GEO5 – 1 step GEO5 – 5 step Versak 2.0
behaviour, with stiff response during the
first and second load increments, becoming Figure 6 L oad test settlement
less stiff for the subsequent load increments.
Figure 7 shows a cross-section of the
profile and the positions of the top three 0.5 m
magnetic extensometers. The magnetic
extensometer had a resolution of 1 mm,
which was low compared to the maximum 1.8 m
measured vertical settlement at extensom-
eter position S8 of 7 mm. Figure 8 shows the S8
settlement measured by the top three exten- 0.8 m
someters, as well as the settlement of the top
of the extensometer borehole casing that was
measured using precise levelling. Despite
the data scatter due to the low resolution of 2.2 m
the extensometers, settlement of the exten-
someters near the footing was observed as
the load was increased. Figure 8 also shows S7

linear trend lines for each extensometer

magnet. Extensometer S6 at a depth of 7.4 m 2.1 m
below the footing measured very little move-
ment, as did the deeper extensometers (S1
S6
to S5) which ranged in depth between 9.5 m
and 26.5 m. The bottom of the casing, taken
as the datum, was 34.5 m below the footing. Rûg Silts
From the lack of movement of extensometers
S1 to S6 it is clear that the soil deformation
27.2 m
was confined to a depth of approximately
1.5 times the width of the footing.
The stiffness of the soil may be computed
between any two extensometer positions. If
axisymmetrical conditions are assumed, the Datum
Youngs modulus (E) is:

∆σ’v – 2v’ ∆σ’r 5 to 15 m

E’ = (1)
∆εv

Where:
E’ = drained Young’s modulus
∆σ’v = vertical effective stress increment
∆σ’r = radial effective stress increment Figure 7 G
 round profile at load test

52 Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
∆εv = vertical strain increment
v’ = Poisson’s ratio.
The vertical strain in the Rûg Silts between
extensometers S6 and S7 ranged between
0.015% for the contact stress of 19.2 kPa
at the first load increment and 0.115% at
176.8 kPa. Using Equation 1, the Young’s
modulus of the Rûg Silts was calculated as
41 MPa. Unfortunately the non-linear stress
strain behaviour of the soil could not be
quantified due to the low resolution of the
magnetic extensometers.
The full-scale load test was completed
in 2003, and on the basis of the results the
decision was made to build the bridge on
shallow foundations founded in the alluvial
gravel. Subsequent to the construction
of the bridge the opportunity arose to
conduct continuous surface wave testing
immediately adjacent to the load test area,
with the aim of assessing the application of
the CSW test for settlement prediction of
shallow foundations.
CONTINUOUS SURFACE
WAVE TESTING
The Continuous Surface Wave (CSW) test
is a seismic test which uses a vibratory
source placed on the ground surface. It
induces Rayleigh waves which radiate along
the ground surface and are detected by a
linear array of geophones also placed on the
ground surface. The vibrator can be used as a
harmonic source at a particular frequency or
as a transient source which sweeps through
a frequency range of interest. The CSW
test has the advantage over impact source
techniques (SASW and MASW) that the
frequency content of the source is known
and can be controlled. This allows the signal
quality to be assessed in the field and a poor
signal or interference from background noise
is easily detected. Also, a vibratory source
provides a significantly higher signal-to-noise
ratio compared to impact sources (Jiang et
al 2015). Simons et al (2002) found the CSW
test to give more repeatable results than the
SASW test.
The geophone response is recorded for
a range of source frequencies, and at each
frequency the Rayleigh wave phase velocity
is calculated. Due to the dispersive nature
of ground profiles, the phase velocity is
frequency-dependent. High frequency
excitation induces short wavelengths which
penetrate only to shallow depth, whereas low
frequencies generate long waves which pen-
etrate deeper. The response at high frequency
is therefore determined by the shallow soils,
whereas the response at low frequency is
also influenced by deeper material. Ground
Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
12
10
8
Extensometer settlement (mm)
6
4
2
0
–2
–4
0 50
Contact stress (kPa)
Top
Figure 8 E xtensometer settlement
motion is insignificant at depths below one
wavelength.
An important part of the analysis of
CSW results is the inversion of the disper-
sion data. Inversion is the process by which
theoretical Rayleigh wave dispersion curves
of a model ground profile are matched with
the measured dispersion data. Calculation of
the theoretical dispersion curves of a model
ground profile requires the solution of the
Rayleigh eigenvalue problem. Thompson
(1950) and Haskell (1953) used the transfer
matrix method and formulated an algorithm
to model surface waves propagated by a
layered medium known as the Thomson-
Haskell algorithm. Kausel and Roësset (1981)
derived a formulation based on the finite
element dynamic stiffness matrix method,
and Kennett (1974) developed the reflec-
tion and transmission coefficient method.
Numerous other techniques have been used
to model surface waves propagated by a lay-
ered medium, including the finite difference
method (Boore 1972), numerical integration
(Takeuchi & Saito 1972) and the boundary
element method (Manolis & Beskos 1988).
These methods determine the Rayleigh dis-
persion curves for fundamental and higher
modes of Rayleigh wave propagation.
100 150 200
S8 S7 S6
Fundamental mode propagation is
dominant in normally dispersive profiles,
where stiffness increases with depth. For
inversely dispersive profiles, where soft
layers occur below stiff layers, higher
modes also contribute to Rayleigh wave
propagation. Correct mode identification
of the experimental data is important to
ensure matching of the same modes of the
theoretical and experimental data. Leong
and Aung (2013) developed a weighted
average velocity method to calculate the
effective Rayleigh phase velocity which
includes the contribution of more than one
mode when mode superposition occurs. The
effective Rayleigh phase velocity implicitly
accounts for the existence of higher modes,
and therefore eliminates the requirement
for mode identification. This is particularly
useful for inversely dispersive profiles, as
higher modes also contribute strongly to
Rayleigh wave propagation. The weighted
average velocity method ignores any contri-
bution from near field effects.
Search algorithms are used as part of the
inversion analysis to find the model ground
profile for which the dispersion curves
best match the measured data. Numerous
algorithms have been proposed, including
53
neighbourhood algorithms (Wathelet 2008), 280
genetic algorithms (Lomax & Snieder
1994) and simulated annealing algorithms 260
(Sen & Stoffa 1991). The inversion process
is ill-posed, which implies that there may 240
be two or more theoretical profiles which

Phase velocity (m/s)

fit the experimental data equally well. It is 220
therefore helpful to have additional infor-
mation, such as borehole logs to constrain 200
the inversion problem.
180
During 2011 the opportunity arose to
conduct CSW testing close to the position
160
of the full-scale load test. The testing was
conducted using a shaker and an array of five
140
4.5 Hz geophones spaced at equal distances
of 0.75 m giving a spread length of 3.0 m. 120
The output from the geophones was passed 0 10 20 30 40 50 60 70 80 90 100
through a pre-amplifier before being logged Frequency (Hz)
using a field computer and data acquisition
hardware. The shaker was a 70 kg mechani- Figure 9(a) CSW dispersion data for Test 01
cal shaker with counter-­rotating balanced
eccentric weights driven by a three-phase
motor (Heymann 2013). The tests were 280

conducted in the frequency range 92 Hz to

260
7.6 Hz, and the induced wavelengths varied
between 2.0 m and 19.1 m. Figure 9 shows
240
the dispersion data for the three tests. The
Phase velocity (m/s)

tests show that the Rayleigh wave phase

220
velocity decreased at low frequency. This is
due to the Rûg Silts being less stiff than the 200
overlying alluvial gravel. The best fit effective
dispersion curve was determined using the 180
weighted average velocity method (Leong &
Aung 2013) and a Monte Carlo search algo- 160
rithm for the inversion analysis. The typical
production rate to conduct three CSW tests 140
is approximately two to three hours for the
field work and a similar length of time for 120
0 10 20 30 40 50 60 70 80 90 100
data analysis. This high production rate
Frequency (Hz)
makes CSW testing significantly more cost-
effective than other tests, including SPT,
Figure 9(b) CSW dispersion data for Test 02
CPT, downhole seismic testing and full-scale
load testing.
The profiles of small-strain shear stiff- 280
ness for the material below the full-scale
load test footing are shown in Figure 10 for 260
the three CSW tests. The stiffness profiles
clearly show the interface between the 240
gravel and Rûg Silts. At the foundation level
Phase velocity (m/s)

of the full-scale load test the small-strain 220

shear stiffness (Go) of the alluvial gravel
ranged between 74.3 MPa and 121.0 MPa 200

