Volume 63 Number 1

of the South African Institution of Civil Engineering March 2021

QQ  eliability basis for assessment of existing building

R
structures with reference to SANS 10160
QQ  easurement of probabilistic ballast
M
particle dynamics using Kli-Pi
QQ  eliability performance of bridges designed
R
according to TMH7 NA load model
QQ  he variability in commercial laboratory aggregate
T
testing for road construction in South Africa
QQ  eview of compatibility between SANS 10400 deemed-
R
to-satisfy masonry wall provisions and loading code
QQ  esign and construction of laboratory-scale activated
D
carbon, gravel and rice husk filter columns for the treatment
of stormwater runoff from automobile workshops
Volume 63 No 1 March 2021 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 11 805 5947, Fax +27 11 805 5971
http://www.saice.org.za
civilinfo@saice.org.za

Contents
Editor-in-chief
Dr Peter Day
Consultant, Jones & Wagener (Pty) Ltd
Adjunct Professor of Geotechnical Engineering
University of Stellenbosch
Tel +27 11 519 0200
day@jaws.co.za
2 Reliability basis for assessment of existing building
joint EditorS-in-chief
Prof Chris Clayton structures with reference to SANS 10160
University of Southampton
c.clayton@soton.ac.uk M Holický, J V Retief, C Viljoen
Prof Gerhard Heymann
University of Pretoria
gerhard.heymann@up.ac.za 11 Measurement of probabilistic ballast
Editor
Verelene de Koker particle dynamics using Kli-Pi
Tel +27 11 805 5947, Mobile +27 83 378 3996
verelene@saice.org.za A Broekman, P J Gräbe
journal editorial Panel
Prof Y Ballim – University of the Witwatersrand
Prof W Burdzik – University of Pretoria 24 Reliability performance of bridges designed
Prof CRI Clayton – University of Southampton
Dr P Day – Jones & Wagener (Pty) Ltd (chairman)
according to TMH7 NA load model
Prof J du Plessis – University of Stellenbosch
Prof GC Fanourakis – University of Johannesburg
R Lenner, S E Basson, M Sýkora, P F Van der Spuy
Prof PJ Gräbe – University of Pretoria
Dr C Herold – Umfula Wempilo Consulting
Prof G Heymann – University of Pretoria 37 The variability in commercial laboratory aggregate
Prof A Ilemobade – University of the Witwatersrand
Prof SW Jacobsz – University of Pretoria testing for road construction in South Africa
Prof EP Kearsley – University of Pretoria
Prof C Roth – University of Pretoria I J Breytenbach, H G Fourie
Prof W Steyn – University of Pretoria
Mr M van Dijk – University of Pretoria
Prof C Venter – University of Pretoria
Prof A Visser – University of Pretoria
45 Review of compatibility between
Dr E Vorster – Zutari
Prof J Wium – University of Stellenbosch
SANS 10400 deemed-to-satisfy masonry
Prof A Zingoni – University of Cape Town wall provisions and loading code
Prof M Zuidgeest – University of Cape Town
Peer reviewing W I de Villiers, G P A G van Zijl, W P Boshoff
The Journal of the South African Institution of
Civil Engineering is a peer-reviewed journal
that is distributed internationally 61 Design and construction of laboratory-scale
Design and reproduction
Marketing Support Services, Ashlea Gardens, Pretoria
activated carbon, gravel and rice husk filter
Papers for consideration should be submitted columns for the treatment of stormwater
online at: http://journal.saice.org.za
runoff from automobile workshops
The South African Institution of Civil Engineering accepts no
responsibility for any statement made or opinion expressed C O Ataguba, I Brink
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 1

TECHNICAL PAPER Reliability basis for
Journal of the South African
Institution of Civil Engineering assessment of existing
ISSN 1021-2019
Vol 63 No 1, March 2021, Pages 2–10, Paper 1041 building structures with
PROF DR MILAN HOLICKÝ obtained his Civil
Engineering degree at the Czech Technical
reference to SANS 10160
University in Prague, his PhD at Waterloo,
Canada, and his DrSc (h.c.) from the
Moscow State University of Civil M Holický, J V Retief, C Viljoen
Engineering. He is currently employed at the
Klokner Institute, Czech Technical University
in Prague, and is an Extraordinary Professor
at Stellenbosch University, South Africa. He is involved in the research of The principles of structural reliability are firmly established to provide the basis for structural
structural reliability and risk assessment. He has written or co-authored performance as incorporated in design standards. Reliability-based procedures provide for
more than 300 scientific and technical publications, textbooks and five variabilities and uncertainties that could reasonably be expected during the design service
monographs, and is actively participating in international research within
life of the structure. However, not all possible conditions to which all structures are exposed
CIB, JCSS ISO, CEN TC250 Structural Eurocodes, and is project leader on the
assessment of existing structures. during such an extended service period can be accounted for effectively and economically
at the design stage. The assessment of the reliability performance of an existing structure
Contact details:
Klokner Institute therefore forms an integral part of structural management and engineering practice. Despite
Czech Technical University in Prague the lag between the development and implementation of a basis for design and assessment
Prague, Czech Republic of structures, the extensive body of information on structural assessment progresses steadily
T: +420 224 353 842, E: holicky@klok.cvut.cz
towards standardisation. This paper provides a review of the progress to provide background
information towards extending the scope of South African structural standards to include
PROF JOHAN RETIEF (Pr Eng, FSAICE) is
Emeritus Professor in Civil Engineering at guidance on structural assessment. The focus is on such developments to include provisions for
Stellenbosch University, South Africa. His existing structures in Eurocode, together with related investigations. In addition to a general
interests are in the application of structural review of background investigations, interrelationships between the basis of design and
reliability in the various fields of structural
assessment for Eurocode (potentially also for South Africa) are considered. The main elements
design standards. Accordingly, he has made
contributions to the development of design of a standardised basis for assessment are defined, and alternative assessment approaches are
standards nationally and internationally. He presented and demonstrated by representative examples. In conclusion, an outline of a possible
holds DEng degrees from Pretoria and Stellenbosch Universities and degrees development path for implementation in South Africa is provided.
from Imperial College, London, and Stanford University, California.

Contact details:
Department of Civil Engineering
Stellenbosch University
Private Bag X1, Matieland, Stellenbosch 7602, South Africa
T: +27 82 446 5084, E: jvr@sun.ac.za
PROF CELESTE VILJOEN (Pr Eng, FSAICE),
Vice-Dean Teaching in the Faculty of
Engineering at Stellenbosch University, was
until recently Head of the Division of
Structural Engineering and Civil Engineering
Informatics at the same university. Her
research considers structural risk and
reliability, including the assessment of various
structural standard provisions in reinforced concrete design, steel design and
loading. She has supervised many postgraduate students to completion and
has authored more than 50 published papers. She is widely involved in
structural standardisation, being a member of various SABS Technical
Committees. She is also convenor of the SANS 10100-3 Working Group, a
member of the Working Group for the Revision of ISO 13824, and a member
of the International Joint Committee on Structural Safety.
Contact details:
Department of Civil Engineering
Stellenbosch University
Private Bag X1, Matieland, Stellenbosch 7602, South Africa
T: +27 +27 21 808 4444, E: celesteviljoen@sun.ac.za
Keywords: structural reliability, existing structures, assessment,
standard procedures, applications
INTRODUCTION
Acceptance of the principles of structural
reliability and risk is manifested by the
introduction of the semi-probabilistic limit
states approach, such as the partial factor
or load and resistance factor methods, not
only for South Africa, but also interna-
tionally by many countries and interna-
tional organisations, such as ISO TC98,
the Joint Committee for Structural Safety
(JCSS) and the European Committee for
Standardisation CEN TC250. Key concepts
of the reliability-based approach are that
structural performance is set in terms of
target levels of reliability, with operational
design procedures that are calibrated to
exceed the target levels, whilst accounting
for the diverse variabilities and uncertainties
of all design or basic variables.
It is evidently not feasible to account,
during the design stage, for all possible condi-
tions to which the structure will be exposed
during a service life of multiple decades. An
extensive set of scenarios can be identified
where conditions to which the structure is
exposed, or even conditions of the structure
as such, deviate from the design basis. Whilst
the same principles of structural reliability on
which the basis of design rests would apply
to the reliability of the existing structure
throughout its service life, the specific condi-
tions requiring assessment of the structure
provide a first indication that operational
procedures for design and assessment are
bound to differ. At the minimum level, the
need for assessment of existing structures
arises from the basic requirement for validat-
ing reliability compliance after decades of
service. For structures designed to present
standards, strategic upper bounds for assess-
ment are required from needs such as life
extension for economic reasons or sustain-
ability objectives and future imperatives to
account for climate change during the service
life. Important classes of differences between
design and assessment are the options for
optimal decision-making under the respec-
tive conditions, including the state of infor-
mation availability and gathering at the stage
of decision-making.
Whilst the emphasis in structural
engineering practice is rightfully placed on

Holický M, Retief JV, Viljoen C. Reliability basis for assessment of existing building structures with reference to SANS 10160.
2 J. S. Afr. Inst. Civ. Eng. 2021:63(1), Art. #1041, 9 pages. http://dx.doi.org/10.17159/2309-8775/2021/v63n1a1
the design of new structures, the bulk of
structures is in use, requiring some degree
of structural management. Based on a
growth rate of a few percent of the stock of
structures over a 50-year service life, the
ratio of existing-to-new structures is around
30, with 3 to 6 structures within their last 10
to 20 years of service life for each new struc-
ture being designed. Additional complica-
tions are early obsolescence or extended
service life that will both have implications
for the relative importance of the reliability
performance of existing structures.
Professional duties regarding both public
and economic interests should include the
reliability of the collection of existing struc-
tures. A standardised approach towards reli-
ability assessment has the potential for the
most effective manner of dispensing with
professional responsibilities, even if only the
requirements for the basis of assessment
are provided in a standardised format. The
primary objective of this paper is to dem-
onstrate that the principles of reliability and
its conversion into operational requirements
and procedures are sufficiently mature in
standardised design to be converted also
to the assessment of existing structures,
in spite of the lag between development of
these two modes of managing and decision-
making on structural reliability.
Eurocode development of
provisions for existing structures
This paper provides a review of recent
developments for the introduction of the
reliability basis for the assessment of exist-
ing structures in the format of standardised
requirements, and alternative approaches
and procedures. The survey includes back-
ground information that served as input to
the development process, including historic
progression of provisions for the closely
related respective fields of design and
assessment. The focus is primarily on the
extension of the Eurocode Head Standard
EN 1990 (EN 2002) to include assessment,
as envisaged in prEN 1990 (EN 2020). The
formulation of Assessment and retrofit-
ting of existing structures (CEN/TS 17440
2020) and the accompanying Background
Documents represent a key step in the
development process. The CEN/TS is based
on a number of national standards related
to the assessment of existing structures,
compiled in a Joint Research Centre report
(JRC 2015). The paper provides an exten-
sion of the overview of recent develop-
ments provided by Holický et al (2019) and
consideration of the applicability of these
Journal of the South African Institution of Civil Engineering Volume 63 Number 1 March 2021
methodologies to South Africa by Retief
et al (2019).
The International Standard ISO 13822
(ISO 2016a), first edition published in 2001,
served as point of departure for the CEN
Eurocode initiative. This standard, adopted
as South African National Standard 13822
(ISO 2016a), represents the most direct
provision for existing standards in this
country. The specific case of the assessment
of existing concrete structures is extensively
developed in the fib Bulletin 80 (2016), with
the proposed methodologies intended to be
compatible with the Eurocode standardised
approach, and with a view of extending the
scope of a future edition of the fib Model
Code for structural concrete. An extensive
review of theoretical concepts for the proba-
bilistic assessment of existing structures
that could serve as basis for operational
partial factor assessment methods is com-
piled by the Joint Committee on Structural
Safety (JCSS 2001a), complementing the
Probabilistic Model Code (JCSS 2001b);
the latter arguably being biased towards
the design process. These standardisation
advances are based on an extensive body
of literature and research, represented for
example by the seminal papers by Allen
(1991) and Ellingwood (1996), and more
recent extensive investigations by Holický
(2005), Vrouwenvelder and Scholten (2010),
Tanner et al (2011), Steenbergen et al (2015),
Sýkora et al (2017), and Holický (2017).
South African applications
The focus for the review and appraisal of
this extensive body of information is provid-
ed by relevance to the Head Standard for the
South African Loading Code SANS 10160-1
(SANS 2018) Basis of structural design
which provides the requirements for the
reliability performance of buildings and
similar structural systems. The background
to SANS 10160 is provided by Retief and
Dunaiski (Eds) (2009), including the reli-
ability principles (Retief & Dunaiski 2009),
the relationship to Eurocode (Retief et al
2009) and an outline of the implementation
of reliability principles in operational partial
factor design in Eurocode by Holický et al
(2009). The most significant differences of
the South African standard compared to
Eurocode are the limited scope of applica-
tion, target reliability levels and resulting
formats for action combinations, and the
relationship between respective standards
for actions and materials-based resistance.
These differences need to be reflected
when establishing the relationship between
design of new structures and assessment of
existing structures.
As a general guideline the conversion
from the wide-ranging principles and
procedures to the specific assessment
procedures proposed for incorporation
into SANS 10160 is intended to provide for
standardised general practice; serving as
a first round of assessment to identify the
need for it and specifics of more refined
investigations, assessment, decision-
making and intervention. As a first step,
procedures are directed towards comply-
ing with all the requirements that can be
regarded as being equivalent to that for
design, such as the relevant limit states and
design situations, yet fully adapted to the
existing structure under investigation.
The intention is that assessment proce-
dures are limited to the semi-probabilistic
approach, whilst requiring an appreciation
of risk and reliability concepts to be able
to adjust design parameters in accordance
with information on the existing structure.
Where the situation may arise that more
advanced methods are justified, such as the
application of full reliability-based assess-
ment or risk-based and risk-optimised
methods, the standardised approach should
serve as initial investigation leading to
specialist assessment using Eurocode pro-
cedures, whilst retaining consistency with
South African requirements.
Since the condition of the existing
structure related to its resistance generally
plays a primary role in its load-bearing
assessment, the scope of this review is
simplified by mainly considering concrete
structures, as opposed to the comprehen-
sive Eurocode approach where all structur-
al materials are included (in addition to all
the structural classes of buildings, bridges
and industrial and civil engineering struc-
tures). Even within the limited scope of this
review, a substantial fraction of existing
building structures will be covered. The
investigation on existing concrete struc-
tures by the International Federation for
Structural Concrete ( fib Bulletin 80 2016)
provides additional background informa-
tion to complement this survey. Some indi-
cation is provided for extending assessment
procedures to other structural materials,
mainly considering steel.
ASSESSMENT PROCESSES
AND PROCEDURES
Initial initiatives to formalise the assess-
ment of existing structures emphasised
3
the process of information gathering and
decision-making, whilst quantitative meth-
odologies were presented in general terms
of the principles of reliability and risk. Flow
charts of iterative processes of applying
more detailed investigations and decision
branches summarise the assessment pro-
cess for both the JCSS (2001a) report and
ISO 13822 (ISO 2016a). The requirements
for a code for the assessment and format
of its contents are provided by the JCSS
report, whilst the ISO Standard provides a
pro forma layout and specification. These
documents represent key contributions to
the advancement of standardised assess-
ment. However, essential quantitative
procedures are provided in theoretical and
conceptual terms, lacking standardised
procedures, such as approach and method-
ologies required to bring assessment on par
with operational design standards.
The development of CEN/TS 17440
(abbreviated as CEN/TS) for the assess-
ment of existing structures represents a
significant advancement in closing the gap
between standardised design and assess-
ment. The presentation of a set of method-
ologies for the verification of the reliability
of an existing structure forms an important
part of this advancement. The verification
methodologies are consistent with the
relationship between the risk-based and
reliability-based approaches for semi-prob-
abilistic partial-factor methods specified
by ISO 2394 (ISO 2016b). Since ISO 13822
(ISO 2016a) has served as starting point
for the CEN/TS, the general format is also
consistent with that International Standard.
The implication is that the adaptation of the
CEN/TS to South African standards will be
consistent with the adopted International
Standards SANS/ISO 2394 (ISO 2016b) and
SANS/ISO 13822 (ISO 2016a).
Integration of the CEN/TS into prEN
1990 (EN 2020) should result in the final
step of the advancement of standardised
assessment. In accordance with the com-
prehensive nature of Eurocode, all classes
and types of buildings and civil engineering
infrastructure, bridges, industrial and geo-
technical structures exposed to the inclusive
range of actions covered by the standard are
included in the scope of the CEN/TS.
The range of CEN/TS topics provides for
the basis of assessment that is equivalent to
the basis for design. The main topics include:
QQ The scope of conditions that may lead to
the need for assessment of the structure
QQ Requirements and prerequisites in
accordance with current standards
4 Volume 63 Number 1 March 2021
QQ The general framework for assessment
to determine their actual conditions
QQ Methodologies for gathering information
and updating of basic variables for the
representation of actions and material
properties
QQ Structural analysis for quantitative verifi-
cation of structural elements and systems
QQ Alternative verification methods
ranging from partial factors to risk
assessment
QQ Construction and operational
interventions
QQ Consideration of the special case of the
assessment of heritage buildings.
Since quantitative reliability verification
plays a central role in the assessment process
and represents the most significant recent
advancement, this paper focuses mainly on
verification processes. Other topics are con-
sidered mainly from the perspective of their
relationship to reliability verification.
Scope of situations requiring
assessment
The scope of conditions and situations
where the need for the reliability assessment
of an existing structure arises can be related
to the reliance of modern societies and
economies on the built environment, where
the structural or load-bearing behaviour
of the facility forms an integral function.
Trends in the continued use of structures,
particularly for public buildings, but often
in conjunction with economic constraints,
require the application of advanced methods
that may exceed the level of sophistication
of design standards. The socio-economic
differences between the design of new struc-
tures and the assessment of existing struc-
tures, as summarised in Table 1, provide an
indication of the context to be applied to the
respective requirements (Holický 2005).
More specifically, the nature and level
of assessment of an existing structure are
closely related to the need and motivation
for such an assessment. Any situation where
the condition of the structure deviates from
the requirements and assumptions used by
present structural standards may require an
Table 1 D
 ifferences between the design of new structures and the assessment of existing structures
Aspect Design of new structures
Marginal costs of reliability
Economic
improvements are usually low
Restrictions are usually less
Social
significant than in existing structures
New materials are often used,
Sustainability
sustainability is difficult
assessment to verify acceptable reliability
and optimal performance. Such conditions
may include exceedance of the design ser-
vice life, outdated standards, deteriorating
and obsolete materials, damage to the struc-
ture, the need for upgrading, refurbishing or
life extension, and enhancing sustainability.
In principle, possible critical situations
for the structure should be identified in
advance in the specification of the future
performance of the structure in terms of
utilisation and safety planning.
Various lists of circumstances leading
to the initiation of assessment of the struc-
ture are provided as an indication of the
scope of the document. These conditions
should serve to determine the objectives,
planning and methodology to be followed.
The following list is based on the CEN-TS
document, as elaborated on by Holický et
al (2019) and extended by the JCSS (2001a)
list, differentiating between circumstances
that may have a bearing on how the assess-
ment is to be done:
QQ Deviations from the original project
description are observed
QQ Adverse results of a recent investigation
or trends from periodic investigations
QQ Concern about the structural safety
caused by evidence of damage
QQ Deterioration due to time-dependent envi-
ronmental actions (e.g. corrosion, fatigue)
QQ Unusual incidents during use which
could have damaged the structure (such
as impact of vehicles, avalanches, fire,
earthquake)
QQ Suspicion of possible impairment of the
structural safety related to structural
materials, construction methods or the
structural system
QQ Discovery of design or construction errors
QQ Planned change of the use of the
structure
QQ Changes in the structural system (e.g.
retrofitting, modifications, extensions
to the structure)
QQ A change of loads and loading conditions
QQ Extension of the design working life or
expiry of residual service life granted
from earlier assessment
Assessment of existing structures
Marginal costs of reliability improvements are
usually high
Restriction of the use and damage of
economic and cultural assets are significant
Allowance for sustainability is enhanced
substantially by using original materials
Journal of the South African Institution of Civil Engineering
QQ Requirement of authorities, insurance
companies or owners, or demanded by a
maintenance plan
QQ Simply because of doubts about the
safety of the structure.
The proper management of a structure
throughout its service life could be added
to the list, with the process starting as early
as the commissioning of a new building.
More detailed specification of the
assessment process can be formulated
for classes of circumstances, such as a
general class where the integral reliability
needs consideration, concern about vari-
ous classes of resistance impairment, and
circumstances related to loading.
Reliability requirements
The central role of risk-based and reliabil-
ity-based procedures for the design of new
structures clearly also applies to the assess-
ment of existing structures with the objec-
tive to verify acceptable levels of structural
performance. Reliability requirements for
assessment should therefore be consistent
with those embedded in design procedures,
both in terms of target levels of reliability,
and in reliability differentiation between
limit states and consequence classes for
structures. Differences in reliability targets
and classes in EN 1990 and SANS 10160-1
are therefore an important issue in adapt-
ing CEN/TS procedures for inclusion in the
South African standard.
The comparison between design
and assessment summarised in Table 1
provides a clear indication of the need
for adapted reliability requirements for
assessment. The derivation of appropriate
reliability levels of existing structures is
actively considered in the literature – see,
for example, Vrouwenvelder and Scholten
(2010), Steenbergen et al (2015), Holický et
al (2015), Sýkora et al (2017), and Holický
(2017). More elaborate reviews of the lit-
erature are captured by background inves-
tigations, such as JCSS (2001a), Lüchinger
et al (2015), fib Bulletin 80 (2016). The
relevant information from the literature is
effectively embedded in the CEN/TS docu-
ment but could serve as background input
towards the extension of SANS 10160-1
to account explicitly for the assessment of
existing structures.
In addition to the general motivation of
refining the rational basis for determining
reliability levels, the main issues related to
assessment are safety costs, varying uncer-
tainties, time-related reference periods and
remaining service life.
Journal of the South African Institution of Civil Engineering Volume 63 Number 1 March 2021
Table 2 Definition of consequences classes
The more severe consequences of
Consequences class
loss of human life
CC4 Highest consequences Extreme
CC3 Higher consequences High
CC2 Normal consequences Medium
CC1 Lower consequences Low
CC0 Lowest consequences Very low
Reliability targets and classes
According to the Eurocode basic require-
ments, the selection of reliability levels
by member states should be based on the
following considerations:
QQ The possible consequences of failure
in terms of risk to life, injury, potential
economic losses
QQ The possible cause and/or mode of
attaining a limit state
QQ Public aversion to failure
QQ The expense and procedures necessary
to reduce the risk of failure.
Reliability levels which are related to the
probability of structural failure are related to
the consequences of failure, which are clas-
sified into five consequence classes (CC0 –
CC4) that depend on societal and economic
effects, as listed in Table 2. Eurocode pro-
vides for CC1 – CC3, considering the two
extreme classes (CC0 and CC4) to be outside
its scope. Indicative reference target reli-
ability levels for the ultimate limit state are
given in prEN 1990 (EN 2020), as listed in
Table 3 for the one-year reliability index β1
and failure probability Pf to be determined
by member states. Procedures are based on
CC2, typically with adjustments indicated
for other consequence classes.
The values in Table 3 are based on pre-
vious studies and Annex C of prEN 1990
(EN 2020); seismic situations are excluded.
It is not specified whether the values in
Table 3 are applicable to accidental and fire
design situations, and recommendations for
Serviceability Limit States are absent.
The final draft of prEN 1990 (EN 2020)
does not include possible transformation
of the reliability level related to one year
to levels related to other reference periods,
even though this is needed to specify reli-
ability elements for design and assessment
of common structures.
Adjustment of reliability level
Transformation of target reliability index
β related to different reference periods
of n years is well known when the main
economic
Huge
Very great
Considerable
Small
Insignificant
Table 3 I ndicative reliability levels related
to one year and ultimate limit states
given in prEN 1990 (EN 2020)
CC1 CC2 CC3
Pf = 10 –5 Pf = 10 –6 Pf = 10 –7
β1 = 4,2 β1 = 4,7 β1 = 5,2
uncertainty comes from actions that have
statistically independent maxima in each
year. Then the reliability (complementary to
failure probabilities) Φ(βn) related to the ref-
erence period of n years is determined from
annual reliability Φ(β1), where the index β1 is
related to one year, as the product of n annual
reliabilities, thus on average as [Φ(β1)]n.
Consequently, the reliability index βn can be
assessed from β1 using the expression indi-
cated in the Eurocode EN 1990 (EN 2002):
Φ(βn) = [Φ(β1)]n(1)
Equation 1 indicates that, for mutually
independent occurrences of failure in subse-
quent years, the commonly used reliability
index βn = 3,8 for n = 50 corresponds to
β1 = 4,7. However, the statistical maxima of
actions (and other time-dependent variables)
in subsequent years are usually correlated.
Consequently, the occurrences of failures
in subsequent years are inter-dependent.
Then the relationship (Equation 1) should be
generalised to take the correlation of failure
events in subsequent years into account. A
procedure for adjustment of reliability to
different reference periods for correlated
events is provided by Holický et al (2018).
Observations on target reliability
The following conclusions are related to
the target reliability level:
QQ The target reliability levels recommend-
ed in various national and international
documents are inconsistent in terms of
the values and reference periods.
QQ In the latest draft prEN 1990 (EN 2020)
of the revision of Eurocode EN 1990 the
5
 target reliability level is indicated only for
one-year and 50-year reference periods.
QQ Transformation formulae for adjustment
of the reliability level to different refer-
ence periods, taking mutual dependence
of failure probabilities in subsequent
years into account, are missing.
QQ The proposed transformation formula for
reliability index βnk depends on the refer-
ence period n and independence interval k.
QQ Reliability index βnk decreases with the
reference period n and increases with
the independence interval k.
QQ When determining the target reliability
index, the assumption of annual inde-
pendence of failures (k = 1) may be unsafe.
RELIABILITY VERIFICATION
Reliability verification of an existing struc-
ture shall be made using valid codes of
practice, as a rule based on the limit state
concept. Attention should be paid to both
the ultimate and serviceability limit states.
Verification may be carried out using partial
safety factor or structural reliability methods
with consideration of the structural system
and ductility of components. The reliability
assessment shall take the remaining working
life of the structure into account, the refer-
ence period, and anticipated changes in the
environment of the structure.
The conclusion from the assessment
shall withstand a plausibility check.
Discrepancies between the results of
structural analysis (e.g. insufficient safety)
and the real structural condition (e.g. no
sign of distress or failure, satisfactory
structural performance) must be explained.
It should be kept in mind that many
engineering models are conservative and
cannot always be used directly to explain
an actual situation.
The target reliability level used for
verification can be taken as the level of
reliability implied by the acceptance criteria
provided in valid design codes. The target
reliability level shall be stated together with
clearly defined limit state functions and
specific models of the basic variables. The
target reliability level can also be established
taking the required performance level for
the structure into account, together with
the reference period and possible failure
consequences. In accordance with ISO
2394 (ISO 2016b), the performance require-
ments applied in the assessment of existing
structures are the same as those used in the
design of new structures. Lower reliability
targets for existing structures may be used
6 Volume 63 Number 1 March 2021
if they can be justified on the basis of eco-
nomic, social and sustainable consideration
(see Annex F to ISO 13822 (ISO 2016a)).
An adequate value of the reliability
index β should generally be determined
considering the appropriate reference peri-
od. For serviceability and fatigue the refer-
ence period equals the remaining working
life, while for the ultimate limit states the
reference period is in principle the same as
the design working life specified for new
structures (50 years for buildings).
Reliability of a structure is given by the
condition g(Xi) > 0, where g(Xi) denotes
the limit state function and Xi represents
the basic variables. Commonly the limit
state function can be considered in a
simplified form:
g(Xi) = R − E > 0 (2)
Here R denotes resistance while E denotes
load effect. The reliability condition
(Equation 2) can be verified by various
methods.
The following procedures are included
in the CEN/TS.
Partial factor method
When using the partial factor method,
the reliability requirement (Equation 2)
(g(Xi) > 0) is substituted by the condition:
g(xdi) = g(xd1, xd2, xd3, ...) > 0,
xdi = xki or xdi = xki γ(β) or
xdi = xki ⁄ γ(β)(3)
Here xdi denotes the design values of basic
variables Xi determined using their charac-
teristic values xki and relevant partial factors
γ(β). The partial factors γ(β) may be adjusted
taking the specified reliability index β and
actual characteristics of the basic variable
into account (see fib Bulletin 80 (2016)).
Assessment value method
The condition (Equation 2) is modified by
the requirement:
g(xdi) = g(xd1, xd2, xd3, …) > 0,
ΦXi(xdi) = Φ(–αiβ)(4)
Here αi denotes the FORM sensitivity fac-
tors and Φ the normal distribution function.
Probabilistic method
The requirement g(Xi) > 0 (Equation 2) is
examined by the failure probability:
P f = P{g(Xi)<0} < P f,t(5)
Here Pf,t denotes the target probability of
failure that is to be specified taking into
account economic and societal conse-
quences of failure, and the costs of improv-
ing structural reliability.
Risk assessment approach
The reliability is examined by acceptable
risk expressed in a symbolic form as:
Risk = Pf C = P{g(Xi) < 0}C < Risk t(6)
Here C generally represents any type of
economic and societal consequences,
and Risk t the relevant target risk level.
Appropriate target risk level Risk t is to be
specified individually, whilst accounting for
the specific condition of an assessed struc-
ture. This may be a complicated task, par-
ticularly in the case of heritage buildings,
where historical and artistic aspects are
usually also involved. A general flowchart
of the risk assessment procedure is shown
in the CEN/TS Annex A. Principles of risk
assessment may be used for cost optimisa-
tion procedures.
APPLICATION EXAMPLE
An existing building constructed in 1970
is to be renovated and a new assessment is
required. The following example is limited
to a simply supported reinforced concrete
panel (a prefabricated hollow core floor
panel) of a span L = 6 m. The panel should
be exposed to additional permanent load
due to a newly designed floor surface. The
following requirements are specified: the
remaining working life is 50 years, and the
target reliability index related to the refer-
ence period of 50 years is βt = 3,8. The tar-
get reliability can be adjusted to the actual
economic and societal conditions. Table 4
indicates expected characteristic values of
actions specified in accordance with valid
standards, the load effect (the mid-span
point bending moment) and the corre-
sponding current capacity of the panel.
The load effect E and the resistance R of
the panel are expressed as bending moments
at the mid-span point of the panel. The
characteristic values of the actions Ek and
resistance Rk alone cannot be used to assess
structural reliability; however, when Ek < Rk
this may indicate some safety margin.
The structural reliability of the consid-
ered concrete panel can be verified consid-
ering the limit state function (Equation 2)
in a common form used for bending capac-
ity of reinforced concrete cross section:
Journal of the South African Institution of Civil Engineering
Table 4 Characteristic values of actions and resistance
Permanent load G
Characteristic values 6.26 kN/m2
Characteristic total actions
Characteristic Ek and Rk
Table 5 Results of the applied assessment methods
Load effect
Assessment method
(kNm)
48.3
Partial factors of Eurocodes
45.1
Adjusted partial factors 41.9
Assessment values 40.3
Probabilistic method – –
g(Xi) = R
 – E = As f y(d − Asf y /(2bfc))
− (g+q) L2/8(7)
Here As denotes reinforcement, f y strength
of reinforcement, fc strength of concrete,
d the effective depth and b the width of the
panel cross section.
The application of the partial factors for
design employing EN 1990 Equations (6.10)
and (6.10a) respectively serves as reference for
the assessment. A set of partial factors, which
are adjusted for assessment, is applied to all
the design variables for verification. Next,
the probability models for the basic variables
are adjusted and applied in accordance with
the assessment value approach. Finally, a
reliability assessment compares the reliability
achieved to the target. The results of the panel
assessment, including recommended deci-
sions based on the applied verification meth-
ods, are indicated in Table 5. The consecutive
steps demonstrate the effects respectively
of reduced standardised assessment partial
possible structural or operational interven-
tions are considered. Then the procedure
Imposed load Q Panel resistance R
for the cost optimisation indicated below
1.50 kN/m2 43.2 kN/m2 may be useful. In such a case the total cost
7.76 kN/m2 – Ctot(x) including possible structural or
operational interventions may be consid-
34.9 kNm 43.2 kNm
ered in the fundamental form (indicated in
COMREL):
Ctot(x) = C0 + C1x + Cf P f (x)(8)
Resistance Reliability
Results
(kNm) index β
Here x = R/R0 denotes the intervention
parameter (material consumption) to
37.6 – Negative
increase the present resistance R0 to a
desired level R, C0 denotes necessary costs
39.3 Assumed 3.8 Negative
of intervention independent of the param-
41.0 Assumed 3.8 Positive eter x, C1 denotes marginal costs per unit of
the parameter x, Cf denotes costs due to fail-
Assessed 4.2 Positive
ure and malfunctioning, and Pf (x) denotes
the probability of failure and malfunction-
remaining working life of 50 years the ing. Figure 1 (obtained using the software
considered panel seems to be sufficiently product COMREL) indicates variation of
reliable. the total costs Ctot(x) with the parameter x,
Cost optimisation procedures may be for selected relative costs C0 = 0, C1 = 10,
particularly effective when the reliability of Cf = 1 000 000 and structural failure Pf (x)
a minor structure is acceptable (as in the of the panel having the resistance x R0.
case of the assessed floor panel); however, Figure 1 indicates that the optimum
to increase the current reliability level increase of the resistance R seems to be
for the required remaining working life, 10 percent (R/R0 ≈ 1,10). However, the
30
25 Ctot = C1 R/R0 + Cf Pf
20
Cost