(midrange value 97.7 MPa). At the top of the

180
Rûg Silts the small-strain shear stiffness for
the three profiles ranged between 13.4 MPa
160
and 39.4 MPa (midrange value 26.4 MPa).
140

SETTLEMENT ANALYSIS 120

The settlement of the full-scale load test 0 10 20 30 40 50 60 70 80 90 100
was analysed using the small-strain stiff- Frequency (Hz)
ness data obtained from the CSW tests, and
the calculated settlement compared to the Figure 9(c) C
 SW dispersion data for Test 03

54 Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017

0 20
0 selected softened strain level and actual
1
Depth below load test foundation (m)
2 settlement analysis
3 settlement calculations to be used to model
4
5
6
7
8 strains from all previous steps are summed
Figure 10 S tiffness profiles from CSW tests
observed behaviour. A number of analyses
were undertaken with a range of complex-
ity. However, all calculation methods were
based on the application of a simple linear
elastic soil model which subdivided the
gravel alluvium into three layers of thick-
nesses (0.6 m, 1.0 m and 1.0 m respectively)
and the underlying Rûg Silts into layers of
thicknesses 0.4 m, 0.5 m, 1.0 m, 3.0 m and
4.5 m respectively. The same layered model
was used for all analyses. A rigid bound-
ary was assumed at 12 m depth equal to
2.4 times the foundation width and below
the depth of any significant stress increase.
The small-strain shear modulus (Go),
as measured by the CSW test, was used
to assign a small-strain drained Young’s
modulus (E’o) to each layer. E’o = 2(1 + v’)‌Go
where the Poisson’s ratio (v’) was taken
as 0.26, which is a typical value for
drained loading of intermediate soils.
Axisymmetrical conditions were assumed
in all analyses, and Boussinesq’s theory
used to calculate the vertical and horizontal
stress increment for each layer below the
centre of the footing.
In all analyses using small-strain
stiffness values, the non-linear stiffness
response of the soil to straining was
accounted for. Many stiffness reduction
curves have been proposed to describe
the non-linear stress strain relationship
observed for soils, including Vucetic and
Dobry (1991), Rollins et al (1998), Clayton
and Heymann (2001), and Archer and
Heymann (2015). For all analyses repre-
sentative stiffness values were calculated by
applying the stiffness reduction function
suggested by Rollins et al (1998, Equation 2
below) to the midrange values determined
from the CSW tests.
Go (MPa)
40 60 80 100 120
Test 01 Test 02 Test 03
E’ 1 An analysis was undertaken using a total
= (2)
E’o 1 + 16γ(1.2 + 10 –20γ)
Where: γ is the current shear strain of a
sub-layer.
Single-stage loading of
manually softened soil
In the simplest analysis method the non-
linear response of the soil was modelled
through a wholesale reduction of all stiff-
ness values to an assumed operational
strain level. Mair (1993) suggested that
strain levels beneath typical shallow foun-
dations are of the order of 0.1% and this
value was used as the basis for reducing the
small-strain stiffness values. Application
of the Rollins et al (1998) stiffness reduc-
tion function (Equation 2) to the midrange
CSW small-strain stiffness values for the
alluvium and Rûg Silts resulted in drained
Young’s modulus values of 83.9 MPa and
22.7 MPa respectively. The settlement of
the load test foundation was then calculated
using these softened stiffness values at each
load stage with the proprietary software
GEO5 Spread Footing which facilitates the
automated linear elastic analysis of a shal-
low foundation on a layer model.
The results of the GEO5 analysis are
shown in Figure 6 and demonstrate a linear
load settlement response corresponding
to the softened soil stiffness values. It can
be seen that at intermediate load stages,
settlements are over-estimated due to the
use of overly softened stiffness values corre-
sponding to a strain in excess of true values.
The error reduces with increased loading
and strains in the ground until, at the final
load stage, the calculated settlement of
18.9 mm is close to the observed settlement
Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
of 18.3 mm. However, it must be noted that
there is a degree of good fortune that the
140
strain level correspond at this loading stage.
Stepwise non-linear
This method allows simple linear elastic
the non-linear behaviour of soil without
the need to assume an operational strain
level. The maximum stress σ v is applied
incrementally in a number of steps, with
the layer stiffnesses in each step being
progressively softened from an initial value
of E 0 to correspond to the calculated strain
level from the previous step. The calculated
to determine the applicable strain for a
particular step. The methodology for the
stepwise non-linear settlement analysis is
shown in Figure 11.
of five steps and the GEO5 Spread Footing
software. The progressive stiffness reduc-
tion of the layer stiffnesses was undertaken
using the calculated vertical strain from
the previous step and not the shear strain.
The results of the 5-step non-linear GEO5
analysis are shown in Figure 6 and demon-
strate a clear non-linear response represent-
ing the progressively softened stiffness
values used in the analysis. It can be seen
that the analysis overestimates the settle-
ments at low stress levels, although there
is a significant improvement in accuracy
over the single-step analysis. Overall there
is a much improved agreement between the
observed data, with the calculated settle-
ment of 17.6 mm at the final loading stage
being close to the observed settlement of
18.3 mm.
To explore the effects of undertaking
stepwise analysis using a larger number of
steps, software (Versak v2.0) was written
to undertake a 100-step analysis where
the stiffness for each layer is automatically
­softened after each load increment. In
contrast to the GEO5 analysis, stiffness
reduction was undertaken using the shear
strain determined from the calculated verti-
cal strains using ε s = ⅔ε v(1 + v’).
The result of the 100-step non-linear
Versak analysis is also shown in Figure 6.
As would be expected the 100-step analysis
provides a smoother non-linear response
curve than the 5-step analysis. Whilst both
the 5-step and 100-step curves provide very
similar overall load settlement curves, it is
noted that the 100-step analysis provides a
slightly softer response, reflecting the more
frequent stiffness reduction. The calculated
settlement of 18.7 mm at the final loading
55
 have been omitted for clarity. Clearly reliable
stiffness parameters could not be obtained
for the alluvial gravel from the standard
penetration test.
σ v3
The effect of disturbance on the SPT
results from the Rûg Silts were clear from
the significantly higher SPT values that
E2 were obtained when mechanisms were put
σ v2
in place to ensure water balance during all
σv

stages of the drilling process. For the first

E1
σ v1
E0
ε v1 ε v2 ε v3
E0
E1
E2
ε s1
log εs
Figure 11 N
 on-linear settlement analysis
stage is close to the observed settlement of
18.3 mm.
DISCUSSION
Conventional site investigation methods to
quantify the design stiffness for the materials
at the location of the bridge at Pont Melin
(constructed in 2008) achieved only limited
success. This was because the materials
were susceptible to disturbance during the
drilling process, and also because the materi-
als were not well suited to conventional
invasive in situ tests. Clayton et al (1995)
pointed out that uncemented sands, silty
sands and gravels are at particular risk of
borehole disturbance. They argued that if
the casing is extended to the bottom of the
borehole it is virtually impossible to prevent
disturbance, due to boiling of the soil
into the casing when the drilling tools are
εv
ε s2
withdrawn from the hole, unless the hole is
constantly recharged with drilling fluid. Data
reported by Mallard (1983) suggests that SPT
N-values may reduce to ¹∕₅th of the correct
value when unsuitable drilling techniques
are used in sands and gravels. The disturbed
zone that occurs below the bottom of a
borehole depends on the size of the borehole
and may be as much as three times the
borehole diameter (Clayton et al 1995). SPTs
conducted at the bottom of the borehole will
therefore be affected by this disturbance.
The SPT had limited success in the
coarse alluvial gravels with a mean particle
size (D 50) of 13.3 mm. Tokimatsu (1988)
showed that SPT results are affected by
particle size, and when the mean particle size
exceeds about 0.5 mm the N-value increases
rapidly. The large scatter of the SPT N-values
in the gravel is immediately apparent from
Figure 2, and cases where refusal occurred
investigation in 1989, without site supervi-
sion, the N-values for the Rûg Silts were
generally below 5, and for the subsequent
investigations in 2002 and 2001, with site
supervision, the N-values were generally
between 5 and 15, with a trend to increase
with depth. This highlights the importance
of sound drilling and testing techniques in
loose cohesionless soils.
Clayton et al (1995) argued that set-
tlement prediction using CPT results are
significantly more accurate than for SPT
results, because the CPT does not suffer
from borehole disturbance. At Pont Melin,
CPT was successfully conducted in the Rûg
Silts, but achieved limited success in the allu-
vial gravels, probably as a result of the large
grain size relative to the size of the cone tip.
The in situ test methods employed at
Pont Melin to characterise the ground
properties were SPT and CPT penetration
tests. Relationships between these tests and
stiffness are empirical and rely on selection
of an appropriate relationship for the mate-
rial type. The stiffness values derived using
this approach are therefore inevitably inac-
curate, and a suitably conservative approach
to parameter selection for design must
be adopted.
Stiffness determined by laboratory testing
on undisturbed samples was not available
for the new bridge at Pont Melin. It is well
known that relatively modest strains induced
during sampling can significantly alter the
stiffness of soils. In addition, high-quality
undisturbed sampling of the alluvial gravels
were not feasible using conventional sam-
pling techniques, and would have required
specialised methods such as ground freezing
or possibly using a piston sampler in a mud-
filled borehole for the soft Rûg Silts.
From the above discussion it is clear that
conventional in situ testing and laboratory
techniques were of limited use to character-
ise the stiffness of the alluvial gravel and soft
Rûg Silts. For this reason a full-scale load
test was conducted. This was an expensive
and time-consuming method, but was justi-
fied, as the cost-saving for using shallow
foundations as opposed to deep foundations
was considerable.
At the time of the project the continuous
surface wave (CSW) test was not available.

56 Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
It is an attractive test method, because it
is a non-disturbing and non-intrusive test
and because the seismic wave velocity of the
material is directly related to stiffness. It is
therefore well suited to quantify the small-
strain stiffness of the geomaterials encoun-
tered at the location of the bridge at Pont
Melin. However, small-strain stiffness (Go
or Eo) cannot be used directly for settlement
calculation, as the stress-strain behaviour of
soil is highly non-linear and an appropriate
stiffness reduction function is required as
part of the settlement analysis.
Settlement analyses to predict the load
settlement behaviour of the full-scale
load test conducted at Pont Melin were
undertaken using a range of simple analysis
techniques in conjunction with the stiffness
parameters determined from CSW testing.
This is in contrast to the more complex
finite element analysis undertaken as part of
the bridge design.
The first single-stage manually softened
analysis requires the average operational
strain level beneath the foundation to be
assumed, and therefore at strain levels
below the assumed value this method over-
estimates settlements and at higher strains
provides an underestimation of settlement.
The method is, however, rapid and uses the
conventional linear-elastic analysis approach
widely adopted by practising engineers for
routine analyses.
The use of a stepwise analysis method
allows the non-linear response of the soil to
be modelled. The use of proprietary software
requires user intervention between each
step to interrogate the output strains in each
layer, undertake reduction of layer stiffnesses
and adjust the input layer stiffnesses for the
next analysis step. Even whilst automating
this process with a spreadsheet as far as
possible, the process is time-consuming. A
small number of load steps are typically used
to keep analysis time within practical limits;
however, the results obtained from a 5-step
analysis provided a reasonable approxima-
tion to the observed load settlement curve.
The automated stepwise non-linear
analysis undertaken using Versak software
allowed a large number of load steps (100)
to be conducted efficiently with no user
intervention. However, whilst the modelled
load settlement curve provided a smoother
response than the 5-step analysis, the
improvement in accuracy to the observed
data was small.
Engineers have to balance resources,
including money, time and effort, between
site investigation and design. Clearly both
these processes are important, but the above
case history suggests that better results were
achieved by focusing on obtaining accurate
Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
material parameters than increasing the
sophistication of the analysis method.
CONCLUSIONS
Valuable lessons were learnt during the series
of site investigations and field testing that
were conducted before and after the con-
struction of the new bridge at Pont Melin.
Quantifying stiffness parameters from
conventional in situ tests such as the SPT
and CPT, as well as laboratory testing, was
difficult and achieved only limited success.
For the alluvial gravel the large scatter in the
SPT and CPT data made selection of the rep-
resentative stiffness difficult. The Rûg Silts
were particularly sensitive to disturbance
and the SPT data from the first site investiga-
tion was anomalously low, due to borehole
disturbance as a result of poor water balance.
Only during later site investigations, when
full time-site supervision was available and
particular care was taken to ensure good
water balance, were more reliable SPT data
obtained. Even then the uncertainty associ-
ated with the selection of an appropriate
empirical relationship between SPT N-values
and stiffness resulted in the adoption of con-
servative stiffness values for design.
The CSW test was successfully used to
rapidly quantify the stiffness of the alluvial
gravel and Rûg Silts, due to the fact that
it is non-invasive and does not cause any
disturbance during testing. In addition, the
CSW test excites a large volume of material
and therefore quantifies the mass stiffness
of the material and avoids the difficulties
which large gravels and cobbles cause for tra-
ditional invasive tests. Clearly, the CSW test
proved to be a more suitable test to quantify
the stiffness of both the alluvial gravel and
the Rûg Silts.
The use of a stepwise non-linear analysis
method over a single-stage method was
shown to provide a significantly more accu-
rate model of the observed settlement. Little
improvement was gained by increasing the
number of analysis steps from 5 to 100.
The use of CSW stiffness data, coupled
with simple analysis techniques, compared
well with observed settlements, and provided
a far better prediction of settlements than
the complex finite element analysis used dur-
ing the original bridge design in conjunction
with conservative soil stiffness values.
ACKNOWLEDGEMENTS
The authors are grateful for the support of
the Welsh Government, sponsor of the new
bridge, who facilitated CSW testing at the
load test site. The supervision of the load
test and design of the bridge foundations
was undertaken by Hyder Consulting (UK)
Limited (now Arcadis Consulting UK
Limited).
REFERENCES
Archer, A & Heymann, G 2015. Using small‑strain
stiffness to predict the load‑settlement behaviour of
shallow foundations on sand. Journal of the South
African Institution of Civil Engineering, 57(2): 28–35.
DOI: 10.17159/2309-8775/2015/v57n2a4.
Boore, D M 1972. Finite difference methods for seismic
wave propagation in heterogeneous materials. In:
Bolt, B.A (Ed.), Methods of Computational Physics,
Vol. 11, New York: Academic Press, 1–36.
Clayton, C R I & Heymann, G 2001. The stiffness of
geomaterials at very small strains. Geotechnique,
51(3): 245–256.
Clayton, C R I 1995. The standard penetration test
(SPT): Methods and use. CIRIA report R143. London:
CIRIA
Clayton, C R I, Matthews, C & Simons, N E 1995. Site
Investigation, 2nd ed. Oxford: Blackwell Science.
Haskell, N A 1953. The dispersion of surface waves on
multilayered media. Bulletin of the Seismological
Society of America, 43: 17–34.
Heymann, G 2013. Vibratory sources for continuous
surface wave testing. In: Coutinho, R Q & Mayne,
P W (Eds.), Geotechnical and Geophysical Site
Characterization (ISC’4), London: Taylor and
Francis, 1381–1386.
Jiang, P, Tran, K T, Hiltunen, D R & Hudyma, N 2015.
An appraisal of a new generation of surface wave
techniques at a test site in Florida. Paper presented
at the International Foundations Congress and
Equipment Expo 2015 (IFCEE), 17–21 March, San
Antonio, TX.
Kausel, E & Roësset, J M 1981. Stiffness matrices for
layered soils. Bulletin of the Seismological Society of
America, 71(6): 1743–1761.
Kennett, B L N 1974. Reflections, rays, and
reverberations. Bulletin of the Seismological Society
of America, 64(6): 1685–1696.
Leong, E & Aung, A 2013. Global inversion of surface
wave dispersion curves based on improved weighted
average velocity method. Journal of Geotechnical and
Geoenvironmental Engineering, 139(12): 2156–2169.
Lomax, A J & Snieder, R 1994. Finding sets of acceptable
solutions with a genetic algorithm with application to
surface wave group dispersion in Europe. Geophysical
Research Letters, 21(24): 2617–2620.
Lunne, T & Christoffersen, P 1985. Interpretation
of cone penetrometer data for offshore sands.
Norwegian Geotechnical Institute, Publication No.
156, Oslo, Norway, 1–12.
Mair, R J 1993. Developments in geotechnical
engineering research: Application to tunnels
and deep excavation. Unwin Memorial Lecture.
Proceedings of the Institution of Civil Engineers, Civil
Engineering, 93: 27–41.
Mallard, D J 1983. Testing for liquefaction potential.
Proceedings, NATO Workshop on Seismicity and
Seismic Risk in the Offshore North Sea Area,
57
 1–4 June 1982, Utrecht, Netherlands. Dordrecht:
Reidel Publishers, 289–302.
Manolis, G D & Beskos, D E 1988. Boundary Element
Methods in Elastodynamics. London: Unwin Hyman.
Nichol, D & Reynolds, J M 2002. Application of
reflection seismology to foundation investigation at
A5 Pont Melin Rûg, North Wales. Quarterly Journal
of Engineering Geology and Hydrogeology, 35, 247–256.
Rollins, K M, Evans, M D, Diehl, N B & Daily, W D III
1998. Shear modulus and damping relationships
for gravels. ASCE Journal of Geotechnical and
Geoenvironmental Engineering, 124(5): 396–405.
Sen M K & Stoffa P L 1991. Nonlinear one-dimensional
seismic waveform inversion using simulated
annealing. Geophysics, 56, 1624–1638.
Simons, N E, Menzies, B K & Matthews, M C 2002.
A Short Course in Geotechnical Site Investigation.
London: Thomas Telford.
58 Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
Solera, S A & Milne, C A 2009. The influence of soft ground
conditions on new bridge design at A5 Pont Melin Rûg,
Corwen. Urban Geology in Wales: 3. National Museum
of Wales Geological Series No. 26, Cardiff, 23–33.
Stokoe, K H, Joh, S & Woods, R D 2004. Some
contributions of in situ geophysical measurements to
solving geotechnical engineering problems. In: Viana
da Fonseca, V & Mayne, P W (Eds.), Proceedings,
ISC‘2 on Geotechnical and Geophysical Site
Characterization held on 19–22 September 2004,
Porto, Portugal. Rotterdam: Millpress, 97–132.
Stroud, M A 1989. The standard penetration test – its
application and interpretation. Proceedings, ICE
Conference on Penetration Testing in the UK.
London: Thomas Telford, 29–46.
Takeuchi, H & Saito, M 1972. Seismic surface waves. In:
Bolt, B A (Ed.), Methods of Computational Physics,
Vol. 11, New York: Academic Press, 217–294.
Thomson, W T 1950. Transmission of elastic waves
through a stratified solid medium. Journal of Applied
Physics, 21, 89–93.
Tokimatsu, K 1988. Penetration testing for dynamic
problems. In: De Ruiter, J (Ed.), Proceedings, 1st
International Symposium on Penetration Testing
(ISOPT-1), 20–24 March 1988, Orlando, FL, Vol. 1,
Rotterdam: Balkema, 117–136.
Vucetic, M & Dobry, R 1991. The effect of soil plasticity
on cyclic response. ASCE Journal of the Geotechnical
Engineering Division, 117(1), 89–107.
Wathelet, M 2008. An improved neighborhood
algorithm: Parameter conditions and dynamic
scaling. Geophysical Research Letters, 35(9),
DOI:10.1029/2008GL033256, 2008.
Effect of the minimum TECHNICAL NOTE
void ratio on the vertical Journal of the South African
Institution of Civil Engineering
intercept of the steady state ISSN 1021-2019
Vol 59 No 2, June 2017, Pages 59–64, TN-77