factors, assessment values accounting mainly 15

for reduction in the coefficient of variation for
the basic variables, and the reliability analysis
to obtain the most likely limit state.
It follows from Table 5 that the partial‑ 10 C1 R/R0
factor methods lead to negative results
when both the partial factors recommended
in Eurocodes and the adjusted partial
factors obtained considering the index 5
β = 3,8 are applied. The assessment value
method (when index β = 3,8 is assumed to
Cf Pf
determine the assessment values) leads to
a positive result. The probabilistic method
0
also indicates a positive result, as the result- 1.00 1.05 1.10 1.15 1.20
ing reliability index β = 4,2 is greater than R/R0
the required reliability level corresponding
to the index β = 3,8. Therefore, for the Figure 1 Variation of the total costs Ctot with the resistance increase R/R0

Journal of the South African Institution of Civil Engineering Volume 63 Number 1 March 2021 7
convincing result should be based on appro-
priate failure cost Cf and marginal costs C1.
This simple illustrative example of a
floor panel may provide guidance to the
assessment of a one-off structure, for
example of a column or bridge element.
It should be emphasised that the actual
conditions and available data concerning
action and resistance variables should
always be accounted for. When only
limited data is available, then advanced sta-
tistical methods (including data updating)
can provide valuable techniques. In general,
the relevant target reliability level (index β)
and sensitivity factors αE and αR should be
re-examined. In some significant cases the
risk assessment method indicated in the
CEN-TS may be advisable. When possible
structural or operational interventions are
considered (to increase the current reliabil-
ity level even when it seems to be accept-
able), then the cost optimisation of possible
interventions may be applied.
Summary of Eurocode assessment
approach
Advanced methods of structural risk and
reliability are required to account for
the extended scope of the basis for the
assessment, in comparison to established
procedures for the design of new structures.
The set of assessment procedures presented
here can be used to account for increasing
levels of refinement associated with more
demanding conditions or reliability classes.
The partial factor method allows for the
direct adjustments to account for modified
reliability targets and uncertainty. The
design value, probability and risk methods
allow for increasingly taking benefit of the
embedded conservatism of the approxima-
tions of the semi-probabilistic approach.
More generally, the advanced reliability
and risk-based methodologies are not only
suitable for the assessment of existing struc-
tures, but also demonstrate the underlying
depth of the basis of design of EN 1990 /
prEN 1990.
The main principles of the upcoming
European document on assessment of
existing structures and building stock can
be summarised as follows:
QQ Currently valid codes for the verifica-
tion of structural reliability should be
applied, while historic codes valid in the
period when the structure was designed
should be used as guidance documents.
QQ Actual characteristics of structural
material, actions, geometric data and
structural behaviour should be
8 Volume 63 Number 1 March 2021
Table 6 Basic reliability elements for the ultimate limit state
Reliability element
Ultimate LS βT (50-year)
Permanent load γG
Variable load γQ
Reinforcement γs
Concrete composite
Concrete γc
Table 7 Underlying probability models for basic variables
Basic variable
Permanent load G
Imposed
Variable load Q
Wind
Reinforcement
Resistance
Concrete
considered; the original design docu-
mentation, including drawings, should
be used as guidance information only.
STANDARDISED ASSESSMENT
FOR SANS 10160-1
Justification for the incorporation of
assessment procedures arises from the
scope definition that SANS 10160 is also
applicable for the structural appraisal of
existing structures, but without providing
any further guidance on the basis for such
an appraisal. This implies rather strict
and arguably unreasonable requirements
and verification rules for these conditions.
The challenge is to derive standardised
procedures for assessment that are effec-
tively equivalent to the requirements for
the design of new structures, but with
appropriate adaptation to conditions apply-
ing to existing structures. The equivalence
requirement is not just an obligation, but
has the advantages of a well-established
reliability framework and built-in margins
that can be exploited to counter many
complications of assessment.
The three components of the conver-
sion of the basis of design into assessment
procedures are: (i) to identify the essential
reliability elements that need to be adapted,
(ii) to revert back to the reliability models
on which they are based, and (iii) to modify
the reliability elements in accordance
with equivalent assessment requirements.
An illustrative example is given here to
indicate how reliability management can
Reliability class
RC2 RC3
3.0 3.5
1.2
× KF = 1.1
1.6
1.15
Quality adjustment
1.5
Distribution Basic function f(β) Mean VX
Normal 1 + αE βT VG 1.05 0.10
0.96 0.24
1 – VQ (0.45 + 0.78LN
Gumbel
(–LN(Φ(–αE βT )))
0.7 0.5
0.10
Lognormal EXP(αR βT VR) 1.0
0.15
be employed to derive standardised partial
factors for assessment.
The β-values for the reference reli-
ability classes for the ultimate limit
state and typical partial factors, as
summarised in Table 6, are taken here
to represent the standardised basis of
design (SANS 10160‑1). Partial factors are
specified for the reference reliability class
RC2, with adjustment for RC3 using the
multiplicative factor (K F). Partial factors for
resistance are specified in the materials-
based standards, represented here by
the values for steel reinforcement and
concrete as given by SANS 10100-1:2000
(SANS 2000), where no reliability dif-
ferentiation is employed, but adjustment of
quality control measures is recommended
in the basis of design.
Notably this sparse set of information
is sufficient to do design verification for
an essential limit state, even allowing for
reliability differentiation for an extensive
range of building structures, such as resi-
dential and office buildings up to 4 and 15
storeys for reliability classes RC2 and RC3
respectively.
Reverse modelling of the partial fac-
tors listed in Table 6 can essentially be
represented by the probability models
for the basic variables as listed in Table 7,
as extracted from background literature
(Kemp et al 1987; Retief & Dunaiski 2009;
Holický et al 2010; Botha et al 2018). The
extensive body of reliability modelling
that resulted in the provisions for the
basis of design is deemed to be essentially
Journal of the South African Institution of Civil Engineering
Table 8 Partial factors for design adjusted to assessment level
γG
Application level
RC2 RC3 RC2
Design 1.2 1.25 1.6
Assessment I 1.15 1.2 1.4
Assessment II 1.15 1.15 1.25
represented by these probability models
through the standardisation decision-
making process. All the essential features
relating the partial factors to the reliability
model are captured by these expressions.
The coarse resolution of the basis of
design procedures are confirmed by the
application of a single partial factor for
variable actions γ Q = 1.6 given in Table 6,
despite the significant differences in the
distribution parameters for the leading
variable actions of imposed and wind load
listed in Table 7. Structural resistance is
represented by concrete as a composite
material with V X of 0.10 and 0.15 repre-
senting the control of resistance by steel
reinforcement and concrete, respectively.
The final main step in adapting design
procedures for assessment consists of
the combination of the information
from Tables 6 and 7, with parameter
values adjusted to represent standardised
assessment situations. The β-value is a
key parameter in the process and will be
used here as an example to demonstrate
the process.
The reference β-values of 3.0 and 3.5
are based on calibration to existing prac-
tice for South Africa (Milford 1985; 1986).
This range of values corresponds with the
SANS/ISO 2394 value of 3.3 for normal
buildings and medium cost of safety,
adjusted from 1-year to 50-year reference
periods according to Equation 1. Extensive
optimisation by Fischer et al (2019) (their
Figure 3) confirms that these values can be
regarded as economically optimised (βO).
The potential for adjustment is indicated
by the observation that βO falls within the
ALARP (as low as reasonably practical)
range between the required societal upper
limit (βR) and the broadly acceptable lower
limit. An indication of adjustment intervals
is obtained from the reduction Δβ of 0.9 –
1.3 between normal and large cost-of-safety
measures for the range of consequence
classes considered in SANS/ISO 2394.
The effect of the adjustment of the
β-values can be demonstrated in a sen-
sitivity analysis at different assessment
Journal of the South African Institution of Civil Engineering Volume 63 Number 1 March 2021
γQ γs γc
RC3 RC2 RC3 RC2
1.8 1.15 1.2 1.5
1.6 1.1 1.15 1.4
1.4 1.05 1.1 1.35
levels. A conservative value of Δβ = 0.5 is
recommended by Diamantidis and Sýkora
(2019) as assessment class I, and Δβ = 1.0
as assessment class II, based on the SANS/
ISO 2394 classes of the cost of safety.
Standardised sensitivity factors α E and α E
of 0.7 and 0.8 are applied to action effects
and resistance respectively. Since the
γ-values listed in Table 7 incorporate provi-
sions for model uncertainty (characteristic
bias amongst others), they are adjusted
in a normalised manner by the ratio
f(β – Δβ) / f(β), where f(β) represents the
probability function listed in Table 7. The
standardised partial factors for design are
compared in Table 8 to the derived values
for the two levels of assessment and the
two reliability classes, rounded off to the
customary values used in standards. The
standardised partial factors are based on
the coarse classification of safety costs and
consequences and generic optimisation. It
can be improved by proper risk optimisa-
tion when specific information is available
by using the methodologies provided by the
CEN/TS.
The adapted assessment partial fac-
tors provide a reasonable and apparently
realistic set of values across the range of
reliability and assessment classes, at least
more nuanced than the K F adjustment used
for design. In principle, a more rigorous
approach would reflect separate scaling
of model uncertainty and characteristic
bias as incorporated in the set of partial
factors for design. However, the objective
is primarily to illustrate equivalence of
assessment with design, which is essentially
achieved by the results given in Table 8.
Notable inconsistencies and approxima-
tions in the design provisions place a limit
on the level of refinement for the derivation
of assessment values. In addition to the
approximate provisions for variable actions
pointed out above, it should be noted that
the SANS 10160-1 partial factors closely
agree with (some) Eurocode values despite
a Δβ = 3.8 – 3.0 between the two stand-
ards. Systematic background investigations
for the formal introduction of provisions
for the assessment of existing structures
should explore the implications for all
these and other issues on the conversion
RC3 from design to assessment. The extensive
Joint Research Centre (JRC 2015) survey
1.6
on international practice should provide
1.5 useful guidance on converting the CEN/TS
procedures into operational rules for South
1.4
African conditions.
CONCLUSIONS
The development of procedures for the
assessment of existing structures that
are based on the principles of risk and
reliability still fits into the partial factor
limit state approach for the design of new
structures presented in this paper, and
represents a significant advancement of
structural engineering practice. These
procedures form the basis for the CEN
Technical Specifications for incorporation
of the assessment of existing structures
into the next version of the Eurocode Head
Standard prEN 1990 (EN 2020). Due to
the harmonised relationship of the South
African Standard SANS 10160-1 to its
Eurocode counterpart, this provides an
opportunity also to introduce standardised
assessment into this standard.
This paper demonstrates the series of
progressively more advanced assessment
methodologies that could be applied to
obtain more refined verification results,
based on the sequence of more rigorous
reliability-based and risk-based analyses.
An example is used to demonstrate the
effectiveness of subsequent methodologies.
The introduction of provisions for
the assessment of existing structures
to South Africa is considered from the
perspective of applying semi-probabilistic
partial factor limit states procedures for
building structures in accordance with
SANS 10160-1 for use in general practice.
The review provides an outline of the steps
that need to be followed in the background
investigations to arrive at appropriate
procedures that could be standardised.
Such a development would be similar to
the adaptation of Eurocode standards to
the South African Loading Code SANS
10160. A South African standard provid-
ing semi-probabilistic procedures that are
consistent with the CEN/TS would make it
possible, with specialist input, to also apply
the reliability and risk-based approaches
from the CEN/TS that are consistent
with South African conditions. A similar
policy is followed of limiting the scope of
9
SANS 10160 to normal design practice and
allowing for the use of specialist Eurocode
procedures where needed, rather than to
adopt Eurocode standards comprehensively
(Retief & Dunaiski 2009).
Two complementary perspectives are
therefore presented in the paper, namely
alternative methodologies for progressively
advanced assessment analysis and verifica-
tion for Eurocode, and a more operational
semi-probabilistic approach that could
be applied in standardised procedures.
The mutual advantages would be the
presentation of procedures that could be
used extensively in normal practice, based
on the advanced procedures that could be
used by specialists where the stakes are
high for assessment verification.
REFEFRENCES
Allen, D E 1991. Limit states criteria for structural
evaluation of existing buildings. Canadian Journal
of Civil Engineering, 18(6): 995–1004.
Botha, J, Retief, JV, Viljoen, C 2018. Reliability
assessment of the South African wind load
formulation. Journal of the South African Institution
of Civil Engineering, 60(3): 30–40.
CEN/TS 17440 2020. Technical Specification.
Assessment and retrofitting of existing structures.
Brussels: European Committee for Standardisation.
Diamantidis, D & Sýkora, M 2019. Reliability
differentiation and uniform risk in standards: a
critical review and a practical appraisal. Paper
presented at the Symposium on Future Trends in
Civil Engineering, 17 October 2019, Zagreb, Croatia.
EN (European Standard) 2002. EN 1990:2002.
Eurocode – Basis of Structural Design. Brussels:
European Committee for Standardisation:
EN (European Standard) 2020. prEN 1990:2002
(supersedes 1990:2002). Eurocode – Basis of
Structural and Geotechnical Design. Brussels:
European Committee for Standardisation.
Ellingwood, B R 1996. Reliability-based condition
assessment and LRFD for existing structures.
Structural Safety, 18(2–3): 67–80.
fib Bulletin 80 2016. Partial factor methods for existing
concrete structures. Lausanne, Switzerland:
International Federation for Structural Concrete
(fib).
Fischer, K, Viljoen, C, Köhler, J & Faber, M H 2019.
Optimal and acceptable reliabilities for structural
design. Structural Safety, 76: 149–161.
10
Holický, M 2005. Chapter 9: Assessment of existing
structures. In Ellingwood, B R & Kanda, J (Eds).
Structural Safety and its Quality Assurance. Reston,
VA: American Society of Civil Engineers (ASCE).
Holický, M, Retief, J V & Dunaiski, P E 2009. The
reliability basis of Eurocode. In Retief, J V &
Dunaiski P E (Eds). Background to SANS 10160.
Stellenbosch: SunMedia.
Holický, M, Retief, J V & Wium, J A 2010. Partial
factors for selected reinforced concrete members:
Background to the revision of SANS 10100-1.
Journal of the South African Institution of Civil
Engineering, 52(1): 36–44.
Holický, M, Diamantidis, D & Sýkora, M 2015.
Determination of target safety for structures.
Proceedings, 12th International Conference on
Applications of Statistics and Probability in Civil
Engineering (ICASP12), 12–15 July 2015, Vancouver,
Canada.
Holický, M 2017. Operational approach to assessment
of existing structures. Proceedings, 12th
International Conference on Structural Safety &
Reliability (ICOSSAR), 6–10 August 2017, Vienna,
Austria.
Holický, M, Diamantidis, D & Sýkora, M 2018.
Reliability levels related to different reference
periods and consequence classes. Beton- und
Stahlbetonbau, 113(1): 22–26.
Holický, M, Viljoen, C & Retief, J V 2019. Assessment
of existing structures. In Zingoni, A (Ed). Advances
in Engineering Materials, Structures and Systems:
Innovations, Mechanics and Applications.
Abingdon, UK: Taylor & Francis, pp 2109–2114.
ISO (International Organisation for Standardisation/
South African National Standard). 2016a. ISO
13822:2010/SANS 13822:2016. Bases for Design
of Structures. Assessment of Existing Structures.
Geneva, Switzerland: ISO.
ISO (International Organization for Standardisation/
South African National Standard). 2016b. ISO
2394:2015/SANS 2394:2016. General Principles on
Reliability for Structures. Geneva, Switzerland: ISO.
JCSS (Joint Committee on Structural Safety) 2001a.
Diamantidis D (Ed.) Probabilistic assessment of
existing structures. Joint Committee on Structural
Safety. SARL, France: RILEM Publications.
JCSS (Joint Committee on Structural Safety) 2001b.
Joint Committee on Structural Safety. Probabilistic
Model Code. http://www.jcss.byg.dtk.dk.
JRC (Joint Research Centre) 2015. Scientific and Policy
Report, No 1247. New European Technical Rules
for the Assessment and Retrofitting of Existing
Structures. Luxembourg: JRC.
Volume 63 Number 1 March 2021
Kemp, A R, Milford, R V & Laurie, J A P 1987.
Proposals for a comprehensive limit states
formulation for South African structural codes.
Civil Engineer in South Africa, 29(9): 351–360.
Lüchinger, P, Fischer, J, Chrysostomou, C et al 2015.
New European technical rules for the assessment
and retrofitting of existing structures. JRC Science
and Policy Report. Luxembourg: Publications Office
of the European Union.
Milford, R V 1985. Calculation of a load index for
ultimate limit state codes. Report 85/6. Pretoria:
CSIR National Building Research Institute.
Milford, R V 1986. A guide for the calibrating SABS
materials codes. Report 86/16. Pretoria: CSIR
National Building Research Institute.
Retief, J V & Dunaiski, P E (Eds) 2009. Background to
SANS 10160. Stellenbosch: SunMedia.
Retief, J V, Dunaiski, P E 2009. The Limit States Basis of
Design for SANS 10160-1. Stellenbosch: SunMedia.
Retief, J V, Dunaiski, P E & Holický, M 2009. Review
of Eurocode from the Perspective of the Revision of
SANS 10160. Stellenbosch: SunMedia.
Retief, J V, Viljoen, C & Holický, M 2019. Standardized
basis for assessment of existing structures. In
Zingoni, A (Ed) Advances in Engineering Materials,
Structures and Systems: Innovations, Mechanics
and Applications. Abingdon, UK: Taylor & Francis,
pp 2120–2125.
SANS (South African National Standard) 2018. SANS
10160-1:2010/2018. Basis of Structural Design and
Actions for Buildings and Industrial Structures.
Part 1: Basis of Structural Design. Pretoria: SABS
Standards Division.
SANS (South African National Standard) 2000. SANS
10100-1: 2000. The Structural Use of Concrete,
Part 1: Design. Pretoria: SABS Standards Division.
Steenbergen, R, Sýkora, M, Diamantidis, D, Holický, M
& Vrouwenvelder, T 2015. Economic and human
safety reliability levels for existing structures.
Structural Concrete, 16(3): 323–332.
Sýkora, M, Diamantidis, D, Holický, M & Jung, K 2017.
Target reliability for existing structures considering
economic and societal aspects. Structure and
Infrastructure Engineering, 13(1): 181–194.
Tanner, P, Lara, C & Prieto, M 2011. Semi-probabilistic
models for the assessment of existing concrete
structures. Proceedings, 11th International
Conference on Applications of Statistics and
Probability in Civil Engineering, 1–4 August 2011,
Zurich, Switzerland.
Vrouwenvelder, T & Scholten, N 2010. Assessment
criteria for existing structures. Structural
Engineering International, 20(1): 62–65.
Journal of the South African Institution of Civil Engineering
 TECHNICAL PAPER
Measurement of
probabilistic ballast particle
Journal of the South African
Institution of Civil Engineering