line of non-plastic soils DR LUIS ALBERTO TORRES-CRUZ obtained his

BSc in Civil Engineering from the Universidad
del Valle in Cali, Colombia, and completed his
L A Torres-Cruz, S Geyer, P R Mackechnie PhD in Geotechnical Engineering at the
University of the Witwatersrand (Wits). His
doctoral research focused on the use of the
The steady state line (SSL) plays a key role in understanding and modelling the mechanical cone penetration test to assess the liquefaction
potential of tailings dams. He is currently a
response of soils. Accordingly, understanding how the SSL correlates to soil index properties is
lecturer in Geotechnical Engineering at Wits.
of primary importance. A previous study reported that the vertical location of the SSL (Γ1) in void
ratio (e) versus mean effective stress (p’) space is correlated to the minimum void ratio (emin). Contact details:
Department of Civil and Environmental Engineering
However, the correlation only included soils with narrow particle size distributions (PSD) and low
University of the Witwatersrand
fines content (FC). In the current study, published data corresponding to 30 non-plastic soils were Private Bag 3
re-processed to further explore the applicability of the Γ1–emin correlation. The results indicate that Wits
the Γ1–emin correlation is linear (R2 = 0.85) and valid regardless of the coefficient of uniformity (Cu), 2050
FC, and particle shape. The Γ1–emin dataset presented herein was also compared to a previously South Africa
T: +27 11 717 7150
published dataset, and good agreement was observed. It is proposed that the Γ1–emin correlation
E: LuisAlberto.TorresCruz@wits.ac.za
can be very useful to understand how the Γ1 of different non-plastic soils compare to one another,
and to minimise the extent of triaxial testing required when characterising a soil deposit from an
STEPHAN GEYER obtained his BSc in Civil
SSL standpoint. Limitations of the Γ1–emin correlation are also discussed. Engineering from the University of the
Witwatersrand. He is currently employed at
Prime Resources Environmental Consultants,
INTRODUCTION q-p’ projection, which reflects steady state where he is involved in the geotechnical and
design aspects of water and mining residue
Soils reach constant values of void ratio (e), frictional properties, is strongly dependent
management, including tailings storage
mean effective stress (p’ = (σ’1 + σ’2 + σ’3)/3), on particle shape (Cho et al 2006) and largely facilities.
and deviator stress (q = σ’1 – σ’3) when independent of particle size distribution (PSD)
Contact details:
sheared to large strains (Castro 1969). The (Carrera et al 2011; Rahman et al 2014). By
Prime Resources Environmental Consultants
locus of steady state (e, p’, q) coordinates contrast, the e-p’ projection, representative PO Box 2316
attained when shearing from different initial of stiffness, is affected by particle shape (Cho Parklands
states, defines the steady state line (SSL). et al 2006), PSD (Thevanayagam et al 2002; 2121
South Africa
Because the SSL represents the stress and void Rahman & Lo 2008; Muir-Wood & Maeda
T: +27 11 447 4888
ratio conditions towards which a soil evolves 2008; Li et al 2013), and void ratio limits (Cho E: stephan@resources.co.za
when sheared, it plays a key role in defining et al 2006; Cubrinovski & Ishihara 2000;
mechanical response. Two projections are Hemer et al 2016). Given the greater number PETER MACKECHNIE obtained his BSc degree in
typically used to define the SSL: q-p’ (stress of factors that affect the e-p’ projection, it has Civil Engineering from the University of the
plane) and e-p’ (compression plane). The received attention from a significant number Witwatersrand.