dynamics using Kli-Pi

ISSN 1021-2019
Vol 63 No 1, March 2021, Pages 11–23, Paper 966

ANDRÉ BROEKMAN is full-time PhD student

A Broekman, P J Gräbe and researcher in the Civil Engineering
Department at the University of Pretoria,
South Africa. In 2018 he obtained his MEng
degree in Transportation Engineering
At first glance the operational performance of ballast appears trivial in its simplicity. However, (cum laude) under supervision of Prof Hannes
Gräbe, having graduated with a BEng Hons
various mechanisms affect the performance of the ballast both on a macroscopic scale and
Transportation degree (cum laude) in 2017
discrete (mesoscale) particle level. The importance of experimental studies to establish the and a BEng Civil Engineering degree (cum laude) in 2016 at the University of
influence of the granular fabric has been highlighted repeatedly by other researchers. This paper Pretoria. He is also responsible for lecturing undergraduate students in the
describes a method by which quantitative metrics and statistics can describe the probabilistic field of instrumentation design and assists with the development of new
instrumentation hardware, software and communications technologies
response of railway ballast. The measurements were obtained with the installation of a set of
(Civiltronics) for undergraduate and post-graduate research projects.
customised wireless inertial measurement unit (IMU), referred to as Kli-Pi, in the granular layers
Contact details:
of a heavy-haul railway line located in South Africa. The results indicate a complex interaction of
Department of Civil Engineering
displacement and rotation, in all three spatial dimensions. The high-frequency measurements Engineering 1, Room 13-10
provided approximations of the particle’s kinetic and potential energy (mechanical work) in University of Pretoria
addition to the indirect quantification of changes to the granular fabric. Finally, the descriptive Pretoria 0002
South Africa
statistics of the mechanical work provided an indirect measure of the confinement and T: +27 12 420 4723
coordination number of the particle, together with supporting evidence of the underlying E: andre@broekmail.com
probabilistic, instead of the expected deterministic response. These results strongly agree with the
findings of existing literature that has, to date, been confined to theoretical study. PROF HANNES GRÄBE (Pr Eng, FSAICE) is a
civil engineer with experience in the fields
of track technology, geotechnology,
advanced laboratory testing, field
INTRODUCTION availability of track deterioration models investigations, maintenance models and
The performance of the track structure as follows: “It is clear from looking at the numerical analysis of track structures. He is
is highly dependent on the operational existing track geometry models that there Associate Professor: Transnet Freight Rail
Chair in Railway Engineering at the
performance and interaction of ballast remains a need for better understanding of University of Pretoria, where he lectures under- and post-graduate
particles. The dynamic response can be the degradation process to be established, courses in railway engineering. He is also responsible for railway research,
investigated from either a macroscopic which can support the development of accu- as well as for continuing professional development in the form of short
courses presented to industry. He holds a PhD degree from the University
scale (considering the track structure as a rate models, based on historical data. This
of Southampton (UK).
whole) or the mesoscale (discrete particle model needs to be capable of including the
Contact details:
behaviour). Attempting to instrument an effects of maintenance models.”
Department of Civil Engineering
individual ballast particle encounters great This paper investigates the application Engineering 1, Room 13-7
difficulty, in part due to the morphology of of a “smart ballast” device – referred to as University of Pretoria
the material and technological limitations. Kli-Pi (derived from the Afrikaans word Pretoria 0002
South Africa
Analytical solutions describing the track for a small stone) – in the investigation of
T: +27 12 420 4723
structure response subjected to complex mesoscale ballast dynamics. The representa- E: hannes.grabe@up.ac.za
dynamic loading requires convoluted tive dynamic response of a track structure
solutions. Rucker (1982) concluded that was recorded for both a laboratory and field
attempts to develop three-dimensional experiment – only the results obtained
theoretical models for the vibrations of from the field test are considered in this
ballast is an extremely difficult, if not paper. These measurements included all six
impossible task. Representative, continuous degrees of freedom (DoF) – three for trans-
empirical measurements are required to lation and three for rotation. Descriptive
understand the fundamental behaviour statistics and energy principles were imple-
of ballast in the process of promoting mented to quantify the performance of the
cost-effective life cycle costs of large-scale ballast response. This investigation provided
infrastructure projects. The importance of insights into the geometry-dependent
studying granular particle interactions and frequencies, lateral and longitudinal forces,
contacts under the influence of quasi-static indirect quantification of the “granular
loading conditions – also referred to as the fabric”, longitudinal principal stress rota-
“granular fabric” – is well known (Cowin & tions (PSR), particle coordination number
Keywords: probabilistic ballast particle dynamics, smart ballast
Satake 1979). Audley and Andrews (2013) (CN) and evidence of the probabilistic (non- instrumentation, Kli-Pi, granular fabric quantification,
concluded their discussion regarding the deterministic) nature of ballast particles. mesoscale ballast dynamics, impact loading

Broekman A, Gräbe PJ. Measurement of probabilistic ballast particle dynamics using Kli-Pi.
J. S. Afr. Inst. Civ. Eng. 2021:63(1), Art. #966,13 pages. http://dx.doi.org/10.17159/2309-8775/2021/v63n1a2 11
LITERATURE
The background literature discusses the rel-
evance of ballast in the track structure and
emerging sensor technologies that enable
quantitative measurements of the ballast
particles. Different modelling approaches for
describing inter-particle contact and behav-
iour – both deterministic and probabilistic
perspectives – are reviewed, and conclude
with recent examples of “smart ballast” to
directly quantify the dynamic ballast prop-
erties subject to quasi-static loading.
Track structure
Most of track maintenance pertains to the
correction of deteriorating track geometry.
The action of repeated cyclic loading
degrades the ballast through inter-particle
attrition, fouling and weathering of the par-
ticles. Although some of the settlement can
be attributed to the underlying earthwork
(subgrade and subballast), most of the track
settlement is caused by the rearrangement
and movement of the unbound ballast parti-
cles. Low-frequency components associated
with the moving load are characterised by
geometry-dependent features of the train,
such as inter-axle, inter-bogie and inter-
wagon dimensions. Intermediate frequencies
are characteristic features of the granular
layers, with higher frequencies associated
with track components and rail defects. For
the investigation of ballast, the frequencies
of interest are below 250 Hz (Kouroussis
et al 2015).
Emerging sensor technologies
Different instrumentation technologies have
been developed, adapted and improved over
time to measure a wide assortment of track
components and associated phenomena.
Although the functional importance of
ballast is well understood and extensively
documented, few studies have attempted to
directly instrument ballast particles. Some
notable examples are that of Bennett et al
(1999) with the development of a cylindri-
cally shaped device to measure strain of
asphalt surfacings, and Nejikovsky and
Keller (2000) with the on-board measure-
ment of lateral acceleration. Straser and
Kiremidjian (1998) developed wireless
monitoring systems for civil structures. Any
attempts to instrument a ballast particle
directly are met with difficulty, owing to
the irregular morphology and freedom of
movement about multiple degrees of free-
dom (DoF). Only the particles located at the
surface, which is not of interest, are suited
for visual study. Recently, Lamas-Lopez et al
12
(2017) investigated the potential replace-
ment of t­ raditional LVDTs (linear variable
differential transformers) altogether using
accelerometers (Bowness et al 2006; Graizer
2010; Kempe 2011). The peak-to-peak
displacement accuracy was reported as 20%
over a wide range of train velocities.
The introduction of inertial navigation
systems and units (IMU), together with the
advancement of micro-electro-mechanical
system (MEMS) manufacturing techniques,
have led to the development of high-
performance sensors in miniaturised form
factors. Woodman (2007) defines inertial
navigation as: “… a self-contained navigation
technique in which measurements provided
by accelerometers and gyroscopes are used to
track the position and orientation of an object
relative to a known starting point, orientation
and velocity.” A rigid IMU system provides
the ability to describe the body frame (of
the sensor), relative to the global reference
frame, through the measurement of three
orthogonal accelerometers and three ortho­
gonal gyroscopes. The linear accelerations
and angular velocities are measured by the
accelerometers and gyroscopes respectively.
MEMS manufacturing technologies refer to
the construction of miniaturised IMUs using
traditional semiconductor and integrated cir-
cuit (IC) fabrication techniques (Nathanson &
Wickstrom 1965). The typical construction
and operation of such a device are discussed
in detail by Lemkin and Boser (1999).
Deterministic mechanics
of granular materials
Traditionally, investigations of granular
materials, particularly numerical models,
laboratory and field investigations, have
been from a deterministic point of view.
The combined probabilistic response of the
particle matrix manifests as macroscopic or
deterministic effects and properties such as
stress, permanent strain and confinement.
The transfer of load from the sleeper
to the ballast produces the same effects as
with triaxial compression (Li et al 2015).
The middle of the ballast layer experiences
a reduction in vertical stress compared to
the ballast directly beneath the sleeper, but
with higher levels of confinement. The lowest
layer of ballast, near the subballast interface,
develops tensile strain incrementally as the
number of load cycles increases. These par-
ticles tend to spread and rearrange, resulting
in the development of voids. The substantial
influence of particle breakage on the
permanent strain was highlighted by Lobo-
Guerrero and Vallejo (2006). Cundall and
Volume 63 Number 1 March 2021
Strack (1979) explored the effect of particle
rearrangement on the distribution of princi-
pal stresses using numerical models. Typical
ballast maintenance activities further disturb
the particle arrangement and interlocking
where the lateral and longitudinal resistance
are reduced by up to 60% and 40% respec-
tively (Tutumluer et al 2006). Furthermore,
the long-term rate of permanent deformation
(Lekarp et al 2000) and cumulative perma-
nent deformation (Gräbe & Clayton 2009) is
highly dependent on the development of PSR.
The most unfavourable stress combination
occurs approximately at the subballast inter-
face depending on the Young’s modulus of
the subgrade and ballast. Gräbe and Clayton
(2009) describe the PSR as a combination of
three distinct mechanisms:
1. Longitudinal PSR is positioned between
two adjacent sleepers for shallow depths
and single-axle combinations.
2. Transverse PSR is present at the discon-
tinuous sleeper ends.
3. The depth of influence of PSR correlates
with the axle count per bogie.
Micromechanics of granular
materials
The deterministic or continuum approach
to analysing soils and granular media is
generally employed to model the response
to loading using theories of elasticity and
plasticity. For the investigation of discrete
particle media, however, analytical solutions
and new instrumentation technologies are
required if the micromechanics of mesoscale
behaviour are to be better understood. The
two most popular numerical methods are
finite element modelling (FEM) and discrete
element modelling (DEM). The complexity
of obtaining realistic results stem from the
significant number of particles present and
the small time-step required for convergence.
To complement the theoretical investigations,
the use of miniaturised, MEMS-based sensor
technologies has recently been explored. Few
comprehensive, constitutive relationships
exist that accurately describe the complexi-
ties involved with a discrete medium, thus
the wide implementation of FEM and DEM.
Cundall and Strack (1979) pioneered the
theoretical establishment of DEM modelling.
Discrete modelling has the advantage of
simulating the relatively large displacements,
rotations and contacts occurring within the
matrix. A comprehensive, two-dimensional
DEM analysis by Zhang et al (2016) inves-
tigated the dynamic ballast behaviour for
high-speed railway applications. The largest
stresses and accelerations were concentrated
Journal of the South African Institution of Civil Engineering
within the first 200 mm of ballast where
peak frequencies were observed in the region
below 100 Hz and in the 150–300 Hz range.
A dimensionless scalar value of β relates the
ratio of the principal stress of a particle to
the average of the three principal stresses.
When the displacements of these particles
are graphed for a moving load, a circular
pattern of movement is produced. The
spectral frequency densities with the greatest
amplitude were observed to be a function
of the inter-axle and inter-bogie spacing of
the train. Other studies improved upon this
research with the implementation of more
representative non-convex particle geometry
(Wilke et al 2016) generated from photo-
grammetric laser scanning (Indraratna et al
2011; Mvelase et al 2012) of the particles.
Some authors have suggested that the
driving force behind the rearrangement,
orientation and dynamic response of
granular media is the manifestation of the
maximisation of the system’s entropy (Cowin
& Satake 1979). This statistical foundation
suggests a possible amalgamation between
the measured deterministic response and
proven probabilistic characteristics exhibited.
Rothenburg and Kruyt (2009) concluded
that the best descriptor of the inter-particle
contact forces, angles and displacements
is with the use of statistics, not simplified
laws of mechanics. The observed particle
displacements, for identical contact angles
(the corresponding conditional probability),
are merely the outcome or realisation of
some independent random variable. The
shape of the probability distribution func-
tion (PDF) was derived using techniques to
maximise the available entropy (Kruyt &
Rothenburg 2002). Depending on the average
coordination number (CN) of the particles,
the PDF is either Gaussian-shaped for dense
systems (5.5 < CN < 6) or exponentially
shaped (4 < CN < 5). Makse et al (2000)
observed similar results to that of Kruyt and
Rothenburg (2002). Kitamura (1980) sum-
marises his findings as follows: “Furthermore,
it is shown that the discontinuous motions
largely contribute to the deformation process
of particulate material, and the method to
quantitatively estimate the discontinuous
motions is proposed based on the probabilistic
considerations for the motions of particles.”
Smart ballast
The fusion of recent research applications
involving smart sensor technologies and
ballast particles is collectively referred to as
“smart ballast” instrumentation. Zhai et al
(2004) provides one of the earliest examples
Journal of the South African Institution of Civil Engineering Volume 63 Number 1 March 2021
of instrumenting field ballast with a uniaxial
MEMS accelerometer, in combination with a
theoretical model. Good agreement between
the results was found. The resonant fre-
quency of the ballast accelerations was found
to be between 80 Hz and 100 Hz. Aikawa
(2009) supplemented this method with an
additional, secondary accelerometer to derive
angular velocities. Rotations exhibited identi-
cal frequencies to that of the accelerations
with more rapid attenuation. Only vertical
accelerations were measured. Li et al (2015)
reported a similar range of resonance fre-
quencies in-between 50 Hz and 210 Hz. Milne
et al (2016a) utilised several wireless, minia-
turised smart ballast particles to investigate
aerodynamic effects on the granular media
for train velocities exceeding 400 km/h.
The most recent and noteworthy example
of smart ballast is the development of
“SmartRock” by Liu et al (2016a). The wire-
less internal unit containing the electronics
is surrounded by a 3D printed exoskeleton,
measuring only 60 × 60 × 60 mm in size.
The unit transmits the IMU measurements
wirelessly using a Bluetooth interface. The
maximum sampling frequency is reported
as 64 Hz which can resolve the peak-to-peak
translational and rotational accelerations.
Short-term testing of 500 loading cycles illus-
trates the complexity in the rearrangement of
the SmartRock in the surrounding granular
matrix. A decrease in lateral stiffness was
observed at the edge of the sleeper resulting
from the decrease in resistance to movement.
Additional research was carried out to move
the SmartRock to different positions beneath
the sleeper (Liu et al 2016b) together with the
influence of geogrids installed within the bal-
last (Liu et al 2016c). The SmartRock proved
vital to measure the mesoscale effects associ-
ated with different geometric and material
variables. The main conclusions reached
were the increased amplitude of rotation by
positioning the SmartRock at the edge of
the sleeper, the significant reduction in the
movement with an increase in depth, and the
complex interaction between the measured
translation and rotation.
KLI-PI DEVELOPMENT
A detailed design methodology outlining
the desired functionality and performance
characteristics of an improved smart ballast
prototype, to meet the required objectives of
the study, was established. The methodology
incorporated the design, construction and
calibration of the instrument. The smart
ballast prototype, named “Kli-Pi”, was
successfully developed (Figure 1) and cali-
brated for this research project (Broekman
& Gräbe 2018). The calibration procedure
provided comparable instrumentation
performance compared to that of Milne et
al (2016b). The name Kli-Pi is derived from
the Afrikaans word for small rock to convey
the miniaturised rock or ballast-shaped ana-
logue. The key performance characteristics
of Kli-Pi can be summarised as follows:
QQ High-performance, Linux-based
microprocessor and MEMS IMU
(H3LIS221DL Motion Sensor).
QQ Sampling rate of 3 000 measurements
per second, divided evenly among
the 6 DoF (tri-axis accelerometer and
­tri-gyroscope). The data for each axis of
each sensor is thus recorded at 500 Hz.
QQ Programmable full scale of ±16 G and
±2 000 degrees per second (DPS) for the
accelerometer and gyroscope respectively.
QQ High-strength, non-convex, 3D printed
PLA (Polylactic Acid) exoskeleton for
ample environmental protection.
QQ Sufficient battery capacity to provide a
two-month standby time or ten hours of
continuous data acquisition.
QQ High-bandwidth, in-situ wireless com-
munication using a Wi-Fi connection
for the transfer of data (post-recording)
over a local area network (LAN).
The 500 Hz sampling rate, or 2 ms between
successive readings, provides the abil-
ity to accurately integrate the data using
numerical methods. A second-order
Lagrange interpolation polynomial was
used to approximate the discrete data
points and is referred to as super-sampling.
Following the super-sampling operation,
Simpson’s method is employed to numeri-
cally integrate the array of data. The linear
acceleration and rotational velocity were
integrated using this method to obtain the
displacements and rotations for the three
orthogonal axes of measurement. Due to the
influence of electromagnetic interference,
only the gravitational vector was employed
in the alignment of the body vector’s z-axis
with that of the global frame. The installa-
tion method ensured that the lateral axes of
the body frame were aligned with that of the
global frame. For both the installation and
experimentation procedure, the following
coordinate system was used:
QQ Longitudinal | parallel to the direction
of the rails | IMU x-axis or roll (φ)
QQ Lateral | parallel to the long axis of the
sleepers | IMU y-axis or pitch (θ)
QQ Vertical | parallel to the gravitational
vector | IMU z-axis or yaw (ψ)
13
Figure 1 Kli-Pi 3D printed shell / exoskeleton and the internal IMU box containing all the electronics
EXPERIMENTAL PROCEDURE
A full-scale field trial was conducted to
investigate the representative dynamic
response of the ballast. The aim of the
experiment was to replicate the typical
characteristics and operating behaviour of
field conditions for heavy-haul conditions.
Figure 2 illustrates the instrumentation
configuration together with the depth of
the instrumentation. The experiment was
carried out during December 2017 near the
Komati Power Station / Broodsnyersplaas
(26°04’32.9”S 29°30’17.8”E). This primary
rail artery is used for exporting coal from
the surrounding mines in the Mpumalanga
Province of South Africa. Gräbe and
Clayton (2009) summarised the track
specifications and dimensions. The track
consists of equidistantly spaced (650 mm)
concrete (PY) sleepers using a standard
Cape gauge of 1 067 mm. The ballast
depth is 300 mm with the subballast and
subgrade material extending to a depth
of 900 mm. This rail corridor started
service in 1976. The direction of travel
Z
AP Y
K4
X
250
PY sleeper
50
K3
Ballast
150
K2
150
Subballast
interface K1
Figure 2 F ield test instrumentation
configuration
14
coincided with an increase in the gradient
and a decrease in train velocity as the train
passed the static reference measurement
point. The granular layers were saturated
following short bouts of rain during the
preceding week. Additionally, during
retrieval of the instruments, the water table
was positioned approximately 100 mm
above the subballast interface.
The section was instrumented with a
total of four Kli-Pis. For ease of installation
and retrieval, the Kli-Pis were installed
directly beneath one of the sleepers in the
shoulder ballast section (Figure 3). This
position also minimises the disturbance
to the ballast matrix. Furthermore, this
position is synonymous with the expected
location where significant rotation (Liu et al
2016b) and PSR (Gräbe and Clayton 2009)
would be measured in addition to the influ-
ence of overlapping stress frusta from the
dual-wheel axle configuration. The ballast
stones were carefully placed back in position
by hand after the installation process to
ensure adequate interlocking of the particles
Figure 3 I nstallation of Kli-Pi 1 and 3 in the subballast interface and beneath the sleeper respectively
Volume 63 Number 1 March 2021
surrounding the instruments. During sub-
sequent retrieval of the instruments, it was
noted that the orientation remained unal-
tered. The Wi-Fi access point (denoted AP)
used for communicating with all the Kli-Pis,
was positioned directly above the vertically
aligned column of instruments to ensure
a reliable connection. The user interface
was situated a safe distance away from the
track using an Ethernet interface. The data
presented was obtained from the passage
of six class 10E locomotives (21 ton/axle)
and 100 loaded wagons (estimated 20 ton/
axle), representing a period of approximately
100 seconds.
The software language of choice was
Python for its versatile set of available
libraries and wide implementation in the
research community. Python was used for
both the calibration of the IMU and the
analysis of the large data sets to extract
meaningful data as efficiently as possible.
The analysis workflow consisted of filtering
methods (Butterworth filtering), numerical
approximation and integration methods,
Z
Y
X
Journal of the South African Institution of Civil Engineering
frequency domain transformations and peak
identification. Similar to Lamas-Lopez et al
(2017) and Milne et al (2016b) a Butterworth
filter of order 3 with a bandpass of 4-200 Hz
0.4
was selected for the acceleration data. To
remove the bias present in the rotational
velocity data obtained from the gyroscope,
0.2

Displacement (mm)
an average measurement over a 3-second
period (with no trains passing) is subtracted
(per axis) from the entire dataset. Any other
calculations derived from these datasets, 0

for example the displacement and energy

metrics and peak-to-peak statistics, are thus
automatically filtered. Over a two-minute –0.2
period, accounting for all six degrees of free-
dom, 360 000 data samples were generated
for each Kli-Pi, or 1.44 million for a single –0.4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3
train passage. For any given degree-of-free-
dom, the 2 000 largest peak-to-peak samples 29 30 31 32 33 34
were extracted for use in the analysis phase. Time (sec)
Using Python, graphs illustrating the data in K4 – 0 mm K3 – 250 mm K2 – 400 mm K1 – 550 mm
both the time and frequency domain were
implemented in quantifying the dominant Figure 4 Time domain deflection: z-axis of all Kli-Pis (with impact load)
modes of rotation and translation, together
with observations of impact loading, rear-
rangement of particles in the ballast matrix
0.04
and ballast resonance.

0.03

DISCUSSION OF THE RESULTS

0.02
With the introduction of Kli-Pi, the experi-
Attitude (degrees)

ment performed aims to augment existing

literature and provide new insights into
the mesoscale behaviour of ballast under
the influence of quasi-static loading. These
results are divided into discussions sur-
rounding the typical displacements and
rotations for both the time and frequency
domains, mechanical work and energy
methods, concluding with the statistical
descriptors for the interpretation of these
metrics. The focus of the discussion is
based on the results from the field test.
Time domain features
Figure 4 illustrates the displacement in the
vertical or z-axis for all the Kli-Pis with the
periodic 4-axle configuration of the wagons
clearly visible; each wagon is denoted by the
bracket with each of the axles numbered.
Impact loading occurs at approximately
31.5 s, whereby the peak-to-peak displace-
ment amplitudes are significantly larger
(for all the Kli-Pis) compared to the pre-
ceding or successive loading patterns. The
largest typical peak-to-peak displacement
is associated with Kli-Pi 4 attached to the
sleeper, measuring approximately 350 μm
(peak-to-peak), reducing with depth. The
peak-to-peak displacements measured
0.01
0
–0.01
–0.02
–0.03 1 2 3 4 1 2 3 4 1 2
29 30
K4 – 0 mm K3 – 250 mm
Figure 5 Time domain rotation: x-axis of all Kli-Pis
among all the instruments show significant
variance over time. This is also true for the
longitudinal (x) and lateral (y) axes, though
the magnitude thereof is substantially
smaller. Even though the displacement
reduces on average with depth, an anomaly
occurs for Kli-Pi 3 and Kli-Pi 2 which are
positioned beneath the sleeper and in the
middle of the ballast layer respectively.
The average vertical displacement (z-axis)
was measured as 109 μm and 104 μm for
Kli-Pi 3 and Kli-Pi 2, respectively. This
3 4 1 2 3 4 1 2 3 4 1 2 3
31 32 33 34
Time (sec)
K2 – 400 mm K1 – 550 mm
net-zero displacement can be explained by
the overlapping of stress frusta of the wheel
loads occurring in-between the positions
of Kli-Pi 2 and Kli-Pi 3. The discontinuous
nature of the particles accommodates this
strain discontinuity, where the particles
beneath the sleeper experience lower levels
of confinement. Reorientation or partial
suspension of the particles is possible,
effectively inducing heaving of the ballast.
The measurement of mesoscale behaviour
limits any simplified approach to quantify

Journal of the South African Institution of Civil Engineering Volume 63 Number 1 March 2021 15
 insights into the geometry of the train,
sleeper-ballast resonance and horizontal
20.0 1.20 × 10 –8 forces. For the vertical deflection (z-axis) of
Kli-Pi 3, positioned beneath the sleeper, the
17.5

Power spectrum density (m2/Hz)

1.00 × 10 –8 distinction between the locomotives and
15.0 wagons is clear (Figure 6). For the first few
Frequency (Hz)

8.00 × 10 –9 seconds, the movement and reorientation of

12.5
the Kli-Pi are represented as a broad range
10.0 6.00 × 10 –9 of frequencies with varying intensities.
Thereafter, each discrete frequency can
7.5 be linked to the inter-bogie, inter-axle and
4.00 × 10 –9
5.0 inter-wagon frequencies or a more complex
combination thereof, as explored in detail
2.00 × 10 –9
2.5 by Milne et al (2017) and Ju et al (2009).
During the passage of the locomotives, a
0
20 40 60 80 100 distinct band of frequencies in the region
Time (sec) of 125–150 Hz was observed and correlates
with the ballast-sleeper resonance noted
Figure 6 Frequency domain deflection: z-axis of Kli-Pi 3 by Zhang et al (2016). The bogie passage
and car bogie bounce are particularly
pronounced between the 5 Hz and 10 Hz
band of frequencies (Kouroussis et al 2015).
20.0 1.20 × 10 –8 The impact load noted before (Figure 4) is
illustrated by the broad range of frequencies
17.5

Power spectrum density (m2/Hz)

1.00 × 10 –8 at the 32 s mark. The largest proportion of
15.0 the energy remained concentrated along the
Frequency (Hz)

8.00 × 10 –9 vertical direction (z-axis), parallel to the axis

12.5
10.0
7.5
5.0
2.5
0
20 40 60
Time (sec)
Figure 7 Frequency domain deflection: x-axis of Kli-Pi 3
the stress and strain of the material. This
provides supportive evidence for the mech-
anism driving the periodic changes on the
ballast matrix as noted by Cundall & Strack
(1979). For the comparative laboratory test,
the same phenomenon was not observed.
Figure 5 illustrates the rotation about
the longitudinal or x-axis for all the Kli-Pis.
The rotational response exhibits more
uniformity than that of the displacements,
although the magnitude is staggered about
depth. As expected (Gräbe & Clayton
2009), the longitudinal PSR is visible when
considering the relatively large peak-to-peak
rotation of Kli-Pi 1 situated within the
subballast-ballast interface (0.02 degrees).
The influence of the sleeper’s transverse PSR
in proximity of Kli-Pi 3 is also illustrated,
with the magnitude even larger than that of
Kli-Pi 1 with a peak-to-peak amplitude of
6.00 × 10 –9
4.00 × 10 –9
2.00 × 10 –9
80 100
0.05 degrees. Kli-Pi 2 and Kli-Pi 4 positioned
within the middle of the ballast layer and
on the surface of the sleeper respectively,
hardly exhibit any rotational response. For
the laboratory tests, significant variations
occurred over the duration of the test where
both the magnitude and preferred axis of
rotation changed throughout the test, simi-
lar to the observations of Liu et al (2016a)
with SmartRock. For the field test, however,
the peak-to-peak rotations increased rapidly
during the initial reorientation after the
installation, stabilising after approximately
20 seconds.
Frequency domain features
Fast Fourier Transformations (FFTs)
highlight the dominant frequencies associ-
ated with the passage of the locomotives
and wagons. This information provides
of loading. Near the end of the train passage,
an increase in the density of the spectra is
observed, likely as a result of an increas-
ing level of compaction of the instrument
within the surrounding ballast matrix.
With the passage of the locomo-
tives, the amplitude of the longitudinal
displacements (x-axis) during this narrow
time frame exceeded that of the vertical
displacements (z-axis) by more than 50%
(Figure 7). This effect can be linked to the
transfer of tractive forces from the locomo-
tive onto the track, inducing a short-lived,
oval-shaped translation pattern (Zhang
et al 2016). Barely any longitudinal forces
(x-axis) are present during the passage of
the wagons. The transfer of longitudinal
(x-axis) and lateral forces (y-axis) dimin-
ishes with depth, although the reduced
influence of these forces was recorded by
Kli-Pi 1. These forces, h­ owever, remain
significant, which is usually absent form
uniaxial cyclic loading typically employed in
laboratory experiments.
The FFT of the rotational response about
the x-axis of Kli-Pi 3 (Figure 8) appears simi-
lar to that of the deflection with the density
profile strongly concentrated around the
discrete spectra associated with the train
geometry. This is indicative of a response
that is strongly linked to the macroscopic
response of the substructure instead of the
mesoscale particle behaviour as with the

16 Volume 63 Number 1 March 2021 Journal of the South African Institution of Civil Engineering
displacements. The negative gradient of the
spectra with the progression of time for all
the FFT-based figures is a direct result of the 20.0 5.00 × 10 –4
decreasing train velocity over time, owing to

Power spectral density (deg2/Hz)

17.5
the positive track gradient.
4.00 × 10 –4
15.0
Mechanical work and

Frequency (Hz)
12.5
energy methods 3.00 × 10 –4
Considering that the motion of the Kli-Pi 10.0
was quantified about all three spatial
7.5 2.00 × 10 –4
dimensions, application of kinetic and
potential energy principles is a cogent 5.0
analysis method. The assumption can be 1.00 × 10 –4
postulated whereby the position of the 2.5
Kli-Pi, over a relatively long period of time,
0
remains unchanged. Even though the 20 40 60 80 100
filtered data represents a relative position Time (sec)
(zero-centred data), the assumption is that
a constant position before and after the Figure 8 Frequency domain rotation: x-axis of Kli-Pi 3
passage of the train represents an absolute
datum. The total work associated with the
Kli-Pi consists of both kinetic (UK) and 1.00
potential energy (UP) components. The
angular moment of the Kli-Pi is ignored 0.75
owing to the difficulty associated with the
calculation of the moment of inertia and the 0.50
small rotation angles. The work generated
by the potential or gravitational field is 0.25

equal to the negative change of the potential

Energy (mJ)

0
energy of a rigid body (Stewart 2012). Also,
following from Newton’s ­second law of
–0.25
motion, the mechanical work exerted on a
rigid body can be shown to equal the change
–0.50
in the kinetic energy. If the total energy is
to be conserved for a Kli-Pi moving along
–0.75
a path C, given by a function r (t), through
a conservative force field F , Equation 1 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3
–1.00
must hold: 64.0 64.5 65.0 65.5 66.0 66.5 67.0 67.5 68.0

∫c ⎫⎪⎭ 2 m∆V 2 – m∆hz g⎫⎪⎭dr = 0

Time (sec)
1 K4 – 0 mm K3 – 250 mm K2 – 400 mm K1 – 550 mm
(1)

Figure 9 Energy response of all Kli-Pis

Where:
∆V = change in body velocity vector
∆hz = change in body elevation with
respect to the potential field
m = mass of the body (400 g for every
Kli-Pi)
g = gravitational constant
The total energy associated with any one
of the Kli-Pis can be expressed as a func-
tion of time. The total energy is illustrated
over a short period of time (Figure 9) in
addition to the isolated kinetic (Figure 10)
and potential (Figure 11) energy over the
entire duration of the test. Note that the
kinetic energy always attains a positive
magnitude, whereas the potential energy
is either positive or negative depending on
the position relative to the datum. Figure 9
illustrates a consistent pattern whereby the
peak-to-peak magnitude of the total energy
decreases with depth for most of the passing
axles. The mean peak-to-peak magnitude
for Kli-Pi 4 positioned on the sleeper, over
the entire duration of the test, is 0.639 mJ,
reducing to 0.287 mJ for Kli-Pi 1 in the
subballast (Table 3 on page 22). The total
energy associated with Kli-Pi 2 and Kli-Pi 3
is nearly identical owing to the overlapping
stress frusta. For the first few load cycles
(Figure 10), the kinetic energy highlights
the initial reorientation and movement of
the installed instruments in combination
with the significant lateral and longitudinal
locomotive traction forces adding to the
overall energy composition. Thereafter,
the kinetic energy constitutes only a small
portion of the total energy, except for the
impact load near 31.5 s and another smaller
impact load closer to the end of the test
at 99 s. This explicit distinction between
the kinetic and total energy provides a
clear illustration of the presence of impact
loads. The potential energy (Figure 11)
contributes the largest proportion of the
total energy. Like the kinetic energy graph,
the increased axle mass of the locomotives
is represented by larger amplitudes, followed
shortly thereafter by the repetitive wagon
axles with a small degree of variation. The
standard deviation of the total energy (peak-
to-peak) reduces from 0.633 mJ for Kli-Pi 4

Journal of the South African Institution of Civil Engineering Volume 63 Number 1 March 2021 17
 the sleeper (Kli-Pi 4). A large component
2.00 of the displacement is associated with the
longitudinal direction (x-axis), caused
1.75
predominantly by tractive wheel-rail forces.
1.50 This longitudinal component, together
with a slightly smaller lateral component,
Kinetic energy (mJ)

1.25 reduces with depth as the force and corre-

sponding displacement dissipates through
1.00
the ballast matrix. Similarly, the vertical
peak-to-peak displacement reduces sub-
0.75
stantially with depth as the stress frusta is
0.50 spread over a larger contact area. As noted,
the displacements associated with Kli-Pi 2
0.25 and Kli-Pi 3 are nearly equal, owing to
the overlapping of stress frusta between
0
0 20 40 60 80 100 these two locations. Considering that the
Time (sec) mean displacement is associated with the
K4 – 0 mm K3 – 250 mm K2 – 400 mm K1 – 550 mm expected outcome of the probability dis-
tribution, isolated events will occur where
Figure 10 Kinetic energy of all Kli-Pis positive or tensile strain is observed at the
interface that can lead to a rearrangement
of the otherwise stable ballast matrix.
6 Figure 13 illustrates the significant influ-
ence of uncontrolled variables, such as the
4 amplitude of the applied load, associated
with the field test. It is intriguing that for
all the axes the CoV decreases with depth,
Potential energy (mJ)

2
0
–2
–4
–6
0 20
K4 – 0 mm
Figure 11 Potential energy of all Kli-Pis
to 0.219 mJ for Kli-Pi 1. The significance of
the statistical characteristics of this metric
follows in the next section.
Statistical descriptors
Reviewing both the measurements and
literature, the underlying probabilistic
nature of the particle response is clear.
Descriptive statistics is employed to
quantitatively compare these results. For
each dataset, a total of 2 000 of the largest
peak-to-peak amplitudes was extracted
to obtain a representative sample for each
DoF. This number was obtained through
an iterative process until a representative
distribution of peak-to-peak amplitudes
had been extracted. This number is larger
than the number of passing axles due
40 60 80
Time (sec)
K3 – 250 mm K2 – 400 mm K1 – 550 mm
to the stochastic nature of the measure-
ments, especially that of the non-uniform
displacements, producing a relatively large
number of peak-to-peak amplitudes that
need to be accounted for. For each DoF,
the minimum, mean, maximum, standard
deviation, coefficient of variance (CoV)
and skewness statistics were calculated. Of
all the statistics analysed, the mechanical
work provided the most representative and
significant results compared to that of the
displacement and rotation statistics.
Displacement
Figure 12 illustrates the difference among
the three spatial dimensions. For any given
spatial direction, the largest mean peak-
to-peak displacement is associated with
likely due to the dissipation of energy and
the increasing number of potential pathways
in the ballast matrix through which the load
can transfer. The magnitude of the CoV
is similar among all three axes, with the
largest measurements associated with the
longitudinal (x-axis) displacement instead of
the expected vertical (z-axis) displacement
associated with the loading vector. The CoV
100 associated with the longitudinal direction
(x-axis) is pronounced only with the sleeper
and the sleeper-ballast interface.
Figure 14 illustrates positive skewness
for all the displacement statistics. The data
is significantly skewed in the longitudinal
(x-axis) and lateral (y-axis) directions
compared to the vertical (z-axis) direction,
primarily caused by the locomotive tractive
forces. These forces are largely dissipated
through the sleeper-ballast interface near
the position of Kli-Pi 3, where after the
skewness decreases noticeably for the
instruments positioned at greater depth in
the ballast.
Rotation
The relationship between the rotation
statistics and depth of installation is
less clear compared to the displacement
statistics. From the time-series data it was
concluded that the preferred axis of rotation
is primarily associated with the behaviour
of the sleeper. The largest mean rotation

18 Volume 63 Number 1 March 2021 Journal of the South African Institution of Civil Engineering
(Figure 15) is associated with Kli-Pi 3 which
is prone to transverse PSR resulting from 0.20
the sleeper above, followed by Kli-Pi 1. 0.18
The mean peak-to-peak amplitude for any
0.16
instrument or DoF remains relatively small,
with only a fraction of a degree rotation 0.14

Displacement (mm)
being measured. This finding was verified 0.12
during the analysis stage, whereby the 0.10
rotational velocity (prior to the integration
0.08
to calculate the attitude or orientations)
was rarely measured to be greater than one 0.06
degree per second. 0.04
The CoV statistics (Figure 16) illustrate
0.02
the rigidity of the sleeper where the CoV is
0
relatively small in the longitudinal (x-axis) Mean (X) Mean (Y) Mean (Z)
and lateral (y-axis) directions. Comparing Spatial direction
the two metrics, the difference in magni- Kli-Pi 4 (0 mm) Kli-Pi 3 (250 mm) Kli-Pi 2 (400 mm) Kli-Pi 1 (550 mm)
tude of the rotational CoV is significantly
smaller than that of the displacement CoV Figure 12 Displacement: mean for all axes
(Figure 13). A small reduction in the CoV
with depth is present when comparing Kli-
Pi 3 and Kli-Pi 4. 2.0
The skewness metric of the rotational
1.8
measurements is much smaller in magni-
tude compared to that of the displacements 1.6
Coefficient of Variance (CoV)

(Figure 17) – the average skewness statistic 1.4

decreases from a typical value of 5 for the 1.2
displacement skewness to less than 1 for
1.0
that of the rotation skewness, indicating
a reduced sensitivity for the rotation of 0.8
the particle to be affected by localised or 0.6
mesoscale effects and forces. The reduc-
0.4
tion in magnitude of the skewness statistic
with depth is only valid for the yaw (z-axis) 0.2

vector; the other DoF appears randomly 0

CoV (X) CoV (Y) CoV (Z)
distributed. The rotational response is
Spatial direction
more strongly influenced by the macro-
Kli-Pi 4 (0 mm) Kli-Pi 3 (250 mm) Kli-Pi 2 (400 mm) Kli-Pi 1 (550 mm)
scopic characteristics, which corresponds
to the frequency domain results (Figure 8);
the rotational response produced a more Figure 13 Displacement: coefficient of variance for all axes
concentrated spectral density around
geometry-dependent frequencies compared
10
to the that of the displacements.
9
Mechanical energy 8
Finally, when considering the mean energy
7
of the field test (Figure 18) the variability
6
introduced by impact loads, varying axle
Skewness

mass and larger particle translations pro- 5

duces a broad range of metrics – the maxi- 4
mum peak-to-peak energy (2.699 mJ) is sig-
3
nificantly larger in proportion to that of the
minimum (0.032 mJ) and mean (0.639 mJ) 2
energy statistics. This variability is also 1
apparent when considering the CoV metric 0
(Figure 19). For the standard deviation, Skewness (X) Skewness (Y) Skewness (Z)
CoV and skewness statistics, a negative cor- Spatial direction
relation exists with depth for the field test, Kli-Pi 4 (0 mm) Kli-Pi 3 (250 mm) Kli-Pi 2 (400 mm) Kli-Pi 1 (550 mm)
exemplifying the probabilistic response of
the ballast. These phenomena likely stem Figure 14 Displacement: skewness for all axes

Journal of the South African Institution of Civil Engineering Volume 63 Number 1 March 2021 19
 from both the dissipation of energy and
1.50 the increasing number of load paths with
1.35 depth as the load is distributed. From the
results presented by Kruyt and Rothenburg
1.20
(2002) it is known that with an increase in
1.05 the confinement stress, the shape of the
Rotation (arcmin)

0.90 parameter’s PDF shifts from an exponential

0.75 to Gaussian distribution. Correspondingly,
this should yield a smaller skewness sta-
0.60
tistic with depth, which is observed with
0.45 the field test results. This also indirectly
0.30 relates the coordination number to the
shape of the PDF. The most confined Kli-
0.15
Pis, positioned at depth in the granular
0
Mean (X) Mean (Y) Mean (Z)
layers, converge towards a skewness value
Spatial direction of 1.1, which is indicative of an exponential
Kli-Pi 4 (0 mm) Kli-Pi 3 (250 mm) Kli-Pi 2 (400 mm) Kli-Pi 1 (550 mm) distribution. The increase of the skewness
statistic corresponds with an increase in
Figure 15 Rotation: mean for all axes both the standard deviation and CoV.

1.0 CONCLUSIONS
The importance of studying and quantifying
0.9
the mechanical behaviour of granular media
0.8 was noted by Cowin and Satake (1979).
Coefficient of Variance (CoV)

0.7 Following the successful development and

0.6
deployment of Kli-Pi, new information
regarding the dynamic behaviour of ballast
0.5
particles within a representative environ-
0.4 ment has been presented. The fundamental
0.3 behaviour of ballast particles on a particle or
mesoscale was characterised by a probabilis-
0.2
tic or non-deterministic response.
0.1 Historically, the influence of lateral
0 and longitudinal forces, principal stress
CoV (X) CoV (Y) CoV (Z)
rotation, particle rearrangement and
Spatial direction
spectral frequencies has been confined
Kli-Pi 4 (0 mm) Kli-Pi 3 (250 mm) Kli-Pi 2 (400 mm) Kli-Pi 1 (550 mm)
to theoretical discussion and numerical
investigations. The results presented in this
Figure 16 Rotation: coefficient of variance for all axes paper, however, provide valuable insight
and supportive evidence for these existing
theories and formulations.
3.0
Few studies account for all the possible
2.8
2.6 DoFs apart from the direction of load appli-
2.4 cation which is always in the vertical direc-
Coefficient of Variance (CoV)

2.2 tion. For this field test, the transfer or large

2.0 longitudinal and lateral forces throughout
1.8
the substructure were measured as increased
1.6
displacements in these spatial directions.
1.4
1.2 During the passage of the locomotives, the
1.0 peak-to-peak magnitude of the displacement
0.8 parallel to the primary load direction was
0.6 measured to be smaller than that of the lon-
0.4
gitudinal direction where traction forces are
0.2
0
exerted. Complex, oval-shaped translational
Skewness (X) Skewness (Y) Skewness (Z) patterns were observed to extend throughout
Spatial direction the ballast structure, extending down to
Kli-Pi 4 (0 mm) Kli-Pi 3 (250 mm) Kli-Pi 2 (400 mm) Kli-Pi 1 (550 mm) the subballast interface, as predicted by
numerical models. Owing to the probabilistic
Figure 17 Rotation: skewness for all axes response of the ballast, in combination with

20 Volume 63 Number 1 March 2021 Journal of the South African Institution of Civil Engineering
the overlapping stress frusta of the wheel
loads, non-uniform strain distribution was 2.0
measured. The variation in the loading, 1.8
material and support conditions in the field
1.6
test significantly amplified the standard

Work (mJ) / Coefficient

deviation, CoV and skewness statistics of 1.4

the peak-to-peak displacement, rotation 1.2

and energy measurements. An increase in 1.0
particle confinement was measured as an
0.8
increase in energy density about specific fre-
quencies for both experiments, which agrees 0.6
with the findings of Liu et al (2016b) with the 0.4
implementation of SmartRock.
0.2
Finally, the introduction of the mechani-
0
cal work energy parameter provided the most Minimum Mean Maximum
representative information. This provides
an unambiguous parameter to quantify the Kli-Pi 4 (0 mm) Kli-Pi 3 (250 mm) Kli-Pi 2 (400 mm) Kli-Pi 1 (550 mm)
magnitude and variance of the quasi-static
loads and confinement characteristics. The Figure 18 Mechanical work: minimum, mean, maximum
distinction between the kinetic and potential
energy highlighted the effects of the depth
of influence for impact forces, locomo-
2.0
tive traction forces and the initial particle
1.8
­rearrangement after the installation of the
instruments. The minimum, mean and 1.6
maximum work energy statistics illustrated
Work (mJ) / Coefficient

1.4
a strong correlation with the depth of instal-
1.2
lation. The skewness statistic of the Kli-Pi
installed at the greatest depth converged 1.0
to a value of 1.1. This corresponds with a 0.8
confined granular state where the probability 0.6
distribution function tends towards an
0.4
exponential distribution, corresponding to
existing literature. The increase of the skew- 0.2
ness statistic corresponds with an increase in 0
both the standard deviation and CoV. Standard Deviation Coefficient of Variance (CoV) Skewness
Test regimes over a longer period should
be perused with the inclusion of typical Kli-Pi 4 (0 mm) Kli-Pi 3 (250 mm) Kli-Pi 2 (400 mm) Kli-Pi 1 (550 mm)

maintenance and corrective geometry

activities. The effects of ballast particle
properties, level of fouling, train speed,
moisture ingress and the substructure
condition, all contributing to both the
macroscopic and mesoscale behaviour,
have not been investigated or considered
in the results. Alternative instrumentation
Figure 19 Mechanical work: standard deviation, coefficient of variance (CoV) and skewness
locations should be included, particularly ADDITIONAL INFORMATION
the centre portion of the sleeper and ballast. The following tables provide a summary
The proof-of-concept to accurately quantify of the field test statistics – statistics of the
mesoscale ballast dynamics in detail has, displacement (Table 1), rotation (Table 2)
however, been successfully demonstrated. and work energy (Table 3).