1.0

Γ1

Contact details:
0.9 Department of Civil and Environmental Engineering
University of the Witwatersrand
1 Private Bag 3
Void ratio (-)

Wits
0.8 λ10
2050
South Africa
T: +27 11 326 0030
E: mackechniep@gmail.com
0.7

0.6
1 10 100 1 000
Mean effective stress (kPa)
Keywords: steady state line, critical state line, minimum void ratio,
Figure 1 Idealisation of the SSL using Equation 1 non‑plastic soils

59
Torres-Cruz LA, Geyer S, Mackechnie PR. Effect of the minimum void ratio on the vertical intercept of the steady state line of non-plastic soils.
J. S. Afr. Inst. Civ. Eng. 2017:59(2), Art. #TN-77, 6 pages. http://dx.doi.org/10.17159/2309-8775/2017/v59n2a7
of researchers and is the focus of this note. 100
The e-p’ projection is commonly modelled
90
with Equation 1:
80
e = Γ1 – λ10log10(p’) (1)

Percentage finer by weight

70

Where λ10 is the slope of the SSL in semi- 60

logarithmic space and Γ1 is the void ratio at 50
p’ = 1 kPa (Figure 1). The current note will
40
explore the correlation between the minimum
void ratio (emin), which is associated with a 30
defined maximum density state, and Γ1.
20
Several studies have investigated how
the SSL is affected by soil index properties. 10
Thevanayagam et al (2002) tested gap- 0
graded mixtures composed of sand and non- 0.001 0.01 0.1 1 10
plastic fines and concluded that the vertical Particle size (mm)
location of the SSLs could be explained by Thevanayagam et al (2002) Yang et al (2006) Li (2003)
the fines content (FC). They noted that as Note: This study used 14 soil types from Li (2013); however, only eight PSD curves are shown because
FC increases from zero, the SSL shifts down- some soil types had identical PSD curves but different particle shape
wards in e-p’ space (Γ1 decreases), and that
beyond a certain FC value it shifts upwards Figure 2 PSD curves of the 30 soils whose data was processed by the current authors
(Γ1 increases). The FC value at which the
SSL shift reverses direction was termed by
1.6
Thevanayagam et al (2002) as the threshold
emin method: ASTM D1557
FC (TFC). The calculation of the parameters
1.4 Γ1 = 1.37emin + 0.04
proposed by Thevanayagam et al (2002) to
R 2 = 0.91
explain the effect of FC required knowledge
1.2
of the SSLs of the different sand-fines mix-
tures. Consequently, the framework lacked
1.0 100/10
predictive power. To overcome this, Rahman
Γ1

& Lo (2008) developed semi-empirical equa- 7/2.1

tions to calculate, as a function of FC and
other PSD descriptors, the parameters in the
framework proposed by Thevanayagam et al
(2002). This allowed the prediction of the
SSL of sands with FC < TFC, provided that
the SSL of another sand-fines mixture with
FC < TFC was known. More recent works
show that the effect of non-plastic fines on
the SSL continues to be investigated (e.g.
Mohammadi & Qadimi 2015; Rahman et al
2014; Yang et al 2015).
Despite the success of Rahman & Lo
(2008) in predicting several SSLs, their frame-
work has limitations that hinder its wider
applicability. For example, the framework is
limited to soils with FC smaller than the TFC,
which tends to be close to 40%. Additionally,
the framework cannot explain the differences
between the SSLs of soils with no fines but
different PSDs or grain shape.
Cho et al (2006) explored the correlation
between Γ1 and emin of 49 natural and crushed
sands with mostly little to no fines (only
six sands had FC > 12%) and a maximum
coefficient of uniformity (Cu) of 6.2 (only
three sands had Cu > 4). They found that the
correlation was linear, independent of particle
shape, and of modest strength (R2 = 0.54).
However, the validity of the Γ1-emin cor-
relation for high FC values or widely graded
soils remains untested. The objective of this
0.8
0.6
0.4
0.2
0.1
Notes: Label next to each data point indicates FC(%)/Cu
A TFC of 36% has been adopted based on the estimation of Rahman & Lo (2008)
Figure 3 Γ1 vs emin plot corresponding to soil types reported in Thevanayagam et al (2002)
note is to present evidence that expands the
applicability of the Γ1-emin correlation to a
wide range of non-plastic soils, regardless of
PSD descriptors such as FC and Cu. The wide
applicability of this correlation is considered
a step towards overcoming the limitations
encountered when using FC and TFC to
explain the location of the SSL.
METHODOLOGY
Datasets from three references were pro-
cessed to explore the validity of the Γ1-emin
correlation: Thevanayagam et al (2002), Yang
et al (2006) and Li (2013). The dataset from
Thevanayagam et al (2002) includes seven
soils composed of foundry sand mixed with
15/14 0/1.7
25/28
60/26
40/47
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
emin
FC < TFC FC > TFC
non-plastic crushed silica fines. The resulting
mixtures have FCs varying from 0% to 100%
and Cu varying from 1.7 to 47. The dataset
from Yang et al (2006) includes nine soils
composed of Hokksund sand mixed with
non-plastic Chengbei silt. The resulting mix-
tures have FCs varying from 0% to 94% and
Cu varying from 2 to 14. The Thevanayagam
et al (2002) and Yang et al (2006) datasets
were used herein to assess the validity of the
Γ1-emin correlation over a wide range of FC
and Cu values. The dataset from Li (2013)
includes 14 soils of which six were made of
glass balls (spherical particles) and eight of
Hostun sand (angular particles). Two soils,
one each of glass balls and Hostun sand,
had FC =10%, whereas the remaining 12 had

60 Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
Table 1 Values of p’ and e used to calculate Γ1
Soil type p’ (kPa) e Soil type p’ (kPa) e Soil type p’ (kPa) e Soil type p’ (kPa) e
5 0.801 13 0.854 277.9 0.577 83 0.534
10 0.796 SIM (100/10) 157 0.821 GB (0/2.5) 543.2 0.566 115 0.560
23 0.774 315 0.767 840.3 0.559 HCM (20/13) 221 0.543
41 0.774 25.9 0.779 131.6 0.470 268 0.556
48 0.765 162.5 0.732 268.6 0.460 515 0.482
HSd (0/1.1/0.9)e GB (0/5)
157 0.746 328.3 0.719 360.4 0.462 0.1 0.626
SIMa (0/1.7)b
473 0.718 639.0 0.700 546.0 0.448 0.1 0.612
937 c 0.668 158.9 0.730 136.6 0.421 0.5 0.603
937 0.645 HS (0/1.1/0.35)e 315.3 0.706 144.8 0.413 0.7 0.588
GB (0/10)
1 099 0.681 632. 6 0.697 271.1 0.404 1.0 0.532
HCM (30/14)
1 217 0.595 11.5 0.780 572.6 0.392 22 0.477
1 407 0.606 158.4 0.726 75.1 0.342 108 0.544
5 0.731 159.2 0.727 131.0 0.332 209 0.536
GB (10/20)
14 0.717 160.6 0.723 274.6 0.321 272 0.530
216 0.670 224.3 0.720 579.1 0.304 528 0.448
HS (0/1.4/0.9)e
SIM (7/2.1) 236 0.665 249.5 0.720 14 0.854 0.1 0.758
434 0.628 316.4 0.713 29 0.850 0.4 0.727
547 0.595 497.5 0.700 77 0.842 1.2 0.746
1 234 0.553 544.1 0.690 HCMg (0/2.4) 111 0.828 1.8 0.667
1 0.617 629.9 0.691 215 0.825 3 0.635
HCM (50/8.9)
17 0.589 165.3 0.723 268 0.807 109 0.683
42 0.575 HS (0/1.4/0.75)e 318.1 0.710 335 0.809 139 0.567
94 0.600 627.1 0.698 9 0.789 165 0.523
SIM (15/14)
124 0.575 9.6 0.774 54 0.771 208 0.673
306 0.522 12.8 0.770 98 0.781 262 0.673
306 0.513 61.3 0.743 HCM (5/3.4) 121 0.779 1.0 0.981
HS (0/2.5)
1 289 0.421 159.7 0.714 236 0.760 1.3 1.006
1 0.477 321.3 0.695 275 0.756 2.2 0.958
11 0.461 640.5 0.675 306 0.752 3 0.815
69 0.430 162.1 0.694 51 0.715 4 0.790
HCM (70/2.2)
73 0.416 323.4 0.671 64 0.693 12 0.748
SIM (25/28) HS (0/5)
83 0.430 326.3 0.681 111 0.675 105 0.900
232 0.404 644.1 0.662 HCM (10/6.6) 115 0.709 126 0.723
690 0.357 164.7 0.618 221 0.650 194 0.877
787 0.357 HS (0/10) 324.7 0.608 286 0.665 253 0.885
4 0.410 652.0 0.606 495 0.679 0.7 1.242
7 0.401 164.5 0.602 7 0.678 2.3 1.267
SIM (40/47) 13 0.390 HS (10/20) 325.1 0.595 7 0.670 2.3 1.219
64 0.390 645.2 0.586 61 0.658 4 0.983
265 0.363 138.1 0.671 70 0.636 5 1.094
HCM (15/11) HCM (94/2)
5 0.547 275.6 0.663 109 0.633 6 1.046
GBf (0/1.1)
5 0.530 556.5 0.655 226 0.607 97 1.150
11 0.539 1 363.7 0.645 283 0.615 152 0.929
SIM (60/26) 112 0.514 119.9 0.643 314 0.584 184 1.125
138 0.514 252.6 0.633 1.4 0.650 240 1.113
GB (0/1.4)
157 0.511 498.3 0.623 2.5 0.638
HCM (20/13)
184 0.497 1 254.8 0.621 7 0.624
SIM (100/10) 10 0.877 GB (0/2.5) 138.2 0.586 32 0.579
a. SIM = Silica sand-silt mixtures tested by Thevanayagam et al (2002)
b. Values in parentheses indicate FC(%)/Cu
c. Values of p’ and e that appear in bold-italic font were not used to calculate Γ1 as they could not be adequately fitted with Equation 1
d. HS = Hostun sand tested by Li (2013)
e. Third value inside the parentheses indicates the mean grain size in mm
f. GB = Glass balls tested by Li (2013)
g. HCM = Mixtures of Hokksund sand and Chengbei silt tested by Yang et al (2006)

Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017 61
FC = 0%. Cu varied from 1.1 to 20. This data- 1.6
set was used herein because: (i) some of the emin method not reported
soils made up of glass balls had considerably 1.4 Γ1 = 1.52emin + 0.02
low emin values which allowed a significant R2 = 0.95
extension of the lower bound of the domain 94/2
1.2
of the Γ1-emin correlation; (ii) given that the
particle shape of the glass balls is distinctly 1.0 0/2.4
different from that of Hostun sand, this

Γ1
dataset enables a straightforward assessment 70/2.2
0.8 15/11
of whether particle shape affects the Γ1-emin 20/13 5/3.4
correlation; and (iii) this dataset also allows 50/8.9 10/6.6
0.6
assessment of the validity of the Γ1-emin 30/14
correlation at different Cu values. Figure 2 0.4
presents the PSDs of the 30 soils considered.
Values of Γ1 were calculated by fitting 0.2
Equation 1 to the (p’, e) points that defined 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
the SSL of each soil (Table 1). Some of the emin
SSLs reported by Thevanayagam et al (2002) Note: Label next to each data point indicates FC(%)/Cu
cannot be modelled with Equation 1, due
to the curvature of the SSL in e-log10p’ Figure 4 Γ1 vs emin plot corresponding to soil types reported in Yang et al (2006)
which some soils exhibit at high stress
levels (e.g. Been et al 1991; Li & Wang 1998).
1.6
Consequently, some (p’, e) points with high
emin method: ASTM D4253
p’ values were excluded from the fitting
1.4 Γ1 = 0.93emin + 0.31
process. Similarly, given the experimental
R 2 = 0.90
difficulties and uncertainties involved in
1.2
performing triaxial tests at very low values
of effective stress, two (p’, e) points with
1.0 0/1.1 0/1.1
p’ = 1 kPa were also disregarded when calcu- 10/20
0/2.5
Γ1

lating Γ1 (see footnote c in Table 1). 0/1.4

0.8 0/5
As annotated in Figures 3 to 5, different 0/1.1
methods were used to determine the emin 0/2.5
0.6 0/10 0/1.4 0/1.4
0/10
values of each dataset. ASTM D1557 refers
to the modified Proctor compaction test, 0/5
0.4
and ASTM D4253 refers to the method of 10/20
soil densification using a vibratory table. 0.2
Although the current authors acknowledge 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
that emin values from different methods are emin
not strictly comparable, it is hypothesised Glass balls Hostun sand
that, regardless of the method, the resulting Note: Label next to each data point indicates FC(%)/Cu
emin provides a reasonable indicator of pack-
ing efficiency. Figure 5 Γ1 vs emin plot corresponding to soil types reported in Li (2013)

RESULTS AND DISCUSSION
Figures 3 to 5 suggest strong (R2 ≥ 0.90)
linear Γ1-emin correlations. The data point
labels further suggest that the correla-
tions are valid regardless of FC or Cu. The
independence of the correlation from FC
observed in Figures 3 and 4 is at odds with
previous works (e.g. Thevanayagam et al
2002; Rahman & Lo 2008; Rahman et al
2014) which have suggested that Γ1 is fun-
damentally correlated to FC. Furthermore,
Figure 3 explicitly shows that essentially the
same Γ1-emin correlation is followed regard-
less of whether FC is smaller or greater than
TFC. Additionally, the angular Hostun sand
and the glass balls follow the same Γ1-emin
correlation despite their significantly dif-
ferent particle shapes (Figure 5). This result
agrees with Cho et al (2006) who reported
the independence from particle shape of the
Γ1-emin correlation for narrowly graded sands
with low values of FC.
When all 30 soils are collectively plot-
ted (Figure 6), a single linear correlation
emerges (R2 = 0.85). The slight decrease
in R2 (compare to Figures 3 to 5) is likely a
consequence of combining SSLs calculated
from triaxial tests conducted in different
laboratories and following slightly different
protocols, and emin values obtained through
different procedures. For example, Prochaska
& Drnevich (2005) showed that the maxi-
mum dry unit weight, which is associated to
emin, can show variances of the order of ± 3%
when estimated from different compaction
techniques. A unique linear correlation
(R2 = 0.77) continues to be apparent when
the data corresponding to Cho et al (2006)
is included (Figure 7). This indicates that
the Γ1-emin correlation observed by Cho et
al (2006) in narrowly graded sands with low
FC values is approximately the same for soils
with significant amounts of non-plastic fines
and large Cu values, such as those repre-
sented in Figures 3 to 5. The data analysed
in this study has also been useful to expand
the lower bound of the domain of the Γ1-emin
correlation reported by Cho et al (2006)
(Figure 7).
The authors suggest that the validity
of the Γ1-emin correlation over such a wide
range of non-plastic soil types is explained by
the similarity in which both Γ1 and emin are
affected by a soil’s fundamental properties.
For example, they are both directly corre-
lated to particle angularity (Li 2013; Biarez &
Hicher 1994; Cho et al 2006), inversely cor-
related to Cu (Li 2013; Biarez & Hicher 1994;
Poulos et al 1985), and respond in a similar

62 Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
 1.6 and valid regardless of FC, Cu, and particle
Γ1 = 1.14emin + 0.20 shape. Comparison of the results presented
1.4 R 2 = 0.85 herein (Figure 6) with the Γ1 versus emin
dataset obtained by Cho et al (2006), indi-
1.2 cates that the Γ1-emin correlation originally
observed by Cho et al (2006) for narrowly
1.0 graded sands with small amounts of fines
may be applicable to all non-plastic soils
Γ1