Table 1 Peak-to-peak displacement statistics for the field test (all Kli-Pis)
Kli-Pi 1 Kli-Pi 2 Kli-Pi 3 Kli-Pi 4
Parameter
X Y Z X Y Z X Y Z X Y` Z

Mean (mm) 0.036 0.030 0.074 0.031 0.028 0.104 0.041 0.053 0.109 0.106 0.074 0.161

Minimum (mm) 0.008 0.006 0.009 0.007 0.004 0.007 0.007 0.004 0.009 0.013 0.008 0.005

Maximum (mm) 0.162 0.137 0.229 0.155 0.150 0.363 0.292 0.355 0.419 1.027 0.563 0.685

Standard deviation (mm) 0.032 0.028 0.056 0.291 0.298 0.925 0.064 0.073 0.100 0.177 0.117 0.187

Coefficient of Variance 0.881 0.916 0.762 0.925 1.050 0.887 1.574 1.385 0.918 1.666 1.591 1.165

Skewness 4.390 4.028 1.119 4.238 3.230 1.124 9.049 4.972 1.440 6.042 6.656 4.498

Journal of the South African Institution of Civil Engineering Volume 63 Number 1 March 2021 21
Table 2 Peak-to-peak rotation statistics for the field test (all Kli-Pis)
Kli-Pi 1 Kli-Pi 2 Kli-Pi 3 Kli-Pi 4
Parameter
X Y Z X Y Z X Y Z X Y Z

Mean (arcmin) 0.879 0.768 0.620 0.625 0.506 0.460 1.330 0.724 0.508 0.548 0.733 0.647

Minimum (arcmin) 0.261 0.332 0.283 0.265 0.185 0.203 0.180 0.239 0.192 0.285 0.390 0.345

Maximum (arcmin) 2.277 1.652 1.358 1.635 1.262 1.100 3.732 2.651 1.583 1.097 1.425 1.344

Standard deviation (arcmin) 0.542 0.322 0.260 0.314 0.264 0.204 1.145 0.481 0.293 0.184 0.241 0.218

Coefficient of Variance 0.617 0.419 0.419 0.503 0.522 0.444 0.861 0.664 0.577 0.336 0.329 0.337

Skewness 0.909 0.864 1.027 1.387 1.062 1.383 0.760 2.691 2.049 0.921 0.955 1.112

Table 3 Work energy statistics (all Kli-Pis) theory. PhD Dissertation, Japan: Kyoto University,
p 17.
Field test
Parameter Kouroussis, G, Caucheteur, C, Kinet, D, Alexandrou, G,
Kli-Pi 1 Kli-Pi 2 Kli-Pi 3 Kli-Pi 4 Verlinden, O & Moeyaert, V 2015. Review of

Mean (mJ) 0.287 0.415 0.424 0.639 trackside monitoring solutions: From strain gauges
to optical fibre sensors. Sensors, 15(8): 20115–20139.
Minimum (mJ) 0.037 0.028 0.034 0.032 Kruyt, N P & Rothenburg, L 2002. Probability density

Maximum (mJ) 0.885
Standard deviation (mJ) 0.219
Coefficient of Variance 0.761
Skewness 1.756
ACKNOWLEDGEMENT
Transnet Freight Rail is gratefully acknowl-
edged for sponsoring the Chair in Railway
Engineering in the Department of Civil
Engineering at the University of Pretoria.
DECLARATION
The authors declare no potential conflicts of
interest with respect to the research, author-
ship, and/or publication of this article.
REFERENCES
Aikawa, A 2009. Techniques to measure effects of
passing trains on dynamic pressure applied to
sleeper bottoms and dynamic behaviour of ballast
stones. Quarterly Report of Railway Technical
Research Institute, 50(2): 102–109.
Audley, M & Andrews, J D 2013. The effects of tamping
on railway track geometry degradation. Proceedings
of the Institution of Mechanical Engineers, Part F:
Journal of Rail and Rapid Transit, 227(4): 376–391.
Bennett, R, Hayes-Gill, B, Crowe, J A, Armitage, R,
Rodgers, D & Hendroff, A 1999. Wireless
monitoring of highways. Smart Structures and
Materials 1999: Smart Systems for Bridges,
Structures and Highways, 3671: 173–182.
Bowness, D, Lock, A C, Powrie, W, Priest, J A &
Richards, D J 2006. Monitoring the dynamic
displacements of railway track. Journal of Rail and
Rapid Transit, 221: 3–22.
1.445 1.641 2.699
0.363 0.390 0.633
0.874 0.921 0.990
1.442 1.122 1.113
Broekman, A & Gräbe, P J 2018. Development and
calibration of a wireless, inertial measurement
unit (Kli-Pi) for railway and transportation
applications. Proceedings, 37th Annual South African
Transportation Conference (SATC 2018), pp 521–531.
Cowin, S C & Satake, M 1979. Continuum mechanical
and statistical approaches in the mechanics of
granular materials. Journal of Rheology, 23(2):
243–256.
Cundall, P A & Strack, O D L 1979. A discrete
numerical model for granular assemblies.
Géotechnique, 29(1): 47–65.
Gräbe, P J & Clayton C R I 2009. Effects of principal
stress rotation on permanent deformation in rail
track foundations. Journal of Geotechnical and
Geoenvironmental Engineering, 135(4): 555–565.
Graizer, V M 2010. Strong motion recordings and
residual displacements: What are we actually
recording in strong motion seismology?
Seismological Research Letters, 81(4): 635–639.
Indraratna, B, Ngo, N T & Rujikiatkamjorn, C
2011. Behaviour of geogrid-reinforced ballast
under various levels of fouling. Geotextiles and
Geomembranes, 29(3): 313–322.
Ju, S H, Lin H T & Huang J Y 2009. Dominant
frequencies of train-induced vibrations. Journal of
Sound and Vibration, 319: 247–259.
Kempe, V 2011. Inertial MEMS: Principles and
Practice. 1st ed. Cambridge, UK: Cambridge
University Press.
Kitamura, R 1980. Analysis of deformation mechanism
of particulate material based on the probability
functions of contact forces for cohesionless
frictional granular materials. International Journal
of Solids and Structures, 39: 571–583.
Lamas-Lopez, F, Cui, Y J, Costa Aguiar, S & Calon, N
2017. Assessment of integration methods
for displacement determination using field
accelerometer and geophone data. Journal of
Zhejiang University, 18(7): 553–566.
Lekarp, F, Isacsson, U & Dawson A 2000. State of the
Art. II: Permanent strain response of unbound
aggregates. Journal of Transportation Engineering,
126(1): 76–83.
Lemkin, M & Boser, B E 1999. A three-axis
micromachined accelerometer with a CMOS
position-sense interface and digital offset-trim
electronics. IEEE Journal of Solid-State Circuits,
34(4): 456–468.
Li, D, Hyslip, J, Sussmann, T & Chrismer, S 2015.
Railway Geotechnics. Boca Raton, FL: CRC Press,
pp 93–100.
Liu, S, Huang, H, Qiu, T & Gao, L 2016a. Comparison
of laboratory testing using SmartRock and discrete
element modelling of ballast particle movement.
Journal of Materials in Civil Engineering, 29(3): 1–7.
Liu, S, Huang, H, Qiu, T & Gao, Y 2016b. Study on
the ballast particle movement at different locations
beneath crosstie using “SmartRock”. Proceedings,
2016 Joint Rail Conference, Columbia, SC.
Liu, S, Haung, H, Qiu, T & Kwon, J 2016c. Effect of
geogrid on railroad ballast particle movement.
Transportation Geotechnics, 9: 110–122.
Lobo-Guerrero, S & Vallejo, L E 2006. Discrete element
method analysis of railroad ballast degradation during
cyclic loading. Granular Matter, 8(3): 195–204.
Makse, H A, Johnson, D L & Schwartz, L M 2000.
Packing of compressible granular materials. Physical
Review Letters, 84(18): 4160–4163.
Milne, D, Le Pen, L, Watson, G, Thompson, D et al
2016a. Measuring ballast acceleration at track

22 Volume 63 Number 1 March 2021 Journal of the South African Institution of Civil Engineering
 level. Proceedings, 3rd International Conference on
Railway Technology: Research, Development and
Maintenance. 5–8 April 2016, Cagliari, Sardinia, Italy.
Milne, D, Le Pen L, Watson, G, Thompson, D et al
2016b. Proving MEMS technologies for smarter
railway infrastructure. Procedia Engineering, 143:
1077–1084.
Milne, D R M. Le Pen L M, Thompson D J, Powrie, W
2017. Properties of train loading frequencies and
their applications. Journal of Sound and Vibration,
397: 123–140.
Mvelase, G M, Anochie-Boateng, J K & Gräbe, P J 2012.
Application of laser-based technology to quantify
shape properties of railway ballast. Proceedings, 31st
Annual Southern African Transport Conference
(SATC), Pretoria, pp 243–254.
Nathanson, H C & Wickstrom, R A 1965. A resonant-
gate silicon surface transistor with high-Q bandpass
properties. Applied Physics Letters, 7(4): 84.
Nejikovsky, B & Keller, E 2000. Wireless
communications-based system to monitor
performance of rail vehicles. Proceedings, IEEE/
Journal of the South African Institution of Civil Engineering Volume 63 Number 1 March 2021
ASME Joint Railroad Conference, Newark, NJ,
pp 111–124.
Rothenburg, L & Kruyt, N P 2009. Micromechanical
definition of an entropy for quasi-static deformation
of granular materials. Journal of the Mechanics and
Physics of Solids, 57: 634–655.
Rucker, W 1982. Dynamic interaction of a railroad-bed
with the subsoil. Proceedings, Soil Dynamics and
Earthquake Engineering Conference, Southampton,
UK, pp 435–448.
Straser, E G & Kiremidjian, A S 1998. A modular,
wireless damage monitoring system for structures.
Technical Report No 128. Stanford, CA: John A.
Blume Earthquake Engineering Center, Stanford
University.
Stewart, J 2012. Calculus, Early Transcendentals.
7th ed. Boston, MA: Cengage Learning.
Tutumluer, E, Huang, H, Hashash, Y & Ghaboussi, J
2006. Aggregate shape effects on ballast tamping
and railroad track lateral stability. Proceedings,
AREMA Annual Conference, 23 May 2006, Salt
Lake City, UT, pp 17–20.
Wilke, D N, Govender, N, Pizette, P & Abriak, N E
2016. Computing with non-convex polyhedra on
the GPU. Proceedings, 7th International Conference
on Discrete Element Methods, 1–4 August 2016,
Dalian, China, pp 1371–1377.
Woodman, O J 2007. An introduction to inertial
navigation. Technical Report No 696. Cambridge,
UK: University of Cambridge Computer Laboratory.
Zhai, W M, Wang, K Y & Lin, J H 2004. Modelling and
experiment of railway ballast vibrations. Journal of
Sound and Vibration, 270: 673–683.
Zhang, X, Zhao, C & Zhai, W 2016. Dynamic
behaviour analysis of high-speed railway ballast
under moving vehicle loads using discrete element
method. International Journal of Geomechanics,
17(7).
BIBLIOGRAPHY
Selig, E T & Waters, J M 1994. Track Geo-technology
and Substructure Management. 1st ed. London:
Thomas Telford Ltd.
23
TECHNICAL PAPER
Journal of the South African
Institution of Civil Engineering
ISSN 1021-2019
Vol 63 No 1, March 2021, Pages 24–36, Paper 1102
PROF ROMAN LENNER (PE) is an Associate
Professor at Stellenbosch University in the
Department of Civil Engineering. His main
research interests include structural
reliability, bridge loading, target reliability
and assessment of existing structures. He is
a member of the Joint Committee on
Structural Safety, serves as Chair of the
IABSE South Africa National Group, is a member of the Editorial Board of
Structural Engineering International, and is a member of the
IABSE TG1.3‑working group.
Contact details:
Department of Civil Engineering, Stellenbosch University
Private Bag X1, Matieland, Stellenbosch 7602, South Africa
T: +27 21 808 9572, E: rlenner@sun.ac.za
SUZANNE BASSON, a Candidate Civil
Engineer, completed her BEng and MEng in
Civil Engineering (cum laude) at Stellenbosch
University in 2017 and 2020, respectively. She
is currently employed as a civil engineer at
Zutari in their Cape Town office. Her research
interests include probabilistic modelling,
structures, and bulk water supply.
Contact details:
Zutari
Cape Town, South Africa
T: +27 21 526 9463, E: suzanne.basson@zutari.com
PROF MIREK SÝKORA is an Associate Professor
at the Klokner Institute, Czech Technical
University in Prague. His research interests
cover the basis of structural design, structural
reliability, probabilistic optimisation, load
modelling, risk assessment of technical
systems, and applications of probabilistic
methods in structural design. He is a member
of the Joint Committee on Structural Safety, fib COM3 Existing Concrete
Structures, CEN/TC 250/WG 2 Assessment of Existing Structures, and CEN/
TC 250/SC 1/WG 3 Traffic Loads on Bridges. He is also a member of the
Editorial Boards of Structural Concrete (official journal of fib) and Acta
Polytechnica (subtitled as Journal of Advanced Engineering).
Contact details:
Klokner Institute
Czech Technical University in Prague
Czech Republic
T: +420 224 353 529, E: miroslav.sykora@cvut.cz
DR PIERRE VAN DER SPUY (Pr Eng, MSAICE) is a
Principal Bridge Engineer with Zutari in Cape
Town. He is registered with the Engineering
Council of South Africa as a Professional
Engineer and is a corporate member of the
South African Institution of Civil Engineering
(SAICE). He recently completed a PhD in bridge
traffic loading at Stellenbosch University.
Contact details:
Zutari
Cape Town, South Africa
T: +27 21 526 9400, E: pierre.vanderspuy@zutari.com
Keywords: bridge loading, structural reliability, WIM data, TMH7,
traffic loads
24
Reliability performance of
bridges designed according
to TMH7 NA load model
R Lenner, S E Basson, M Sýkora, P F Van der Spuy
The current code of practice for the design of highway bridges in South Africa is called Technical
Methods for Highways 7 (TMH7). It was first published in 1981 and is largely based on the
loading provisions of the BS5400. The purpose of this study is to carry out an investigation into
the reliability performance of new highway bridges designed according to TMH7. Past studies
have identified deficiencies in the traffic load model. A case study considering isolated traffic
load effects and sagging moment on a simply supported bridge indicates that NA loading
generally performs well for spans ranging from 15 to 50 metres. However, a poor reliability
performance is seen for short-span narrow bridges, especially for 5 m and 10 m spans where
the number of traffic lanes is equal to the number of notional lanes for NA loading. To study
the reliability performance of a critical element, flexural behaviour of a 20 m twin-spine simply
supported deck is then investigated. The results agree with the first case study. It is, however,
recommended to verify the presented findings by investigation based on data from other
heavy-freight routes and to account for continuous structures and shear load effects.
Introduction traffic conditions. The performance of a
The structural performance of bridges bridge at the ultimate limit state (ULS)
is essential in ensuring a sound road is also investigated, with the traffic load
transport network. The current code effect taken as static bending moment
of practice for the design of highway at midspan of a simply supported bridge
bridges in South Africa is called Technical (excluding dynamic effects arising from
Methods for Highways 7 (TMH7) (CSRA the complex vehicle-structure interac-
1998). It was first published in 1981 and tion). Furthermore, this study investigates
is largely based on the loading provisions only the free-flow traffic, which is typi-
of the British Code BS5400, which has cally assumed to govern extreme traffic
since been updated and superseded by the load effects for short- to medium-span
Eurocodes. Since TMH7’s last revision bridges ranging from 5 to 50 metres
in 1988, the road freight industry has (Bruls et al 1996).
expanded significantly, encouraged by an In achieving the study’s purpose, reli-
economic deregulation of road transporta- ability analyses are performed for two
tion in 1988, and further exacerbated by case studies based on actual traffic load
changes to the traffic regulations in South effects derived from site-specific weigh-
Africa. As a result, concern is raised as in-motion (WIM) data. For each analysis,
to whether TMH7-based design delivers a reliability index is obtained which is
adequately safe bridges for the current indicative of the safety level of the struc-
traffic demand. ture. The results are compared to target
The purpose of this study is to carry reliability indices from existing standards
out an investigation into the reliability to draw inferences regarding the reliability
performance of new highway bridges performance of TMH7 for normal traffic
designed according to the NA load model conditions.
of TMH7 (Basson & Lenner 2019). Past The two case studies are based on WIM
studies have identified deficiencies in data obtained from stations of two different
TMH7’s traffic load model for normal national routes. The first case study is based
traffic conditions, known as NA load- on WIM data from Roosboom Station on
ing (Oosthuizen et al 1991; Anderson National Route 3 (N3) between Durban
2006; Van der Spuy & Lenner 2018). In and Johannesburg – this was selected as
response, this study focuses on the reli- Roosboom had served as a reference sta-
ability performance of TMH7 for normal tion in previous studies (Lenner et al 2017;
Lenner R, Basson SE, Sýkora M, Van der Spuy PF. Reliability performance of bridges designed according to TMH-7 NA load model.
J. S. Afr. Inst. Civ. Eng. 2021:63(1), Art. #1102, 13 pages. http://dx.doi.org/10.17159/2309-8775/2021/v63n1a3
Van der Spuy & Lenner 2018; Pérez &
Lenner 2019). The second case study is
based on WIM data from National Route 1
(N1) near Kilner Park, as used in the study
of multiple-lane presence factors (Van der
Spuy et al 2019b).
The scope of each case study is dictated
by the number of traffic lanes measured at
the WIM site. For Roosboom only the slow
(outer) lane data is available, which is a
limitation for a full probabilistic reliability
analysis. This is why the focus is only on
the modelling of traffic load extremes.
Using the semi-probabilistic approach,
EN 1990 (EN 2002) and ISO 2394
(ISO 2015), the traffic load effects are
assigned with a fixed sensitivity factor α
and the design value is derived from the
data as an appropriate fractile of a fitted
probability distribution. A second case
study based on the data from Kilner Park
then provides a fully probabilistic reliability
analysis. Two lanes per direction are mea-
sured at this site and it is therefore possible
to develop a meaningful loading due to
collected WIM data on a typical bridge
deck. A 20 m two-lane, single-span bridge
is selected as a representative example.
A critical element reliability analysis is
carried out including the resistance, per-
manent and traffic load effect variables.
A sensitivity analysis follows to determine
the relative significance of the basic vari-
ables on the obtained reliability indices.
This study is based on the derived load
effects from previous work by Van der
Spuy & Lenner (2019a), who used cleaned
and calibrated WIM data in a moving load
analysis to derive the traffic load effects
for various span lengths and structural
systems. Utilisation of already derived
load effects is somewhat limiting, as it
is difficult to investigate subsets of the
data and to identify the critical loading
events such as two vehicles in the same
lane at a critical gap. Van der Spuy &
Lenner (2019a) also found it difficult to
distinguish in WIM data between illegally
overloaded standard trucks and permit
trucks. Since overloading is a growing
concern in South Africa (DOT 2013;
Lenner et al 2017) and the NA loading in
TMH7 should provide the provision for
overloading of legal vehicles (Anderson
2006), the whole data set of traffic load
effects is considered in this study.
Uncertainty in WIM is neglected
in the presented analysis. Two previ-
ous studies by Jacob and Feypell-de La
Beaumelle (2010), and Morales-Nápoles
Journal of the South African Institution of Civil Engineering Volume 63 Number 1 March 2021
40
Qa = Average load per metre of notional lane in kN
36
32
28
24
20
0 20
L = Effective loaded length in metres
Figure 1 L oading curve for type NA loading (CSRA 1998)
and Steenbergen (2014) classified WIM
systems into low-speed (5–15 km/h) and
high-speed (normal speeds of the traffic
flow). The latter – of interest here – are
expected to have an accuracy of 10–25%
for 95% of vehicles, while the accuracy of
the former is much higher (estimated to
be 3–5%). The accuracy depends on the
calibration of a WIM system; the key issue
is to remove possible biases, particularly
for high vehicle weights. When the bias is
correctly removed (as is assumed in this
study), uncertainty in WIM measure-
ments is deemed to have a small influence
on predicted traffic load effect extremes.
De Wet (2010) shows that WIM errors in
South Africa are generally less than 10%.
The accuracies mentioned here fall within
the B(10) accuracy class of COST 323
which are suitable for the development of
bridge live load studies (Jacob et al 2002).
O’Connor and O’Brien (2005) show that
an accuracy as low as C(15) does not have
an appreciable effect on predicted extreme
values. Yet, this is to be investigated in
detail in further research.
Overview of na loading
TMH7’s traffic load model is provided
in Section 2.6 of Part 2 of TMH7 (CSRA
1998), where three types of live loading are
defined: normal loading (NA), abnormal
loading (NB) and super loading (NC).
Typical requirement by the authorities for
all highway bridges is at least an envelope
180
Qa = +6
√L
36 40 60 80 100
of load effects due to NA loading and a
36-unit of NB loading (CSRA 1998). The
NA load model utilises a floating notional
lane width, where the number of notional
lanes depends on the carriageway width
(refer to Section 2.6.2. in TMH7). The
notional lanes do not represent actual traf-
fic lanes, but are used for the application
NA and NB loading which is characterised
by a uniformly distributed load applied
to any continuous part of a notional lane,
together with a single axle load applied
to each notional lane. The uniqueness of
the uniform loading is that its intensity
depends on the effective loaded length
defined as the aggregate length of all the
loaded parts on a single or combination of
notional lanes. Here “parts” refer to sec-
tions of the notional lanes that are loaded
to obtain the most adverse load effects at a
specific location on a bridge. The loading
curve describing the application of the uni-
formly distributed load Qa , in accordance
with the effective loaded length L, is illus-
trated in Figure 1. The axle load per each
notional lane is given by the expression
144/√n, where n is the sequence number
of the notional lane under consideration
(CSRA 1998).
In order to obtain ultimate limit state
values for the design, partial factors of
γfL = 1.5 and γf3 = 1.1 as provided in
Section 5 of TMH7 are used. As the NA
loading intrinsically accounts for dynamic
effects of moving traffic by means of a
“Swiss Formula” (see Section 2.6.1.3. in
25
TMH7), this amplification is removed
from the NA load model specifically for
this study to enable a comparison with the
load effects derived on the basis of static
WIM data.
The investigation of uncertainty in
dynamic effects of heavy traffic at the ulti-
mate limit states is beyond the scope of this
contribution; for further discussion see for
instance work by Lenner & Sýkora (2016)
and Ivanković et al (2019).
Roosboom case study (RCS)
The first case study consists of a reliability
analysis considering only the traffic load
effect based on seven years of WIM data
for a single lane. It is first necessary to
obtain the distribution function describing
the load effect and then formulate the limit
state function. This enables the calculation
of the reliability index by means of the First
Order Reliability Method (FORM).
Overview of traffic load
effect calculation
The work presented in this study is based
on the load effects obtained previously
by means of moving-load analysis which
accounts for single-truck events and mul-
tiple-truck events (Van der Spuy & Lenner
2018). Only the sagging bending moment in
increments of 5 m for 5–50 m span lengths
is presented in this work to limit the scope.
Probabilistic modelling of
traffic load effects
Consideration of single-loading events
leads to a distribution of the load effects
which need to be described by a probability
distribution. There are different approaches
that vary from utilising basic fitting of a
normal distribution to the population or to
the tail (Nowak 1993; Nowak & Rakoczy
2013; Soriano et al 2016), to the more com-
mon extreme value theory and tail fitting
(Coles 2001; Caprani et al 2008; Enright
2010; O’Brien et al 2015).
A block maxima approach is adopted
in this study as it is a proven and effective
technique in studying extreme traffic
events (Crespo-Minguillón & Casas 1997;
Caprani et al 2003; Getachew & O’Brien
2007; Hajializadeh et al 2012). It essentially
isolates a maximum event in each block.
To utilise the technique, the underly-
ing assumption is that the data must be
independent and identically distributed
(iid) (Coles 2001). The minimum size of
a block is one day (Caprani 2012; Van der
26
2.0
Fréchet
α > 1.14
1.5
Gumbel α = 1.14
Skewness (α)
1.0 Three-parameter
lognormal for any α
Weibull
α < 1.14
0.5
0
Coefficient of Variation (V)
Figure 2 R
 elation between skewness and coefficient of variation for monthly maxima for each
span length of RCS; the points indicate V and α for various span lengths (Basson 2020)
Spuy 2020). However, larger blocks tend to
remove variations in observations and yield
data that better adheres to the iid require-
ment. Considering seven years of available
recordings, monthly maxima of the bend-
ing moment are further used in this study.
The block is large enough to reduce the
variations, and there are enough blocks for
the statistical description. Detailed analysis
of statistical inference methods and proba-
bilistic modelling of load effect maxima are
offered by O’Brien et al (2015).
A simplified approach exploring a
wide range of extreme value distribution
as developed by Holický (2013) is adopted
here. It is particularly useful in finding
applicable models by plotting the skew-
ness and the coefficient of variation of
the data set against plotted distributions.
Considering the monthly blocks, the
sample skewness and coefficient of varia-
tion, denoted by α and V respectively, are
used to plot data points on the developed
diagram for each span length.
Figure 2 clearly indicates the fit of the
data to various distributions including
the GEV family, i.e. the Gumbel, Fréchet
and Weibull distribution. The V ranges
between 0.08 and 0.10, which indicates
a relatively narrow distribution shape in
comparison to the characteristics of daily
and weekly maxima where V ranges 0.12 to
1.16, and 0.10 to 0.14 respectively. All span
lengths have a positive skewness (right tail)
equal to or less than 0.8, except for a 15 m
span that has a small negative skewness
(left tail). Furthermore, the skewness for
the 10 m and 15 m spans can be observed
Volume 63 Number 1 March 2021
Two-parameter
lognormal
α = 3V + V 3
Gamma
α = 2V
Beta with zero
lower bound α < 2V
as close to zero, which indicates a distribu-
tion that approaches a normal distribution
(refer to Basson 2020 for full details).
Most of the points in Figure 2 are
located above the two parameter lognor-
mal line, which suggests that the three
parameter lognormal (LN3) distribution
or Weibull distribution (for maxima) is
appropriate for modelling the monthly
maxima. The GEV distribution can be
used conveniently, as it approaches the
Weibull distribution when the skewness
is less than 1.14. Both the LN3 and GEV
distribution are asymmetrical and have
three model parameters. Both models are
flexible in allowing for both positive and
negative skewness and can thus cater for
different sample characteristics obtained
from the different span lengths. As a
result, the LN3 and GEV distributions are
further investigated with an aim to decide
on an appropriate model.
Assessment of selected
probabilistic distributions
Goodness-of-fit tests performed in R
(R Core Team 2018) assess the suitability
of both the LN3 and GEV distribution in
­representing the monthly maxima. The
tests consist of diagnostic plots and the
modified Anderson-Darling (AD) (Ang
& Tang 2006) hypothesis testing. For
each considered distribution, the model
parameters are inferred by the maximum
likelihood estimation (MLE).
Visual inspection of the diagnostic
plots for all span lengths shows that both
models fit the data well. The points on
Journal of the South African Institution of Civil Engineering
 Emperical and theoretical density Q–Q plot
0.020
180

160
0.015

Empirical quantiles
140
Density

0.010
120

100
0.005
80

0 60
60 80 100 120 140 160 180 200 80 100 120 140 160 180
Data Theoretical quantiles

Emperical and theoretical CDFs P–P plot

1.0 1.0

0.8 0.8
Empirical probabilities

0.6
0.6
CDF

0.4
0.4

0.2
0.2

0 0
60 80 100 120 140 160 180 200 0 0.2 0.4 0.6 0.8 1.0
Data Theoretical probabilities

Figure 3 Diagnostic plots for the monthly maxima of sagging bending moment for a 5 m span RCS represented by the LN3 distribution (Basson 2020)

the probability plots adhere to a straight The AD test is further used as a numeri- Table 1 M
 odified Anderson-Darling p-values for
line and a good fit is seen for the density cal measure to substantiate the results LN3 distribution and GEV distribution
plots. For illustrative purposes, the Q-Q obtained from the diagnostic plots. When fitted to monthly maxima of RCS
plots obtained for a 5 m span are shown the p-value obtained from the AD test Span p-value p-value
in Figure 3 for LN3 and in Figure 4 for exceeds the significance level, the data length (m) (LN3) (GEV)
GEV. In the figures, the following plots are can be represented by the selected model.
5 0.73 0.75
included (clockwise from top left): A significance level of 0.05 is typically
1. Histogram with a fitted probability accepted (Fisher 2006). The LN3 distribu- 10 0.27 0.45
density function tion is converted to a normal distribution
15 0.15 0.33
2. Plot with empirical and theoretical (Holický 2013). The AD test applied to the
quantiles – probability paper of a theo- transformed data obtains p-values as pro- 20 0.06 0.09
retical distribution vided in Table 1. The p-values exceed 0.05,
3. Empirical and theoretical cumulative which indicates that the monthly maxima 25 0.58 0.58

distribution function can be represented by an LN3 distribution. 30 0.95 0.96

4. Plot with probabilities corresponding Similar p-values were obtained for the GEV
to empirical and theoretical quantiles, distribution, although higher values are seen 35 0.93 0.92
providing a similar insight as the for 10–20 m spans (see Table 1). The results
40 0.42 0.42
second plot. agree with the diagnostic plots that both the
In the second and fourth plots a straight LN3 distribution and the GEV distribution 45 0.35 0.39
line would indicate a perfect fit of the data can be used to adequately represent the
50 0.56 0.60
by the theoretical distribution. monthly maxima.

Journal of the South African Institution of Civil Engineering Volume 63 Number 1 March 2021 27
 Probability plot Quantile plot
1.0

360
0.8
340

0.6

Empirical
320
Model

0.4 300

280
0.2

260
0
0 0.2 0.4 0.6 0.8 1.0 260 280 300 320 340 360
Empirical Model

Return level plot Density plot

0.020
380

360
0.015

340
Return level

f(z)

320 0.010

300

0.005
280

260
0
1e-01 1e+00 1e+01 1e+02 1e+03 240 260 280 300 320 340 360 380
Return period z

Figure 4 Diagnostic plots for the monthly maxima of sagging bending moment for a 5 m span RCS represented by the GEV (Basson 2020)

The Q-Q plots in Figures 3 and 4 suggest
that upper tail behaviour is described approx-
imately only. Detailed analysis that might
improve the model for traffic load extremes
will be the scope of further research.
Design load effect
The overall reliability level, described by
the reliability index β, can be split into the
resistance part R and load effect part E
(EN 2002; ISO 2015). For this case study,
the load effect part is of interest, where
the reliability level is expressed with the
load effect index βE equal to αE β. The
variable αE is a sensitivity factor of traffic
load effects obtained from FORM. This
describes the relative importance of the
load effects in obtaining β and is accepted
in this work as –0.7 for dominant action
as recommended by EN 1990 (EN 2002)
and ISO 2394 (ISO 2015). This αE-value
is deemed appropriate for shorter spans,
while it may be reduced for longer spans
with the dominant dead load.
When only traffic load effects are ana-
lysed, a reliability criterion can be defined
as the instance when the actual traffic load
effects exceed the design load effect. A ref-
erence period of 50 years is chosen in order
to provide for comparison with the target
reliability commonly listed in codes such
as EN 1990 (EN 2002) or SANS 10160‑1
(SANS 2019). Therefore, the actual load
effects derived from the WIM data should
correspond to a 50-year maximum. The
design load effect can then be described
as a fractile of the 50-year maximum load
effect distribution F50 (x). The probability
of failure pf is then approximated by the
probability that the design load effect ed
is exceeded by the traffic load effects E, as
shown in Equation 1. Equation 2 is used to
estimate the overall β, where the probabil-
ity distribution function of a standardised
normal distribution FU(x) relates β to pf .
pf = P(E > ed ) = 1 – F50(ed )(1)
pf = FU(+αEβ) = FU(–0.7β)(2)
where FU (·) denotes the cumulative distri-
bution function of the standardised normal
variable and F50 (·) is the 50-year maximum
load effect distribution.
50-Year maximum load effect distribution
To obtain the maximum load effect
distribution for a 50-year period, the
monthly maxima distribution is statistically
projected by raising the probability
distribution function F(x) to an appropriate
power n (per Equation 3). In Equation 3,
n is the number of monthly maxima in

28 Volume 63 Number 1 March 2021 Journal of the South African Institution of Civil Engineering
 Notional lane
Traffic lane
Carriageway width = 3 m
Design scenario 1: 1 traffic lane; 1 notional lane
Notional lane Notional lane
Shoulder Traffic lane
Carriageway width = Above 7.4 m up to and including 11.1 m
Design scenario 2: 1 traffic lane; 3 notional lanes
Figure 5 Design scenarios
a reference period, i.e. n = 600, and Fn(x)
denotes the probability distribution function
of the 50-year maxima, F50 in Equation 1.
Fn(x) = [F(x)]n(3)
For each span length, the 50-year maxi-
mum load effect distribution was obtained
from both the LN3 distribution and the
GEV distribution of monthly maxima.
Depending on the selected model, the char-
acteristics of the projected maximum load
effect distribution will differ. According to
Castillo et al (2005), a lognormal distribu-
tion fitted to maximum values converges to
a Gumbel distribution as the power n tends
to infinity.
The Weibull distribution for maxima,
represented by the GEV distribution, has
a finite upper bound and its maximum
domain of attraction remains a Weibull
distribution (Castillo 1988). The concern
with using the maximum Weibull distri-
bution is that the upper bound remains
unchanged irrespective of the reference
period projected to. This means that the
50-year maximum load effect distribution
has the same upper bound as the monthly
maxima distribution. This could potential-
ly be problematic, as the design load effects
may exceed the bound determined by the
distribution function parameters.
The physical argument that might jus-
tify the upper bound could be formulated
on the basis of legal limits for axle and
vehicle loads. However, the concern about
frequent overloading of trucks makes this
argument doubtful. Further, the LN3 dis-
tribution shows a similar fit to the monthly
maxima as the GEV distribution (Figures 3
Journal of the South African Institution of Civil Engineering Volume 63 Number 1 March 2021
Notional lane
Shoulder
Carriageway width = Above 7.4 m up to and including 11.1 m
Design scenario 3: 2 traffic lane; 3 notional lanes
Notional lane Notional lane
Shoulder
Carriageway width = 4.8 m up to and including 7.4 m
Design scenario 4: 2 traffic lanes; 2 notional lanes
and 4), and its unbounded right tail (for
positive skewness) allows for the thorough
investigation of the design load effects. As a
result, an LN3 distribution is adopted here
for the monthly maxima.
Reliability analysis for
traffic load effects
Formulation of limit state function
The limit state function Z defines the
desired failure mode for the investigation
as per Equation 4. Model uncertainty θE is
introduced in this contribution according to
a lognormal distribution with a mean of 1.0,
and a COV of 0.1 is commonly used in lit-
erature (JCSS 2001; fib Bulletin 80 2016; Von
Scholten 2004). The design load effect ed is
a deterministic value provided by TMH7 for
NA loading, while E50 is described by the
50-year maximum load effect distribution
with an LN3 distribution.
E = θE E50
Z = ed – θE E50(4)
FORM is implemented to estimate failure
probability for each span length from
Equation 4; the overall reliability index β is
then derived from Equation 2.
Overview of reliability analysis
for different design scenarios
The actual traffic load effects are derived
from the WIM data for a single traffic lane.
To make a meaningful comparison with
TMH7 it is necessary to investigate different
scenarios. The concept of a notional lane as
defined in TMH7 is different to the actual
traffic lane, meaning that a notional lane
Notional lane Notional lane
Traffic lane Traffic lane Shoulder
Notional lane
Traffic lane Traffic lane
is often narrower than a traffic lane. As a
result, a number of notional lanes can be
fitted to a bridge deck which in reality only
carries a single traffic lane (bridges are built
to accommodate not only a traffic lane,
but also shoulders). Four design scenarios
are defined where initially an assumption
of one traffic lane corresponding to one
notional lane is investigated, followed by the
opposite notion of one traffic lane occupy-
ing a deck which is in reality designed for
three notional lanes. The same is carried
out for two possible traffic lanes and three
and two notional lanes respectively. A sum-
mary of the scenarios under consideration is
depicted in Figure 5 and provided in Table 2.
Table 2 Overview of design scenarios for RCS
Design
Description
scenario
1 1 traffic lane, 1 notional lane
2 1 traffic lane, 3 notional lanes
3 2 traffic lanes, 3 notional lanes
4 2 traffic lanes, 2 notional lanes
The design values of traffic load effects, ed,
for each scenario given in Table 3 represent
the rounded maximum obtained bending
moments according to TMH7 NA loading
for the midspan of a beam when designing
for one, two and three notional lanes. It is
important to note that dynamic effects are
excluded from the NA load model based
on the Swiss formula, and the ULS partial
factor γf = 1.65 is obtained as a product
of γfL = 1.5 and γf3 = 1.1. The calculated
reliability indices β for each span length are
given in Table 4.
29
 11 950 Overall bridge width

10 900 Carriageway width

525 2 500 3 700 3 700 1 000 525

1 000

1 750

250

300

200
1 800
4 450 2 000

Figure 6 Cross-section of bridge deck with traffic lane arrangement (units in mm) of KPCS (Basson 2020)

Table 3 D
 esign traffic load effects (in kNm)
Traffic lane 1 Traffic lane 2 for ULS determined from NA loading
3.7 m 3.7 m in TMH7
–0.18 Span One Two Three
length notional notional notional
(m) lane lanes lanes
5 360 650 920
10 1 050 1 960 2 810
LLDF2 = 0.49 15 2 080 3 950 5 740
20 3 470 6 900 8 900
25 5 200 9 720 12 500
LLDF1 = 0.95 30 7 280 12 900 16 600
35 9 720 16 400 21 200
1.18
40 13 100 20 200 26 200
45 15 700 24 300 31 600
Figure 7 L ateral load distribution factors corresponding to the position of traffic load effects in
50 18 500 28 700 37 300
two lanes (Basson 2020)

In order to make a meaningful com-
parison of the obtained results for a 50-year
reference period, EN 1990 (EN 2002)
recommends βt = 3.8 while the South
African National Standard, SANS 10160-1
(SANS 2019), uses a βt of 3.0 for the same
reference period. It is readily observed that
the way each scenario is defined is funda-
mental to the obtained reliability index. The
extreme case of one notional lane to one
traffic lane shows very low β (therefore high
probability of failure) for short spans, while
for spans ≥ 25 m the performance is deemed
satisfactory. The poor performance in this
scenario can be contributed to the way a
notional lane is defined, and that it is physi-
cally impossible to have one notional lane
equal to one traffic lane. It is neglecting the
mandatory shoulders on the roads – mean-
ing the deck width is always wider than a
traffic lane and therefore necessitates more
notional lanes during the design.
On the other hand, the case of wide
shoulders results in a scenario where three
notional lanes are compared to a single
traffic lane. In this case the obtained values
show high reliability, mostly in excess of
β = 10, leading to very low probability of
failure. Care must be exercised in this case,
as this is pointing only to the global action
effects and it might not necessarily be true
to a critical element of a bridge, for instance
a single girder. Yet, it does provide an insight
into the performance of NA loading and the
inherent conservatism in the definition of
notional lane width. Design scenarios 3 and
4 exhibit similar results, where two traffic
lanes are considered and again, when the
number of notional lanes is matched with
the number of reflected traffic lanes, a poor
reliability performance is observed. At the
same time, for all considered scenarios, the
longer span lengths prove to show satisfac-
tory performance when a target β of 3.0 or
even 3.8 is considered.
Kilner Park Case Study (KPCS)
The second case study utilises three
consecutive years of WIM data for the two
outer traffic lanes in both directions (Van
der Spuy et al 2019c). A 20 m reinforced
concrete bridge with a twin spine deck is
considered as a representative example for
bridges in South Africa. The effect of lane
loading and load distribution between two
main girders is investigated.
Bridge deck analysis and design
The cross-sectional properties of the
designed bridge deck, together with the
traffic lane arrangement, are illustrated in
Figure 6. Two actual traffic lanes of 3.7 m
width each are considered with necessary
surfaced shoulders on both sides. The
deterministic analysis reveals that the spine
beam is the critical member. Note that this
bridge is not skew. As the maximum of
design load effects to load effects per meas-
ured vehicles is considered in this inves-
tigation, it is expected the results of the
analysis would be similar for skew bridges.
The permanent loads for the design
include the dead load of the bridge
deck, typical F-shape Type A parapets
(SANRAL 2012) and asphalt road surfacing
with a thickness of 40 mm (SANRAL 2010).
NA loading (excluding dynamic effects)

30 Volume 63 Number 1 March 2021 Journal of the South African Institution of Civil Engineering
is applied to the grillage model using Table 4 O
 verall reliability indices β provided for each design scenario and corresponding to a
three notional lanes that correspond to 50-year period of RCS
the considered deck width as per TMH7 Design Design Design Design
requirements. The ULS mid-span moment Span length (m)
scenario 1 scenario 2 scenario 3 scenario 4
for the considered beam according to the
5 –1.2 9.7 3.8 –0.7
analysis equals to 11 600 kNm, which
10 –1.6 10.8 4.2 –0.7
results in 29 Y40 (450 MPa yield strength)
steel bars required for flexural resistance. 15 1.2 15.0 7.5 2.3
The thickness of the deck here is the main 20 1.7 10.6 6.1 3.2
factor influencing the amount of moment
25 2.8 10.3 6.2 3.6
distribution per spine. The next section
30 3.8 10.6 6.6 3.9
explains lateral load distribution factors
used in this case. 35 4.3 10.4 6.4 3.9
40 5.1 10.1 6.3 3.8
Influence of traffic load effects 45 5.2 10.2 6.5 4.0
on critical spine beam
50 5.6 10.7 6.9 4.4
The transverse stiffness of the considered
superstructure is a key factor for resulting
load effects due to WIM-based loading on
the spine beam. By using the transverse Traffic lane 1 Traffic lane 2 Traffic lane 3
3.5 m 3.5 m 3.5 m
influence line for the critical element
(Figures 7 and 8) it is possible to capture the –0.18
load sharing effect and determine how much
of the total load is resisted by the spine.
The lateral load distribution factor (LLDF) LLDF3 = 0.09
numerically describes the contribution of
each WIM lane to the bending moment in
the investigated girder and is located at the LLDF2 = 0.52
centre of the traffic lane. Uncertainty in the
lateral load distribution is accounted in the
probabilistic analysis by load effect model
LLDF1 = 0.96
uncertainty, θE, similarly to Equation 4.
Apart from the scenario depicted in 1.18
Figure 6, a further loading arrangement is
investigated for the case of a bridge utilised Figure 8 L ateral load distribution factors corresponding to the position of traffic load effects in
as a three-lane deck providing a reduced three lanes (Basson 2020)
lane width. The critical arrangement of
WIM traffic lanes, along with the transverse
influence line in this case corresponds to 2.0
Figure 6. This loading scenario is provided
in order to capture the possibility that a
Fréchet Two-parameter
bridge initially designed for two traffic lanes α > 1.14 lognormal
with three notional lanes may actually be, 1.5 α = 3V + V 3
under special circumstances, utilised as a
three-lane bridge with reduced shoulders. Gumbel α = 1.14
Skewness (α)

It is possible to have a different thickness

1.0 Three-parameter
of the deck which would result in different Gamma
lognormal for any α α = 2V
LLDF factors. Due to the adopted linear
shape of transverse distribution depicted Weibull
in Figures 7 and 8, the change in deck α < 1.14
0.5
thickness will result in accordingly scaled Beta with zero
distribution factors and scaled results. lower bound α < 2V

Probabilistic modelling of
0
traffic load effects 0 0.1 0.2 0.3 0.4
The monthly maxima combined for two Coefficient of Variation (V)
opposite travel directions, representing the
extreme traffic load effects for the critical Figure 9 A
 diagram where the relation between skewness and coefficient of variation is used to
spine beam, are modelled and assessed find applicable models to represent the combined monthly maxima for each load case
similarly to the RCS. Figure 9 illustrates the of KPCS; the points indicate V and α for various span lengths (Basson 2020)

Journal of the South African Institution of Civil Engineering Volume 63 Number 1 March 2021 31
 Emperical and theoretical density Q–Q plot
0.0035

0.0030 600

0.0025 500

Empirical quantiles
0.0020
Density

400
0.0015
300
0.0010

0.0005 200

0
100 200 300 400 500 600 700 100 200 300 400 500 600 700
Data Theoretical quantiles

Emperical and theoretical CDFs P–P plot

1.0 1.0

0.8 0.8
Empirical probabilities

0.6 0.6
CDF

0.4 0.4

0.2 0.2

0 0
100 200 300 400 500 600 700 0 0.2 0.4 0.6 0.8 1.0
Data Theoretical probabilities

Figure 10 Diagnostic plots of sagging moments provided in R for two southbound lanes combined of KPCS fitted with the LN3 distribution (Basson 2020)

characteristics of each load case and relates
the sample skewness to the coefficient of
variation. It is observed that the sample
skewness varies considerably, which can be
conveniently captured by the LN3 distribu-
tion. Consistently with the RCS, the use of
the Weibull distribution is avoided, as the
fixed upper bound is deemed to be unjusti-
per Table 5) are considered. As in the first Table 5 M
 odified Anderson-Darling p-values
case study, the adopted probabilistic model for LN3 distribution fitted to
provides only an approximation of the upper combined monthly maxima of KPCS
tail behaviour, and detailed analysis might Load
improve predictions of traffic load extremes. Permutations p-value
Case
The modified AD test results are sum-
1 Nb1, Nb2 0.11
marised in Table 5. Nb1 and Nb2 respective-
ly refer to the slow lane and fast lane in the 2 Sb1, Sb2 0.98

fied. As only extracted load effects were
available for this project, it is impossible to
determine the cause of the high skewness
for the single-load effect situated in the
Frechet domain.
The moment parameters for the LN3
distribution are estimated using MLE. Visual
inspection of the diagnostic plots shows
that the LN3 distribution fits the combined
monthly maxima well. For illustrative pur-
poses, Figure 10 shows the diagnostic plots
for the load case when the two traffic lanes
in the southbound direction (Load Case 2
northbound direction. The same reference 3 Nb2, Nb1 0.88
is made for Sb1 and Sb2 in the southbound 4 Sb2, Sb1 0.63
direction. The traffic lane combinations
5 Nb1, Sb1 0.14
for each load case are described in Table
5, where the traffic lanes are ordered from 6 Sb1, Nb1 0.89
left to right in accordance to the closest and 7 Nb1, Nb2, Sb2 0.11
furthest traffic lane from the critical spine
8 Sb1, Sb2, Nb2 0.71
beam. It shows the recorded traffic used in
each of the considered lanes. The p-value for
each load case exceeds a significance level of The results show relatively high p-values
0.05, which confirms that the LN3 distribu- for all the load cases, except for Load Cases
tion can be used to represent the combined 1, 5 and 7 that have much lower p-values.
monthly maxima. This observation concurs with the diagnostic

32 Volume 63 Number 1 March 2021 Journal of the South African Institution of Civil Engineering
Table 6 C
 onventional probabilistic models describing the basic variables in the limit state function
Symbol Standard
Distribution Mean
Name of basic variable of basic Unit deviation Reference
type (μ X)
variable (X) (σX)

Dead load G kNm N Gk 0.05μX

Von Scholten et al 2004; Holický 2009
Superimposed dead load Gw kNm N Gwk 0.1μX

Traffic load (50 years) Q kNm LN3 Based on moment parameters of 50-yr maximum load effect distribution

Concrete strength fcu kPa LN fcuk + 1.645σX 0.18μX Holický 2009

Yield strength fy MPa LN f yk + 1.645σX 30 Holický 2009

Effective flange width b m Det bk 0 Measured in-situ

Effective depth to reinforcement d m N dk 0.02μX Lenner & Sýkora 2017

Area of reinforcement As m2 N 1.02As 0.02μX Holický 2009

Steenbergen & Vrouwenvelder 2010;

Permanent load model uncertainty θG – N 1 0.07μX
Lenner & Sýkora 2017

JSCC 2001; Holický & Sýkora 2012;

Traffic load model uncertainty θQ – LN 1 0.1μX
fib Bulletin 80 2016

Resistance model Model 1 θR – LN 1 0.06μX fib Bulletin 80 2016

uncertainty Model 2 θR – LN 1.1 0.1μX Sýkora et al 2015

Note: abbreviations for distributions: N (normal), LN (lognormal), Det (deterministic)

plots. A shared characteristic between these
load cases is that the concurrent traffic
load effects in the northbound slow lane
(Nb1) have the largest contribution to the
combined monthly maxima. The results for
these load cases suggest that the combined
monthly maxima imposed on the critical
spine beam may originate from a mixture of
loading events. For example, the combined
monthly maxima may include side-by-side
truck events together with single or following
truck events. The 50-year maximum load
effect distribution can again be obtained
by Equation 3, where F(x) is the probability
distribution function of the LN3 distribution
describing the combined monthly maxima.
Formulation of limit state function for
the critical element reliability analysis
The limit state function in Equation 5 is
used to determine the reliability perfor-
mance of the bending moment capacity of
the critical spine beam at ULS. The func-
tion consists of independent basic variables
that describe the resistance, permanent
load, traffic load effects and model uncer-
tainties relevant to the critical spine beam.
f y As
Z=θ
 R fy As 1 – d – θG(G + Gw)
1.34fcubd
– θQQ(5)
Table 6 defines the basic variables pre-
sented in Equation 5 with respective
probabilistic models. Due to ambiguities
in model uncertainty models presented by
various researchers, two different models different load cases to find the governing
of uncertainty for the resistance are further load case for the critical member under
investigated in order to assess the sensitiv- consideration. For future studies, it is rec-
ity of their parameters. ommended to identify and separate different
loading event types. By fitting distributions
Reliability analysis results and discussion to the individual loading event types, the
Reliability analysis of the critical element accuracy of the probabilistic models should
is again performed using FORM. Overall improve. This will also lend further sup-
reliability indices β obtained for each load port to the assumption of independent
case depend on the selected model of the identically distributed extremes of traffic
resistance model uncertainty θR. The result- load effects. The results also show that the
ing values for β for each load case are pre- coefficient of variation of θR influences β,
sented in Table 7. For Model 1, the obtained which agrees with literature (Allaix 2007;
β-values range between 6.2 and 7.8, whereas Holický et al 2010).
for Model 2, the values range between When comparing the results to target
6.2 and 7.0. The lowest β is obtained for reliability indices βt of 3.8 and 3.0, the
Load Case 8, when three traffic lanes are
positioned on the bridge and the slow lane Table 7 R
 eliability indices β, corresponding to a
in the southbound direction has the largest 50-year reference period, provided for
loading contribution on the critical spine each load case depending on the model
beam. It is seen that the addition of another of the resistance model uncertainty θR
traffic lane is not as influential as in the RCS
β
due to its reduced influence according to the Load Case
transverse stiffness. Model 1 Model 2
The range of β-values is caused by con- 1 7.1 6.6
siderably varying statistical moments for Q
amongst the load cases under consideration. 2 6.6 6.5
The higher the mean, coefficient of variation 3 6.3 6.4
and skewness of the traffic load effects, the
4 7.6 6.8
lower the obtained β-values are. Depending
on whether the traffic moves in the north- 5 7.8 7.0
bound or southbound direction and whether
6 7.8 6.9
a slow lane or a fast lane is located above the
critical spine beam, the moment parameters 7 7.3 6.7
describing the traffic load effects vary. The
8 6.2 6.2
results show the importance of investigating

Journal of the South African Institution of Civil Engineering Volume 63 Number 1 March 2021 33