0.8 (Figure 7). The similarity in the way in which

both Γ1 and emin respond to changes in the
0.6 fundamental properties of a soil is believed
to be the reason why the Γ1-emin correlation
0.4 is valid over a wide variety of soil types.
Given that the effect of non-plastic
0.2
fines on the SSL continues to be intensely
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 researched, it is an important finding of
emin this work that the Γ1-emin correlation is not
Thevanayagam et al (2002) Li (2013) – Glass balls affected by FC (Figures 3 and 4). This is
Yang et al (2006) Li (2013) – Hostun sand not entirely surprising though, as it seems
unlikely that a fundamental soil property (Γ1)
Figure 6 Γ1 vs emin plot including all the data points from Figures 3 to 5 can be explained by a property such as FC
which is arbitrarily defined as the percentage
of particles smaller than 75 μm (or 63 μm,
1.6
depending on the standard). Accordingly, it
Γ1 = 1.34emin + 0.16
must be expected that the predictive power
1.4 R2 = 0.77
of FC and associated concepts like the
TFC will have important limitations. The
1.2
results presented herein suggest that a better
Expanded domain understanding of the SSL can be achieved by
1.0
correlating Γ1 to emin rather than to FC. The
Γ1

authors are currently analysing an extended

0.8
database and conducting triaxial experi-
ments to continue exploring the strengths
0.6
and limitations of the Γ1-emin correlation.
0.4

0.2
ACKNOWLEDGEMENTS
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 The authors gratefully acknowledge Dr Nico
emin Vermeulen (Jones & Wagener) and Dr Irvin
30 soils processed by the current authors 49 soils processed by Cho et al (2006) Luker (University of the Witwatersrand) for
Note: The emin values reported by Cho et al (2006) were obtained using a variety of methods their valuable insights and discussions.

Figure 7 Γ 1 vs emin plot including all the data points from Figures 3 to 5 and the data points from
Cho et al (2006)
manner to changes in FC (Lade et al 1998;
Rahman & Lo 2008). It is also important to
acknowledge that the use of emin as a predic-
tor of Γ1 may be limited by the differences
between the remoulding and particle crush-
ing mechanisms that a soil undergoes when
sheared to steady state and when compacted
to emin.
The authors are not recommending the
use of the correlations in Figures 3 to 7 to
replace triaxial testing to determine the
SSL, as doing so can result in significant
errors. For example, in Figure 7 a deviation
from the best fit line of ± 0.1 is observed in
Γ1, implying a potential error of 0.2. This
value is significantly higher than the error
of ± 0.01 in void ratio, which was suggested
by Jefferies & Been (2006) as a reasonable
target error when calculating the SSL experi-
mentally. Notwithstanding these potential
errors, the correlation is very useful to
qualitatively understand how the Γ1 values of
different non-plastic soil types will compare
to one another. Additionally, when studying
a heterogeneous soil deposit that includes
a variety of soil types, the correlation can
help reduce the number of soil types whose
SSLs have to be experimentally determined
to fully characterise the deposit from an SSL
standpoint (e.g. Hemer et al 2016).
CONCLUSIONS
The Γ1-emin correlation of 30 non-plastic
soils has been investigated. The results indi-
cate that the correlation is linear (R2 = 0.85)
REFERENCES
Been, K, Jefferies, M G & Hachey, J E 1991. The critical
state of sands. Geotechnique, 41(3): 365–381.
Biarez, J & Hicher, P-Y 1994. Elementary Mechanics
of Soil Behaviour: Saturated Remoulded Soils.
Rotterdam: Balkema.
Carrera, A, Coop, M & Lancellota, R 2011. Influence
of grading on the mechanical behaviour of Stava
tailings. Geotechnique, 61(11): 935–946.
Castro, G 1969. Liquefaction of sands. PhD thesis,
Cambridge, MA: Harvard University.
Cho, G, Dodds, J & Santamarina, J 2006. Particle shape
effects on packing density, stiffness, and strength:
Natural and crushed sands. Geotechnical and
Geoenvironmental Engineering, 132(5): 591–602.
Cubrinovski, M & Ishihara, K 2000. Flow potential of
sandy soils with different grain compositions. Soils
and Foundations, 40(4): 103–119.
Hemer, J C, Mincione, N L M & Torres-Cruz, L A
2016. Determination of steady state lines for

Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017 63
 non-plastic platinum tailings. In: Jacobsz, S W. (Ed.),
Proceedings, 1st Southern African Geotechnical
Conference Held at Sun City, 4–6 May 2016.
London: CRC Press/Taylor & Francis, 275–280.
Jefferies, M G & Been, K 2006. Soil liquefaction: A
critical state approach. London: Taylor & Francis.
Lade, P V, Liggio, C & Yamamuro, J A 1998. Effects of
non-plastic fines on minimum and maximum void
ratios of sand. Geotechnical Testing Journal, 21(4):
336–347.
Li, G 2013. Étude de l’influence de l’étalement
granulometrique sur le comportement mecanique
des materiaux granulaires. PhD thesis, Nantes,
France: Ecole Central de Nantes [In French].
Li, G, Ovalle, C, Dano, C & Hicher, P-Y 2013. Influence
of grain size distribution on critical state of granular
materials. In: Yang, Q, Zhang, J-M, Zheng, H & Yao,
Y (Eds.), Constitutive Modeling of Geomaterials,
Berlin, Heidelberg: Springer Verlag.
Li, X S & Wang, Y 1998. Linear representation of
steady state-line for sand. Geotechnical and
Geoenvironmental Engineering, 124(12): 1215–1217.
Mohammadi, A & Qadimi, A 2015. A simple critical
state approach to predicting the cyclic and
64 Journal of the South African Institution of Civil Engineering • Volume 59 Number 2 June 2017
monotonic response of sands with different fines Yang, S L, Sandven, R & Grande, L 2006. Steady-state
contents using the equivalent intergranular void lines of sand–silt mixtures. Canadian Geotechnical
ratio. Acta Geotechnica, 10(5): 587–606. Journal, 43(11), 1213–1219.
Muir-Wood, D & Maeda, K 2008. Changing grading of Yang, J, Wei, L M & Dai, B B 2015. State variables for
soil: Effect on critical states. Acta Geotechnica, 3(1): silty sands: Global void ratio or skeleton void ratio?
3–14. Soils and Foundations, 55(1): 99–111.
Poulos, S J, Castro, G & France, J W 1985. Liquefaction
evaluation procedure. Geotechnical Engineering,
111(6): 772–792. NOTATION
Prochaska, A B & Drnevich, V P 2005. One-point e = void ratio
vibrating hammer compaction test for granular soils. p’ = mean effective stress (σ’1 + σ’2 + σ’3)/3
Proceedings, ASCE Geo-Frontiers Conference held in q = deviator stress (σ’1 – σ’ 3)
Austin, TX, 24–26 January. Reston, VA: ASCE. SSL = steady state line
Rahman, M M & Lo, S R 2008. The prediction of Γ1 = steady state void ratio that corre-
equivalent granular steady state line of loose sand sponds to a mean effective stress of
with fines. Geomechanics and Geoengineering, 3(3): 1 kPa
179–190. λ10 = slope of the SSL in e-log10(p’) space
Rahman, M M, Lo, S R & Dafalias, Y F 2014. Modelling FC = fines content (< 75 μm)
the static liquefaction of sand with low-plasticity PSD = particle size distribution
fines. Geotechnique, 64(11): 881–894. emin = minimum void ratio
Thevanayagam, S, Shenthan, T, Mohan, S & Liang, J TFC = threshold fines content
2002. Undrained fragility of clean sands, silty sands, Cu = coefficient of uniformity (D60/D10)
and sandy silts. Geotechnical and Geoenvironmental
Engineering, 128(10): 849–859.
Guidelines for the preparation of papers and technical notes
Authors should comply with the following guidelines when preparing papers for publication in the journal
The Journal of the South African Institution of Civil
Engineering is published quarterly in March, June,
September and December. Articles submitted for
publication are reviewed by a panel of referees under the
guidance of the SAICE Journal Editorial Panel. The jour-
nal publishes research papers covering all the disciplines
of civil engineering (structural, geotechnical, railway,
coastal/marine, water, construction, environmental,
municipal, transportation) and associated topics that
are relevant to the civil engineering profession, and that
preferably have particular relevance to civil engineering
in southern Africa and the African continent.
When preparing articles for publication, authors
should please take note of the following and comply with
the guidelines as set out:
CLASSIFICATION OF ARTICLES
CONSIDERED FOR PUBLICATION
■■ Technical papers are well-researched, in-depth, fully
referenced technical articles not exceeding 6 000
words in length (excluding tables, illustrations and
the list of references). Related papers that deal with
‘softer sciences’ (e.g. education, social upliftment, etc)
are accepted if they are of a technical nature and of
particular interest to the civil engineering profession.
The latter type of paper will be subject not only to
peer-review by civil engineers, but also to review by
non-engineering specialists in the field covered by
the paper.
■■ Technical notes are short, fully referenced technical
articles that do not exceed 2 000 words. A typical
technical note will have limited scope often dealing
with a single technical issue of particular importance
to civil engineering.
■■ Review papers are considered for publication as
either technical papers or technical notes on condi-
tion that they are the original work of the author
and will assist the reader with the understanding,
interpreting or applying of the subject under review.
A review paper must contain criteria by which the
work under review was evaluated, and contribute
by synthesising the information and drawing new
conclusions from the dissemination of the previously
published work.
■■ Discussion on published articles is welcomed up to
six months after publication. The length of discus-
sion contributions is limited to 1 500 words. Where
appropriate, discussion contributions will be subject
to the normal reviewing process and will be for-
warded to the authors of the original article for reply.
POLICY REGARDING LANGUAGE AND
ORIGINALITY OF SUBMITTED ARTICLES
■■ Language: Manuscripts should preferably be
presented in English, as the journal is distributed
internationally. Articles submitted in any of the other
official South African languages should be accompa-
nied by an expanded abstract in English.
■■ Original work: Papers and technical notes must be
original contributions. Authors must affirm that sub-
mitted material has not been published previously,
is not under consideration for publication elsewhere
and will not be submitted elsewhere while under
consideration by the SAICE Journal Editorial Panel. It
is the responsibility of the authors to ensure that pub-
lication of any paper in the journal will not constitute
a breach of any agreement or the transgression of any
law. The corresponding author should confirm that
all co-authors have read and approved the manuscript
and accept these conditions. Authors are responsible
for obtaining permission to publish experimental
data and other information that may be confidential
or sensitive. Authors are also responsible for obtain-
ing permission from copyright owners when repro-
ducing material that had been published elsewhere.
Proof of such permission must be supplied.
SUBMISSION PROCEDURES
AND REQUIRED FORMAT
■■ Online submission: Manuscripts must be uploaded
as PDF files (http://journal.saice.org.za). Individual
file sizes may not exceed 4 MB. Should you experi-
ence problems uploading your paper, please contact
the editor (verelene@saice.org.za).
■■ Format: Manuscripts should be prepared in MS
Word and presented in double line spacing, single
column layout with 25 mm wide margins. Line
numbers must be applied to the whole document.
All pages should bear the authors’ names and be
numbered at the bottom of the page. With the excep-
tion of tables and figures (see below) the document
should be typed in Times New Roman 12 pt font.
Contributions should be accompanied by an abstract
of not more than 200 words.
■■ First page: The first page of the manuscript should
include the title of the paper, the number of words of
the main text (i.e. excluding figures, tables and the
list of references), the initials and surnames of the
authors, professional status (if applicable), SAICE
affiliation (Member, Fellow, Visitor, etc), telephone
numbers (landline and mobile), and e-mail and
postal addresses. The name of the corresponding
author should be underlined. Five keywords should
be suggested.
■■ Figures, tables, photos and illustrations: These
should preferably be submitted in colour, as the jour-
nal is a full-colour publication.
■■ Their positions should be clearly marked in the text
as follows: [Insert Figure 1].
■■ Figures, tables, photos, illustrations and equations
should be numbered consecutively and should
appear in the text directly after they have been
referred to for the first time.
■■ Illustrations must be accompanied by appropriate
captions. Captions for tables should appear above
the table. All other captions should appear below
the illustration (figures, graphs, photos).
■■ Only those figures and photographs essential to
the understanding of the text should be included.
All illustrations should be referred to in the text.
■■ Figures should be produced using com-
puter graphics. Hand-drafted figures will not be
­accepted. Lettering on figures should be equiva-
lent to a Times New Roman 9 pt font or slightly
larger (up to 12 pt) if desired. Lettering smaller
than 9 pt is not acceptable.
■■ Tables should be typed in Times New Roman 9 pt
font. They should not duplicate information already
given in the text, nor contain material that would
be better presented graphically. Tabular matter
should be as simple as possible, with brief column
headings and a minimum number of columns.
■■ Mathematical expressions and presentation
of symbols:
■■ Equations should be presented in a clear form
which can easily be read by non-mathematicians.
Each equation should appear on a separate line
and should be numbered consecutively.
■■ Symbols should preferably reflect those used in
Microsoft Word Equation Editor or Mathtype,
or should be typed using the Times New Roman
symbol set.
■■ Variables in equations (x, y, z, etc, as well as lower
case Greek letters) should be presented in italics.
Numbers (digits), upper case Greek letters, sym-
bols of metric measurement units (m for metres,
s for seconds, etc) and mathematical/trigono-
metrical functions (such as sin, cos and tan) are
not written in italics, but in upright type (Roman).
Variables and symbols used in the body of the text
should match the format used in the equations,
i.e. upright or italics, whichever is applicable.
■■ Metric measurement abbreviations/units should
conform to international usage – the SI system of
units should be used.
■■ Decimal commas may be used, but decimal points
are preferred.
■■ Symbols should preferably be defined in the text,
but if this is not feasible, a list of notations may be
provided for inclusion at the end of the paper.
■■ Headings: Sections and paragraphs should not be
numbered. The following hierarchy of headings
should be followed:
HEADING OF MAIN SECTION
Heading of subsection
Heading of sub-subsection
■■ References: References should follow the Harvard
system. The format of text citations should be as
follows: “Jones (1999) discovered that …” or “recent
results (Brown & Carter 1985; Green et al 1999)
indicated that …”
 References cited in the text should be listed in
alphabetical order at the end of the paper. References
by the same author should be in chronological order.
The following are examples of a journal article, a
book and a conference paper:
■■ Donald, A W 1954. Pore pressure coefficients.
Géotechnique, 4(4):143–147.
■■ Terzaghi, K 1943. Practical soil mechanics. New
York: Wiley.
■■ Smith, A W, Black, C & White, A 1999. Factors
affecting the strength of soils. Proceedings, ASCE
Conference on Shear Strength of Soils, Colorado,
pp 503–532.
Papers published previously in the Journal of the
South African Institution of Civil Engineering should
be cited if applicable.
■■ Footnotes, trade names, acronyms, abbreviations:
These should be avoided. If acronyms are used, they
should be defined when they first appear in the text.
Do not use full stops after abbreviations or acronyms.
■■ Return of amended papers: Papers requiring
amendments will be accepted up to six months after
the referee reports had been sent to authors, after
which the paper will be withdrawn from the system.
FINAL ARTICLE
■■ Copyright: On acceptance of the paper or technical
note, copyright must be transferred by the author/s to
the South African Institution of Civil Engineering on
the form that will be provided by the Institution.
■■ Photos of authors: The final corrected version of
the paper should be accompanied by recent, high-
resolution head and shoulders colour photographs
and a profile not exceeding 100 words for each of
the authors.
■■ Proofs: First proofs of papers will be sent to authors
in PDF format for verification before publication.
No major re-writes will be allowed, only essential
minor corrections.