90
80
70
Relative importance α2 (%)
60
50
40
30
20
10
0 for sagging moments, they need to be
Model 1 Model 2
Load case 2
θR 10.24 43.56
fy 9.61 14.44
Q 65.61 20.25
θQ 10.24 9.61
 elative importance α2 of the resistance model uncertainty θR, yield strength f y , traffic
Figure 11 R
load (50 years) Q and traffic load model uncertainty θQ for Load Cases 2, 3 and 8
obtained β-values are significantly higher.
This observation shows that TMH7 NA
load model exhibits satisfactory reliability
performance for the considered bending
capacity of the critical elements of a 20 m
twin spine deck but might be deemed
uneconomic. It is, however, important to
consider shear forces as well. Yet, according
to Tabsh and Nowak (1991), and Matos et
al (2019), uneconomically high β-values
(larger than 5–6) could be expected for
structures in a good condition. The high
β-levels are also in agreement with the
results of a previous study by Teichgräber
et al (2019) who found many sources of
“hidden safety” in the traffic-load model
LM1 in EN 1991-2, leading to ~25% over­
design. Future research should investigate
other bridge deck types for a 20 m span to
determine whether the reliability results
concur with the results of the twin spine
deck. In addition, it is also recommended
to investigate shear as a failure mode (see
the Discussion further down).
Sensitivity analysis results and discussion
The FORM analysis also provides the
sensitivity factors α for all basic variables.
The factor indicates the influence of a vari-
able on the obtained β. As |α| numerically
increases, so does the significance of the
basic variable. Note that ∑α 2 = 1, when
34
Model 1 Model 2 Model 1 Model 2
Load case 3 Load case 8
6.25 19.36 8.41 27.04
5.76 6.25 7.84 9.00
77.44 64.00 70.56 49.00
7.29 6.76 9.00 9.00
all the basic variables in the limit state
function are considered in the summation.
Thus, α 2 describes the relative importance
of each basic variable. The primary focus
of the sensitivity analysis is Load Cases
2, 3 and 8, where the lowest β-values are
obtained. Variables Q, θR, f y and θQ show
the largest significance, with Q and θR
having the highest α-values. Considering
the two models for θR, Figure 8 provides
the relative importance α 2 of the most
significant variables for Load Cases 2, 3
and 8. From Model 1 to Model 2, the coef-
ficient of variation of θR increases and, in
response, the relative importance of θR
also increases, whereas Q becomes less
significant (Figure 11). This would justify
considering a reduced value of the sensitiv-
ity factor for traffic load effect, well below
|0.7|. However, as a next step of future
research it is recommended to verify and
improve the theoretical probabilistic mod-
els used for the basic variables, especially
θR. Actual data would improve the uncer-
tainty quantification of a variable, which in
turn would improve the accuracy of α 2 and
the obtained β.
Discussion
This study is limited to the investigation
of static effects only. If DAF were included,
Volume 63 Number 1 March 2021
β-levels would mostly increase, as the typi-
cally conservative nominal value of DAF
is considered in design (Van der Spuy et
al 2019c), while measurements typically
suggest very low dynamic amplification
due to heavy traffic (besides effects on
local or short-span members). DAF com-
monly exhibits an inverse proportionality
between the dynamic amplification and
vehicle weight and a number of lanes, and
a reduction in its scatter with increasing
weight and number of lanes (Lenner &
Sýkora 2016; Ivanković et al 2019; O’Brien
et al 2009; O’Connor & Enevoldsen 2007).
As the results of this study are presented
interpreted with caution for other types of
load effects:
QQ Similar results are expected for hogging
moments.
QQ For shear, particularly at shorter spans
and local verifications, a detailed analy-
sis of axle loads is needed. This would
likely result in a lower reliability index
(Van der Spuy 2020).
Reliability of steel bridges is commonly
dominated by fatigue, and modelling of
traffic load extremes becomes less impor-
tant; an example of fatigue verification of a
steel bridge using WIM was provided in a
recent study (Ma et al 2018).
Conclusions and
recommendations
The Roosboom Case Study, considering
sagging moments as dominating traffic
load effects with an assumed sensitivity
factor αE = –0.7, indicates that NA loading
generally performs well for spans ranging
from 15 to 50 metres in length. However,
a poor reliability performance is seen for
short-span narrow bridges, especially for
5 m and 10 m spans where the number
of traffic lanes is equal to the number of
notional lanes for NA loading. The find-
ings agree with Oosthuizen et al (1991),
Anderson (2006), and Van der Spuy and
Lenner (2019a), who also found deficien-
cies in the NA loading for short-span and
narrow bridges. The generally acceptable
performance of the rest of the bridges
can be contributed to the relatively high
partial factor, but mostly to the nature of
the notional lane definition of the NA load
model. Owing to the geometry of a typical
highway bridge, more notional lanes of the
NA model are used for design than the
bridge can physically carry. The Roosboom
case study identifies deficiencies in TMH7
Journal of the South African Institution of Civil Engineering
for normal traffic conditions, and it also
identifies span lengths that achieve a high
reliability performance when considering
sagging moments. This implies that the
design load model could be optimised to be
more cost-efficient for a range of bridges.
To study a reliability performance of a
critical element, the Kilner Park case study
entailed investigation of a single spine of
a 20 m twin-spine deck. The results agree
with the Roosboom case study and satisfac-
tory performance is observed. In fact, the
obtained β-values range between 6.2 and
7.8, and far exceed the target levels βt of
3.8 and 3.0 as required by the Eurocode
and the SANS codes, respectively. A revi-
sion of the load model could therefore lead
to potential cost savings. To extend the
background information for such revision,
it is recommended to verify the findings
of this study by studies based on WIM
data from other heavy-freight routes in
South Africa. It should be further verified
whether similar conclusions can be drawn
for hogging moments as expected. For
shear, particularly at shorter spans and
local verifications, the detailed analysis of
axle loads is needed and lower reliability
indices may be obtained.
ACKNOWLEDGEMENTS
This study has been supported by the
Wilhelm Frank Trust in South Africa, and
partly by the Ministry of Education, Youth
and Sports of the Czech Republic under
Grant LTT18003, and by the Czech Science
Foundation under Grant 20-01781S. The
authors thank the N3TC for access to the
Roosboom WIM data. and Bakwena conces-
sion for use of the Kilner Park WIM data.
REFERENCES
Allaix, D L 2007. Bridge reliability analysis with an
up-to-date traffic load model. PhD Thesis. Turin,
Italy: Politecnico di Torino.
Anderson, J R 2006. Review of South African live load
models for traffic loading on bridge and culvert
structures using Weigh-in-Motion (WIM) data.
MEng Dissertation. University of Cape Town.
Ang A & Tang, W 2006. Probability Concepts in
Engineering: Emphasis on Applications to Civil
and Environmental Engineering. 2nd ed. New York:
Wiley.
Basson, S E 2020. An investigation into the reliability
performance of bridges designed according to TMH7.
MEng Dissertation. Stellenbosch University.
Basson, S E & Lenner, R 2019. Reliability verification of
bridges designed according to TMH7. Proceedings,
7th International Conference on Structural
Engineering, Mechanics and Computation, Cape
Town, pp 1865–1870.
Bruls, A, Croce, P & Sanpaolesi, L 1996. ENV1991.
Part 3: Traffic loads on bridges: Calibration of
load models for road bridges. Proceedings, IABSE
Colloquium on the Basis of Design and Actions
on Structures. Delft, The Netherlands: IABSE,
pp 439–454.
Caprani, C C 2012. Calibration of a congestion
load model for highway bridges using traffic
microsimulation. Structural Engineering
International, 22(3): 342–348.
Caprani, C C, Belay, A & O’Connor, A J 2003.
Site-specific probabilistic load modelling for
bridge reliability. Proceedings, 3rd International
Conference on Current and Future Trends in
Bridge Design, 28–30 April 2003, Shanghai, China,
pp 341–348.
Caprani, C C, O’Brien, E J & McLachlan, G J 2008.
Characteristic traffic load effects from a mixture
of loading events on short to medium span bridges.
Structural Safety, 30(5): 394–404.
Castillo, E 1988. Extreme Value Theory in Engineering.
Boston, MA: Academic Press.
Castillo, E, Hadi, A S, Balakrishnan, N & Sarabia, J M
2005. Extreme Value and Related Models with
Applications in Engineering and Science. New York:
Wiley.
Coles, S 2001. An Introduction to Statistical Modelling
of Extreme Values. London: Springer.
Crespo-Minguillón, C & Casas, J R 1997. A
comprehensive traffic load model for bridge safety
checking. Structural Safety, 19(4): 339–359.
CSRA (Committee of State Road Authorities) 1998.
TMH7. Code of Practice for the Design of Highway
Bridges and Culverts in South Africa. Parts 1 and 2.
Pretoria: Department of Transport.
De Wet, D P G 2010. Post-calibration and quality
management of weigh-in-motion traffic data. MEng
Dissertation. Stellenbosch University.
DOT (Department of Transport) 2013. National
Transport Master Plan (NATMAP) 2050. Chapter 7:
Freight Transport. Pretoria: DOT.
EN (European Standard) 2002. EN 1990:2002.
Eurocode – Basis of Structural Design. Brussels:
European Committee for Standardisation.
Enright, B 2010. Simulation of traffic loading on
highway bridges. PhD Thesis. Dublin, Ireland:
University College Dublin.
fib Bulletin 80 2016. Partial factor methods for
existing concrete structures. Lausanne, Switzerland:
International Federation for Structural Concrete
(fib).
Fisher, R A 2006. Statistical Methods for Research
Workers. Edinburgh: Genesis Publishing.
Getachew, A & O’Brien, E J 2007. Simplified site-
specific traffic load models for bridge assessment.
Structure and Infrastructure Engineering,
3(4): 303–311.
Hajializadeh, D, O’Brien, E J, Enright, B, Caprani,
C C, Sheils, E & Wilson, S P 2012. Probabilistic
study of lifetime load effect distribution of bridges.
Proceedings, 6th International ASRANet Conference
for Integrating Structural Analysis, Risk and
Reliability, 2–4 July 2012, London.
Holický, M 2009. Reliability Analysis for Structural
Design. Stellenbosch: SunMedia.
Holický, M 2013. Introduction to Probability and
Statistics for Engineers. Berlin, Heidelberg: Springer.
Holický, M, Retief, J & Wium, J 2010. Partial factors for
selected reinforced concrete members: Background
to a revision of SANS10100-1. Journal of the South
African Institution of Civil Engineering, 52(1): 37–44.
Holický, M & Sýkora, M 2012. Conventional
probabilistic models for calibration of codes.
In Faber, M, Koehler, J & Nishijima, K (Eds).
Applications of Statistics and Probability in
Civil Engineering, Boca Raton, FL: CRC Press,
pp 969–976.
ISO (International Organisation for Standardisation)
2015. ISO 2394:2015. General Principles on
Reliability for Structures. Geneva, Switzerland: ISO.
Ivanković, A, Skokandić, D, Žnidarič, A & Kreslin, M
2019. Bridge performance indicators based on traffic
load monitoring. Structure and Infrastructure
Engineering, 15(7): 899–911.
Jacob, B & Feypell-de La Beaumelle, V 2010. Improving
truck safety: Potential of weigh-in-motion
technology. IATSS Research, 34(1): 9–15.
Jacob, B, O’Brien, E J & Jehaes, S 2002. COS 323.
Weigh-in-motion of road vehicles. Final report. Paris:
Laboratoire Central des Ponts et Chaussées.
JCSS (Joint Committee on Structural Safety)
2001. Probabilistic Model Codes. JCSS Working
Document. http://www.jcss.ethz.ch
Lenner, R & Sýkora, M 2016. Partial factors for loads
due to special vehicles on road bridges. Engineering
Structures, 106: 137–146.
Lenner, R & Sýkora, M 2017. Partial factors for
imposed loads in areas for storage and industrial
use. Structure and Infrastructure Engineering,
13(11): 1425–1436.
Lenner, R, De Wet, D P G &Viljoen, C 2017. Traffic
characteristics and bridge loading in South Africa.
Journal of the South African Institution of Civil
Engineering, 59(4): 34–46.
Ma, R, Xu, S, Wang, D & Chen, A 2018. Vehicle models
for fatigue loading on steel box-girder bridges
based on weigh-in-motion data. Structure and
Infrastructure Engineering, 14(6): 701–713.
Matos, J C, Moreira, V N, Valente, I B, Cruz, P J,
Neves, L C & Galvão, N 2019. Probabilistic-based
assessment of existing steel-concrete composite
bridges: Application to Sousa River Bridge.
Engineering Structures, 181: 95–110.
Morales-Nápoles O & Steenbergen R 2014. Analysis of
axle and vehicle load properties through Bayesian
networks based on weigh-in-motion data. Reliability
Engineering and System Safety, 125: 153-164.

Journal of the South African Institution of Civil Engineering Volume 63 Number 1 March 2021 35
Nowak, A S 1993. Load model for highway bridges.
Structural Safety, 13(1–2): 53–66.
Nowak, A S & Rakoczy, P 2013. WIM-based live load
for bridges. KSCE Journal of Civil Engineering,
17(3): 568–574.
O’Brien, E J, Rattigan, P, González, A, Dowling, J &
Žnidarič, A 2009. Characteristic dynamic traffic
load effects in bridges. Engineering Structures,
31(7): 1607–1612.
O’Brien, E J, Schmidt, F, Hajializadeh, D et al 2015. A
review of probabilistic methods of assessment of
load effects in bridges. Structural Safety, 53: 44–56.
O’Connor, A & Enevoldsen, I 2007. Probability-based
bridge assessment. Proceedings of the Institution of
Civil Engineers – Bridge Engineering, 160: 129–137.
O’Connor, A J & O’Brien, E J 2005. Traffic load
modelling and factors influencing the accuracy
of predicted extremes. Canadian Journal of Civil
Engineering, 32(1): 270–278.
Oosthuizen, A P, Meintjies, C J, Trumpelmann, V,
Peters, D, Ullmann, K K & Oppermann, G H 1991.
TMH7 Part 2: Traffic Loading (1991). Proposed
Substitution of Section 2.6. Pretoria: Committee of
State Road Authorities, Department of Transport.
Pérez, S & Lenner, R 2019. Bridge reduction factors
based on Monte Carlo routine with copulas.
Engineering Structures, 198. doi.org/10.1016/j.
engstruct.2019.109530
R Core Team 2018. R: A language and environment for
statistical computing, version 3.5.0. Vienna, Austria:
R Foundation for Statistical Computing.
36
SANRAL (South African National Roads Agency
Limited) 2010. Table 6000/1 Project document:
Project specifications. Pretoria: SANRAL.
SANRAL (South African National Roads Agency
Limited) 2012. SANRAL Structural Drawings.
Pretoria: SANRAL. https://www.nra.co.za/live/
content.php?Category_ID=206.
SANS (South African National Standard) 2019. SANS
10160-1:2019. Basis of Structural Design and Actions
for Buildings and Industrial Structures. Pretoria:
SABS Standards Division.
Soriano, M, Casas, J R & Ghosn, M 2016. Simplified
probabilistic model for maximum traffic load from
weigh-in-motion data. Structure and Infrastructure
Engineering, 13(4): 454–467.
Steenbergen, R D & Vrouwenvelder, A C 2010.
Safety philosophy for existing structures and
partial factors for traffic loads on bridges. Heron,
55(2): 123–140.
Sýkora, M, Holický, M, Prieto, M & Tanner, P 2015.
Uncertainties in resistance models for sound and
corrosion-damaged RC structures according to EN
1992-1-1. Materials and Structures/Materiaux et
Constructions, 48(10): 3415–3430.
Tabsh, S W & Nowak, A S 1991. Reliability of highway
girder bridges. Journal of Structural Engineering,
117(8): 2372–2388.
Teichgräber, M, Nowak, M, Köhler, J & Straub, D
2019. The effect of traffic load model assumptions
on the reliability of road bridges. In Life-Cycle
Analysis and Assessment in Civil Engineering:
Volume 63 Number 1 March 2021
Towards an Integrated Vision. Proceedings of the
6th International Symposium on Life-Cycle Civil
Engineering (2018), Ghent, Belgium: IALCCE,
pp 2837–2845.
Van der Spuy, P 2020. Derivation of a traffic load
model for the structural design of highway bridges in
South Africa. PhD Thesis. Stellenbosch University.
Van der Spuy, P & Lenner, R 2018. Developing a
new bridge live load model for South Africa.
Proceedings, 9th International Conference on Bridge
Maintenance, Safety and Management (IABMAS),
9–13 July 2018, Melbourne, pp 1405–1410.
Van der Spuy, P & Lenner, R 2019a. Towards a new
bridge live load model for South Africa. Structural
Engineering International, 29(2): 292–298.
Van der Spuy, P, Lenner, R, De Wet, T & Caprani, C C
2019b. Multiple lane reduction factors based on
multiple lane weigh in motion data. Structures,
20: 543–549.
Van der Spuy, P, Lenner, R & Meyer, M 2019c.
Dynamic amplification factor for South African
bridges. Proceedings, 7th International Conference
on Structural Engineering, Mechanics and
Computation, 2–4 September, Cape Town,
pp 1859–1864.
Von Scholten, C, Enevoldsen, I, Arnbjerg-Nielsen, T et
al 2004. Reliability-based classification of the load
carrying capacity of existing bridges. Copenhagen:
Danish Ministry of Transport, Road Directorate.
Journal of the South African Institution of Civil Engineering
The variability in commercial TECHNICAL PAPER
laboratory aggregate Journal of the South African
Institution of Civil Engineering

testing for road construction ISSN 1021-2019

Vol 63 No 1, March 2021, Pages 37–44, Paper 1038

in South Africa DR IZAK BREYTENBACH (Pr Sci Nat) obtained

his PhD in engineering geology at the
University of Pretoria in 2016 and is a
registered professional natural scientist. He
I J Breytenbach, H G Fourie works as an engineering geological and
geotechnical consultant in multiple fields.
He has extensive experience in identifying
and investigating potential new hard rock
Simple data comparisons were undertaken using project data to illustrate the need for duplicate aggregate sources, or investigating existing material sources, specifically
test analyses in road construction material investigations for hard rock aggregate sources. for the supply of road construction aggregate.
Test data was sourced from projects commissioned by the South African National Roads Contact details:
Agency SOC Limited (SANRAL). All test results except the polished stone value were reported PO Box 73478
by commercial geotechnical laboratories that are accredited with the South African National Lynnwood Ridge
Pretoria 0040
Accreditation System (SANAS). Only samples subjected to duplicate test analyses were used, as
South Africa
this method allows direct comparisons. The tests used for comparison included the polished T: +27 12 991 0426
stone value, wet and dry versions of the Aggregate Crushing Value test, wet and dry versions E: izak@soilkraft.co.za
of the 10% Fines Aggregate Crushing Value test, the ethylene glycol-soaked version of the 10%
Fines Aggregate Crushing Value test, and the water absorption test of both the fine (–5.00 mm) GERHARD FOURIE (Pr Tech Eng) has 24
years’ experience in road pavements and
and coarse (+5.00 mm) material fractions.
materials. As Project Manager: Materials
The results show that duplicate testing is firmly justified, as test results reported were and Pavement Engineering at SANRAL
often highly variable, despite all participating laboratories being SANAS-accredited and using (South African National Roads Agency
standardised methods. Further consideration was then given to the additional cost of duplicate Limited) since 2010, his core function is to
provide specialist input on designs for
testing at project stage investigations compared to the cost implications of having erroneously
asphalt, chip seals and pavements,
reported materials failing in service or having good-quality materials rejected. including geotechnical projects, drilling and materials investigations.
Other experience includes materials specifications, construction and
quality control. He serves on several industry working groups in
developing and promoting the latest design standards, specifications and
INTRODUCTION QQ Test or calibration methods and
best practice. He obtained his BTech degree in Civil Engineering at the
Intrusive material investigations are fre- method validation Tshwane University of Technology in 1999 and is registered as a
quently undertaken in South Africa with the QQ Equipment Professional Engineering Technologist at the Engineering Council of
South Africa.
intention of procuring construction gravel QQ Measurement traceability
or aggregate materials for road construc- QQ Sampling Contact details:
Private Bag X19
tion. However, due to challenging economic QQ Handling of test and calibration items
Bellville 7535
conditions globally and locally, project funds Variability of test results increases with South Africa
for comprehensive material investigations inter-laboratory testing due to all the pos- T: +27 21 957 4600
are often constrained or allocated to other sible variables listed above. For this reason, E: fourieg@nra.co.za

aspects of a road construction project. This laboratories are required to participate

is unfortunate, as the materials used for
road construction literally form the founda-
tion of such construction projects. With the
limited funds, it is even more difficult to
motivate that material investigation labora-
tory tests should be performed in duplicate
and verified at an additional cost.
The South African National Standard
(SANS 17025 2018) – which specifies
general requirements for competence of
testing and calibration for laboratories –
recognises that many factors may influence
the correctness and reliability of tests per-
formed. Hence, the specification addresses
each of the following factors separately:
QQ Human influencing factors
QQ Accommodation and environmental
factors
in inter-laboratory testing or proficiency
testing schemes. SANAS (2016a; 2016b)
requires that accredited laboratories should
participate in proficiency testing schemes
covering the scope of accreditation over a
period not exceeding five years.
The issue of duplicate testing has
become hotly contested, with many argu-
ing that laboratory test results should
be accepted as issued, because SANAS
accreditation for the commercial labo-
ratories performing the test analyses is
mandatory (as specified on project tender
requirements). Those in favour of duplicate
testing counter that unverified test results
come with inherent risk, and that the cost
of duplicate testing is negligible compared Keywords: hard rock aggregate, aggregate crushing value, 10%
to the cost implications of a problematic fines aggregate crushing value, polished stone value

Breytenbach IJ, Fourie HG. The variability in commercial laboratory aggregate testing for road construction in South Africa.
J. S. Afr. Inst. Civ. Eng. 2021:63(1), Art. #1038, 8 pages. http://dx.doi.org/10.17159/2309-8775/2021/v63n1a4 37
aggregate material which prematurely fails
in service.
Properties of soil – and in this case
aggregate – materials are affected by a num-
ber of aspects, including the inherent nature
of the material itself, sampling disturbances,
a measurement error and statistical uncer-
tainty associated with the number of tests
performed (Juang et al 2019). Jacobsz and
Day (2008) highlighted the variability in the
results of Atterberg Limits reported on the
same sample by different commercial geo-
technical laboratories in South Africa and
the implications of using erroneous results
in geotechnical design. A similar exercise
was undertaken by Theron et al (2015) who
considered the results of hydrometer tests
performed by multiple laboratories on the
same sample materials – the outcome of the
comparison was described as troublesome.
Another example is given by Takács and
Mahler (2011) who compared results pro-
duced by 63 laboratories in Hungary, testing
more than 500 samples on eight separate
sample groups.
However, while literature on the reli-
ability or reproducibility of soil materials is
fairly readily available, similar discussions
on aggregate material tests were found
to be extremely limited, especially in the
South African environment, and more
specifically for South African aggregate
tests. Moreover, literature related to
aggregate tests mainly focus on concrete
production – for example Grieve (2002) –
and not on road construction aggregate,
despite some overlapping tests between the
two applications. Dumas (2000) compiled
a proposed appendix to the Technical
Methods for Highways (TMH1) series
widely used in South Africa. The document
comprehensively considers the reliability
of test results and, amongst other aspects,
takes into account the statistical signifi-
cance of test results and their variability.
The TMH1 series of tests was replaced
by SANS 3001, which includes a series
on the determination of uncertainty of
measurement, repeatability, reproducibility
and bias, i.e. SANS 3001-PR1 (SANS 2010),
as well as an assessment of repeat, check
or duplicate tests, i.e. SANS 3001-PR2
(SANS 2011). The latter standard provides
guidelines to assess duplicate tests by cal-
culating the permissible range between the
primary and duplicate test results.
The approach followed in this compari-
son differs from that followed by Jacobsz
and Day (2008) and Theron et al (2015)
in that where these authors submitted the
38
same sample to multiple laboratories for
comparison, this study considers multiple
samples tested in duplicate only, and as
such a detailed statistical analysis on vari-
ance between different laboratories’ results
is not possible due to limited data. Instead,
the discussion intends to highlight the high
degree of variation in test results on indi-
vidual samples tested in duplicate, despite
participating laboratories being SANAS-
accredited for the tests performed, with the
exception of the polished stone value (PSV)
for which none of the laboratories had
SANAS accreditation.
METHODOLOGY
Data was captured from existing projects
commissioned by SANRAL tasked with
identifying and proving sources of suitable
hard rock aggregate materials to be quar-
ried for various purposes (e.g. base course,
surfacing stone, etc). During the course of
the projects, duplicate tests were performed
on selected samples, and only certain
critical tests or samples were duplicated.
Samples were produced from drill core
retrieved during intrusive drilling, which
was grouped into batched samples and
crushed using a laboratory crusher. Test
duplication was to be done by splitting the
nominated sample after sample prepara-
tion. One half of the sample was tested by
the primary laboratory (i.e. primary tests)
while the second half was sent to a second,
independent laboratory (i.e. duplicate tests)
to perform the same test, using the same
standardised test method. The reasoning
behind this approach was that by splitting
the sample, the test material should be
practically identical and therefore compar-
ing test results directly should be possible.
Preference for duplicate testing was
given to tests which have lower or upper
limit specifications in COLTO (1998) and/
Table 1 Number of data sets
Geological material
Test
Dolerite
PSV 15
ACV (dry) 15
ACV (wet) 15
10% FACT (dry) 15
10% FACT (wet) 15
10% FACT (ethylene glycol) 15
Water Absorption (+5.00 mm) 15
Water Absorption (–5.00 mm) 15
Volume 63 Number 1 March 2021
or SANRAL (2013). After reviewing the
available project data, which included
duplicate test values, the following tests
were selected for comparison:
QQ Polished Stone Value (SANS 5848)
QQ Aggregate Crushing Value (SANS 3001:
AG10) – dry test
QQ Aggregate Crushing Value (SANS 3001:
AG10) – wet test
QQ 10% Fines Aggregate Crushing Value
(SANS 3001: AG10) – dry test
QQ 10% Fines Aggregate Crushing Value
(SANS 3001: AG10) – wet test
QQ 10% Fines Aggregate Crushing Value
(SANS 3001: AG15) – ethylene glycol-
soaked test (basic crystalline materials
only)
QQ Water Absorption of the +5.00 mm
sample fraction (SANS 3001: AG20)
QQ Water Absorption of the –5.00 mm
sample fraction (SANS 3001: AG21)
Material Sample Selection
Following the identification of preferred test
methods, suitable project result data had
to be selected. The selected test data was
sourced from fourteen project sites on vari-
ous geological bedrock types. The tests were
performed by a total of seven different com-
mercial laboratories, all of which are SANAS-
accredited for the tests under consideration,
except the PSV test. The geological composi-
tion of the data set is summarised in Table 1.
Statistical Analyses
Data sets were entered into a spreadsheet
and each material test was used for a sepa-
rate analysis. The test result data for each
test method was used to perform a general
statistical analysis (i.e. descriptive statistics)
which was supplemented by a simple cor-
relation analysis between the primary and
duplicate test results in order to determine
the correlation coefficient. The results of
the statistical analyses will not be discussed
Granite/ Total
Tillite Quartzite
Gneiss
17 2 6 40
18 3 6 42
18 3 6 42
18 3 6 42
18 3 6 42
15
17 3 6 41
17 3 6 41
Journal of the South African Institution of Civil Engineering
in depth, but relevant aspects will be high-
lighted in the following sections. 80
Lab 1
75 Lab 2
Direct Simple Comparison Lab 3
In order to directly compare results of 70 Lab 4
primary and duplicate tests, it was decided Lab 5
that the simplest and most direct method 65
would be to illustrate the results graphically.

PSV
Scatter plots were prepared for each type of 60
test used, plotting the sample number on the
55
x-axis and the two test results (i.e. primary
and duplicate) on the y-axis. In this manner 50
the primary and duplicate test results can
be compared easily. The scatter plots were 45
further refined to distinguish between the
seven participating laboratories that supplied 40
0 5 10 15 20 25 30 35 40
results. Results were then further assessed Sample number
according to SANS 3001 PR2 (SANS 2011) in
order to assess the variability in the results. Figure 1 Polished Stone Value comparison

RESULTS 25
Lab 1
23 Lab 3
Polished Stone Value (PSV)
Lab 5
The comparative plot of results for the PSV 21
Lab 6
tests is illustrated in Figure 1. From the
19 Lab 7
graph it is clear that the correlation between
the primary and duplicate test results is 17
ACV (%)

extremely poor. This is reflected by a correla- 15

tion coefficient of –0.19. Lab 4, in particular,
13
reported highly variable results that were
generally strongly contradicted by duplicate 11
testing. Sample 9, for example, returned 9
primary and duplicate test results of 51 and
79, respectively. Results from remaining 7

laboratories were somewhat less variable, but 5

when considering the sensitivity of the PSV 0 5 10 15 20 25 30 35 40
Sample number
test, the results are still considered question-
able. Perhaps the best correlation of results in
this data set is reflected by Lab 2 and Lab 5. Figure 2 A
 ggregate Crushing Value (dry) comparison
However, in the entire data set the primary
and duplicate test results matched only on
25
two instances (i.e. sample 29 and sample 31).
Apart from the poor correlation between 23
test results, the case of sample 30 needs 21
further consideration. For the projects from
which data was sourced, a minimum PSV 19

of 50 was required. In the case of sample 17

ACV (%)

30, the primary test reported a PSV of 51;

15
however, the duplicate test result reported a
PSV of 48. This difference could be pivotal 13
in deciding whether a potential aggregate 11 Lab 1
source will be used for the production of Lab 3
9
surfacing stone or not, as one result satis- Lab 5
fies material specifications and one does 7 Lab 6
not. The conundrum is then which of the Lab 7
5
two results should be regarded as accurate 0 5 10 15 20 25 30 35 40
and how many additional verification tests Sample number
should be undertaken to comprehensively
prove the true result with confidence. Figure 3 A
 ggregate Crushing Value (wet) comparison

Journal of the South African Institution of Civil Engineering Volume 63 Number 1 March 2021 39
 Aggregate Crushing Value (ACV)
500 The comparative scatter plots for the dry
and wet ACV tests are shown in Figures 2
450
and 3, respectively.
400
The results of the dry ACV test had a
correlation coefficient of 0.50, which is a
350 notable improvement compared to the PSV
10% FACT (kN)

results, but still indicate unreliable results.

300 Of the entire data set only sample 3 had a
perfectly matching result (i.e. 11.9%) and
250
samples 31 through 36 tested by Lab 5 and
Lab 1
200 Lab 6 showed reasonably good correla-
Lab 3
Lab 5
tion. However, the same two laboratories
150 Lab 6 also reported the largest discrepancies
Lab 7 (i.e. samples 28 to 30) with the maximum
100 difference being encountered in sample
0 5 10 15 20 25 30 35 40
Sample number
28. The primary test reported an ACV
result of 12.7%, which easily satisfies the
Figure 4 1 0% Fines Aggregate Crushing Value (dry) comparison COLTO (1998) maximum (i.e. 21%) for
surfacing aggregate, while the duplicate
test result reported 22.2 %, which fails
450 the same specification. The duplicate test
Lab 1 result is therefore almost double that of
Lab 3 the primary test result for the same mate-
400
Lab 5
rial sample. Throughout the remainder
Lab 6
350 of the data trends can arguably be seen,
Lab 7
but data is simply too erratic to make
10% FACT (kN)

300 sensible inferences. Applying the principles

of SANS 3001 PR2 (SANS 2011), 13 of
250 the 42 test results (i.e. 31%) exceeded the
permissible range, rendering those results
200 “unacceptable”.
Results of the wet ACV tests were
150 unsurprisingly similar to that of the dry test,
though the correlation coefficient between
100 the primary and duplicate test results was
0 5 10 15 20 25 30 35 40 fractionally lower (i.e. 0.47). The results of
Sample number
Lab 1 and Lab 3 showed marked improve-
ment for samples 1 through 16, compared
Figure 5 1 0% Fines Aggregate Crushing Value (wet) comparison to the corresponding dry test results. In
contrast to this, the conflicting results of
samples 28 to 30 persisted, with sample 29
450
showing a difference of 11.4% in reported
results. This discrepancy – for both the dry
400 and wet test – was ultimately traced back
to an error made by the testing technician.
350 Applying the principles of SANS 3001 PR2
10% FACT (kN)

(SANS 2011), 7 of 42 test results (i.e. 17%)

exceeded the permissible range, rendering
300
those results “unacceptable”.

250 10% Fines Aggregate Crushing

Lab 1 Value (10% FACT)
200 Lab 3 The comparative results of the 10% FACT
Lab 5 dry and wet tests are shown in Figures 4
Lab 6 and 5, respectively, and the two figures
150
0 5 10 15 have corresponding sample test numbers
Sample number plotted on the x-axis. Figure 6 shows a
comparison of the ethylene glycol-soaked
Figure 6 1 0% Fines Aggregate Crushing Value (ethylene glycol) comparison version of the test, albeit with far fewer

40 Volume 63 Number 1 March 2021 Journal of the South African Institution of Civil Engineering
data sets, as only basic crystalline materials
were subjected to this test and hence sam- 1.4
Lab 1
ple numbers for Figure 6 do not correspond
Lab 3
to those of Figures 4 and 5. 1.2
Lab 5
The data for the dry tests shows very Lab 6
poor correlation, with discrepancies 1.0 Lab 7

Water Absorption (%)

between the primary and duplicate test
results of up to 181 kN (i.e. sample 23). 0.8
Sample 23 consisted of dolerite material
and hence the minimum specification limit 0.6
for crushed stone base course of the 10%
FACT (dry) test is 110 kN (COLTO 1998). 0.4
While both the primary result (308 kN)
and the duplicate test result (489 kN) for 0.2
this sample easily satisfy the specified
limit, the fact that the variation between 0
0 5 10 15 20 25 30 35 40
the two results is larger than the actual
Sample number
minimum specification value is disconcert-
ing. Only one sample (i.e. sample 40) had
Figure 7 W
 ater Absorption (+5.00 mm) comparison
an exact match between the primary and
duplicate test results, and overall the cor-
relation between the remaining data sets is
2.0
considered poor at best, with a correlation Lab 1
coefficient of 0.40. Of interest, only one 1.8 Lab 3
result (sample 23) exceeded the permissible Lab 5
1.6
range for 10% FACT when tested in accord- Lab 6
1.4 Lab 7
Water Absorption (%)

ance with SANS 3001 PR2 (SANS 2011).

However, it was concluded that the method
of assessment becomes less sensitive at
higher values, resulting in the results being
considered acceptable despite the poor cor-
relation. The materials assessed here are of
a high quality, with 10% FACT results well
above the required specification, but should
results be notably lower (e.g. at or below
specification values), the analysis would be
notably more sensitive.
Results of the wet 10% FACT tests
exhibited similar properties to those of the
dry tests, though variations were marginal-
ly less severe and the correlation coefficient
improved slightly to 0.58. Applying the
principles of SANS 3001 PR2 (SANS 2011),
4 of the 42 test results (i.e. 10%) exceeded
the permissible range, rendering those
results “unacceptable”.
In contrast to the dry and wet versions
of the 10% FACT test, the ethylene glycol-
soaked tests had a correlation coefficient of
0.82 between the primary and duplicate test
sets. This is a substantial improvement over
the dry and wet versions of the test, but the
results should be interpreted with caution,
as only 15 test sets were available for the
ethylene glycol-soaked version of the test
(compared to 42 test sets for the dry and wet
tests). SANS 3001 AG15 does not include
evaluation formulae for the calculation of
uncertainty and hence SANS 3001 PR2
(SANS 2011) could not be applied.
1.2
1.0
0.8
0.6
0.4
0.2
0
0 5
Figure 8 W
 ater Absorption (–5.00 mm) comparison
Water Absorption
The final comparison is between the water
absorption reported for the coarse fraction
(Figure 7) and the fine fraction (Figure 8) of
the test samples.
The correlation coefficient of the coarse
fraction was recorded as –0.26; however,
the data is overall skewed by the very poor
correlation of results for samples 26 through
40. Lab 6, in particular, produced highly
erratic results compared to the remaining
results which in general showed differences
of between 0.1% and 0.2%. The principles of
SANS 3001 PR2 (SANS 2011) indicate that
19 of 41 test results (i.e. 46%) exceeded the
permissible range, rendering those results
“unacceptable”. It should be noted that
the uncertainty of measurement for water
absorption tests is based on the bulk density
10 15 20 25 30 35 40
Sample number
and apparent density of the material, and
not on the actual water absorption. The
acceptable range in this case was based on
the apparent density for aggregate.
Critically, the erratic results reported
by Lab 6 exceeded the COLTO (1998) limit
for coarse aggregate for asphalt (i.e. 1.0%)
significantly on a number of occasions (i.e.
samples 30, 32, 33, 34 and 40), while the
duplicate test results for the same samples
were all below 0.4% and well within the
maximum specification. This clearly
illustrates a serious problem. Primary and
duplicate test sets reported identical results
on seven occasions and, considering the
fairly simple nature of the test, it would be
expected that this would have happened
more frequently. The origin of the large
discrepancy resulting from results issued by

Journal of the South African Institution of Civil Engineering Volume 63 Number 1 March 2021 41
 PSV ACV Dry (%)
80 24
75 22
Secondary test results

Secondary test results

70 20
65 18 R2 = 0.2447
60 16
55 R2 = 0.0373 14
50 12
45 10
40 8
48 53 58 63 68 73 78 83 8 10 12 14 16
Primary test results Primary test results

ACV Wet (%) 10% FACT Dry (kN)

24 500

22 450
Secondary test results

Secondary test results

20 R2 = 0.2231 400 R2 = 0.1629

18 350

16 300

14 250

12 200

10 150
9 14 19 24 250 300 350 400 450 500
Primary test results Primary test results

10% FACT Wet (kN) 10% FACT EG (kN)

400 450

350
400
Secondary test results

Secondary test results

300 R2 = 0.3411
350
250
300
200

250 R2 = 0.6698
150

100 200
150 200 250 300 350 400 450 200 250 300 350 400 450
Primary test results Primary test results

Water Absorption +5.00 mm (%) Water Absorption –5.00 mm (%)

0.8 1.8
1.6
0.7
Secondary test results

Secondary test results

1.4
0.6
1.2
0.5 1.0 R2 = 0.5919
R2 = 0.0651
0.8
0.4
0.6
0.3
0.4
0.2 0.2
0.2 0.4 0.6 0.8 1.0 1.2 1.4 0.2 0.7 1.2 1.7
Primary test results Primary test results

Figure 9 P
 rimary (x-axis) versus secondary (y-axis) test results

42 Volume 63 Number 1 March 2021 Journal of the South African Institution of Civil Engineering
Lab 6 was ultimately identified to be a devia-
tion from the test method by an inexperi-
enced technician. This in itself highlights
the importance of technician training and
certification to perform specific tests.
Despite the much-improved correlation
coefficient of the water absorption for the
fine fraction data (i.e. 0.79), results still
compare relatively poorly. Primary and
duplicate test results were identical on only
three occasions, while the largest single
difference in results was reported as 0.9%
(i.e. sample 10). Applying the principles of
SANS 3001 PR2 (SANS 2011) once more,
30 of 41 test results (i.e. 73%) exceeded the
permissible range, rendering those results
“unacceptable”.
In this data set seven samples conflicted
with regard to passing or failing the maxi-
mum water absorption of 1.5% allowed
by the COLTO (1998) specification, while
one sample was confirmed to fail by both
primary and duplicate tests. As before, the
degree of variation for such a relatively sim-
ple test is a serious concern. For example,
sample 10 reported a primary result of 1.6%
water absorption while the duplicate result
reported only 0.7%. Though this is an
extreme example, it illustrates the problem
well in that one result is less than half of
the other.
DISCUSSION
In order to further illustrate the difference
between primary and duplicate test results
concisely, Figure 9 shows direct compari-
sons between the test results. From the
regression coefficients (R2) it is clear that
there are large shortcomings in reliability.
The significance of the poor relationship
between the primary and duplicate test
results is not only of theoretical signifi-
cance. One should consider the practical
and financial implications of using only a
single set of unverified and possibly errone-
ous test results. To illustrate the point, the
following two scenarios can be considered:
QQ Scenario 1: A site is investigated as
potential hard rock quarry; however,
unverified test results inaccurately
report lower than true values which
happen to fall below the minimum
material specification. The site is dis-
qualified from use, and the funds and
time utilised investigating the site are
wasted. Even worse, additional funds
and time are required to investigate
and prove a new site, while the original
site was in fact suitable, and the pro-
ject is delayed with contractual and
cost implications.
QQ Scenario 2: A site is investigated as
potential hard rock quarry. The unveri-
fied laboratory analyses inaccurately
overestimate test results and report
them to be within specification when,
in fact, they fall below specification.
The source is subsequently approved
for material production and is used
in construction, but ultimately the
aggregate material fails prematurely in
service because of its inferior quality.
Costs are incurred to identify a new
suitable source of aggregate and the
road must be repaired, or worse, rebuilt
at great expense.
The scenarios above hold substantial prac-
tical and financial risk for virtually all par-
ties involved; however, the second scenario
clearly holds the largest risk as the implica-
tion of premature pavement failure is dire.
Regardless of the scenario, duplicate testing
could be used to identify inaccuracies or
problems with test results which could be
addressed timeously and without incurring
potentially severe financial repercussions
and project delays. Coincidentally, dupli-
cate testing identified at least three serious
discrepancies in project data used for the
analyses reported here and allowed the
source of the discrepancies to be identified
and addressed timeously. The cost of dupli-
cate testing amounts to only a fraction of
the cost of repairing or rebuilding a failed
road section or repeating the investigations
for an alternative material source.
The cost of periodic maintenance (e.g.
resealing) for a single carriageway road
with an assumed width of ten metres
typically ranges from R2.0 million to R3.0
million per kilometre, as summarised by
SANRAL’s Annual Integrated Report for
2019 (SANRAL 2019) and indicated in
Table 2. Similarly, the cost of rehabilitation
or partial reconstruction would be at least
ten times the cost of resurfacing. The total
cost of all duplicate tests considered in this
research comparison amounted to less than
R1.0 million, which is negligible compared
to the high cost of road construction and
maintenance. Therefore, insisting on dupli-
cate tests to ensure reliable test results,
will recoup the costs of the duplicate tests
within the first kilometre of a typical reseal
project if scenarios 1 or 2, as mentioned
earlier, can be avoided.
In cases where large variations occur
between primary and duplicate test results,
additional tests can be performed to deter-
mine – with confidence – which set of test
results is reliable and reproducible. In this
manner, a laboratory can also make sure
that equipment and reporting techniques
are as they should be. This, however, has
practical problems, too, as funds have

Table 2 Typical costs for road maintenance and construction (SANRAL 2019)
Type of road project Frequency Cost

Capital projects

Strengthening: Increases structural capacity of pavement through recycling of existing layers or addition
Every 20 – 25 years R2m – R3m per kilometre
of new granular layers or structural asphalt layers

Improvement: Improves road with unacceptable quality of service. Examples: addition of passing lanes,
Every 20 – 25 years R12m – R18m per kilometre
addition of paved shoulders, improved intersections. Is often combined with rehabilitation of pavement

New facilities: Improve capacity to accommodate traffic and expand road network. Examples: greenfields
roads, upgrades from single to dual carriageway, new bridges, replacement of intersections with Every 20 – 25 years R20m – R40m per kilometre
interchanges

Road maintenance

Routine maintenance: Cleaning drains and culverts, cutting vegetation, repainting road markings,
Ongoing R0.1m a kilometre per year
repairing guard rails and signs, patching and sealing cracks

Periodic maintenance: Scheduled waterproofing of roads by application of surface seals and thin
8 – 12 years R2m – R3m per kilometre
functional asphalt layers

Journal of the South African Institution of Civil Engineering Volume 63 Number 1 March 2021 43
finite limits and verification testing cannot
be conducted on an indefinite number
of samples until satisfactory (i.e. statisti-
cally acceptable) results are produced. It
is further time-consuming and project
timelines do not make provision for
additional testing.
CONCLUSION AND
RECOMMENDATIONS
Actual project data was presented as
proof that confirmation test analyses are
required in order to verify material test
results. While minor discrepancies are
inevitable and test results are practically
unlikely to match precisely under real-
world conditions, variations were often not
acceptable. In this case, data from fourteen
high-quality hard rock quarry projects was
used where duplicate tests were performed.
A simple comparison between specifica-
tion test results clearly illustrated that the
relationship between primary and duplicate
test results is often inconsistent and that
the test results are therefore treated with
scepticism, despite all participating labora-
tories being SANAS-accredited and using
standardised test methods. In this case
study, material sources were of a high qual-
ity and hence the majority of test results
fall well within material specification,
despite frequent large variations. However,
the implications of unreliable test results
will become more severe where materials
are within closer bounds (i.e. borderline
cases) to material specification limits.
It is also disconcerting that technical
signatories of the commercial laboratories
involved with the projects used in this
comparison issued material test results –
which reflected both the primary and
duplicate test sets on one document –
without revisiting results that were clearly
highly variable or questionable.
Based on the findings of this comparison
it is recommended that material test sched-
ules for hard rock quarry material investiga-
tions also make provision for (at least limit-
ed) duplicate testing as a measure of quality
control. The principles of SANS 3001 PR1
(SANS 2010) and SANS PR2 (SANS 2011)
must be applied to determine if the range
between primary and duplicate test
results are within acceptable limits for the
44
respective test. Large discrepancies in test
results must be identified, and where neces-
sary additional tests of the same sample
material must be undertaken to identify
which test result set is in fact correct, and
also to back-trace the origin of the errone-
ous results to establish the root cause. As
specified in SANS 17025 (SANS 17025 2018),
the responsibility rests with the laboratory
to undertake a root cause analysis.
Estimation of uncertainty of measure-
ment for soils and gravels is better defined,
probably due to the high frequency and
volume of testing concerned with these
materials. However, in the case of aggregate
testing, further research is required to
determine the maximum systematic error
to enable laboratories to identify erroneous
test results. This is not possible without
verification testing. Uncertainty of meas-
urement is often underestimated or even
ignored by design engineers, technologists
and technicians when assessing material
test results.
Compared to the cost of repairing or
rebuilding a road section due to premature
failure ascribed to the use of inferior
aggregate material, the additional cost of
duplicate tests is negligible. Seen in a bigger
context, the additional cost of duplicate
test analyses can massively reduce the
financial and practical risk to any road
construction project.
ACKNOWLEDGEMENT
The authors acknowledge the South
African Roads Agency SOC Limited for
permission to use selective project data.
DISCLAIMER
The views expressed by authors are in
their personal capacity and do not reflect
SANRAL’s viewpoint or policy.
REFERENCES
COLTO (Committee of Land Transport Officials)
1998. Standard Specifications for Road and Bridge
Works for State Road Authorities. Halfway House:
South African Institution of Civil Engineering
(SAICE).
Dumas, B (Provincial Administration, Western Cape)
2000. General technical information: Proposed
Volume 63 Number 1 March 2021
document as an appendix to the TMH series.
Unpublished document.
Grieve, G R H 2002. Inter-laboratory proficiency
evaluations between several Gauteng-based
concrete and aggregate testing laboratories. Journal
of the South African Institution of Civil Engineering,
44(1): 31–34.
Jacobsz, S W & Day, P 2008. Are we getting what
we pay for from geotechnical laboratories? Civil
Engineering, 16(4): 8–11.
Juang, C H, Zhang, J, Shen, M & Hu, J 2019.
Probabilistic methods for unified treatment of
geotechnical and geological uncertainties in
a geotechnical analysis. Engineering Geology,
249: 148–161.
SANAS (South African National Accreditation System)
2016a. Regulatory document R-48-04: Proficiency
Testing and other Comparison Programme
Requirements for Calibration Laboratories. Pretoria:
SANAS.
SANAS (South African National Accreditation System)
2016b. Regulatory document R-80-03: Proficiency
Testing and other Comparison Programme
Requirements for Testing and Medical Laboratories
and Blood Transfusion Services. Pretoria: SANAS.
SANRAL (South African National Roads Agency
Limited) 2013. SAPEM 2013. South African
Pavement Engineering Manual. Chapter 4:
Standards. Pretoria: SANRAL.
SANRAL (South African National Roads Agency
Limited) 2019. SANRAL Integrated Report
2018/2019. Volume 1. Pretoria: SANRAL.
SANS (South African National Standard) 2010.
SANS 3001 – PR1 (2010). Part PR1: Determination
of Uncertainty of Measurement, Repeatability,
Reproducibility and Bias. Pretoria: SABS Standards
Division.
SANS (South African National Standard) 2011. SANS
3001 – PR2 (2011). Part PR2: Use and Assessment of
Repeat, Check or Duplicate Tests. Pretoria: SABS.
SANS (South African National Standard) 2018.
SANS 17025:2018. General Requirements for the
Competence of Testing and Calibration Laboratories.
Pretoria: SABS Standards Division.
Takács, A & Mahler, A 2011. Statistical evaluation
of geotechnical laboratory round robin tests in
Hungary. Proceedings, 15th European Conference
on Soil Mechanics and Geotechnical Engineering,
Athens, Greece, Vol 1, pp 293–297.
Theron, E, Badenhorst, W & Stott, P 2015. Duplicate
testing conducted on the input parameters for
the estimation of potential expansiveness of clay.
Proceedings, 16th African Regional Conference on
Soil Mechanics and Geotechnical Engineering, 27
April 2015, Hammamet, Tunisia.
Journal of the South African Institution of Civil Engineering
Review of compatibility TECHNICAL PAPER
between SANS 10400 Journal of the South African
Institution of Civil Engineering
deemed-to-satisfy ISSN 1021-2019
Vol 63 No 1, March 2021, Pages 45–60, Paper 1062

masonry wall provisions DR WIBKE DE VILLIERS is a senior lecturer at

and loading code

Stellenbosch University, where she obtained
her BEng (2006), MSc Eng (2008) and PhD
(2019), the latter in the development of
performance-based specifications for
alternative masonry units. As part of the
Unit of Construction Materials, her research
W I de Villiers, G P A G van Zijl, W P Boshoff
interests lie in alternative and sustainable
construction materials, and their structural and thermal performance. She
has supervised a number of postgraduate students and authored or
South Africa has a housing shortage estimated in excess of 2 million units. This backlog is being co-authored several publications over the past nine years. Prior to this,
she spent two years at Aurecon as a structural design engineer.
addressed predominantly with the construction of 40 m2 low-income, single-storey, detached,
state-subsidised houses built with conventional concrete masonry units, regulated by the Contact details:
Department of Civil Engineering, Stellenbosch University
Application of the National Building Regulations, SANS 10400. However, several developments
Private Bag X1, Matieland 7602, Stellenbosch, South Africa
warrant a reconsideration of SANS 10400 deemed-to-satisfy masonry wall provisions. Two T: +27 21 808 4072, E: wdv@sun.ac.za
critical configurations of single-storey, unreinforced, bonded masonry walls are generated,
based on these deemed-to-satisfy provisions. Subsequently, a simplified micro-scale finite PROF GIDEON VAN ZIJL (Pr Eng) is a professor
element model is used to analyse these configurations under serviceability and ultimate limit in structural engineering at Stellenbosch
University. He obtained his Bachelor’s (1986)
state loading conditions. Characterisation tests of the concrete masonry material, together with
and Master’s (1990) degrees in Civil
numerical fitting to test data and data taken from literature, generate the necessary parametric Engineering from Stellenbosch University,
input. The numerical analyses reveal that in half of the load cases, the walls’ resistances and his PhD (2000) from Delft University in
failed to achieve the design load as required by the South African loading code. A significant the Netherlands. As Director of the Centre for
Development of Sustainable Infrastructure,
shortfall was found for the out-of-plane resistance against the wind load, as well as structural
his research interests are 3D construction printing, structural and
vulnerability under seismic loading due to the geometric layout permitted by the deemed-to- computational mechanics, including the development, characterisation, and
satisfy rules. This indicates that the SANS 10400 provisions for masonry wall panel geometries constitutive and durability modelling of advanced construction materials.
require reconsideration, especially given the recent revision of the South African loading code. Contact details:
Department of Civil Engineering, Stellenbosch University
Private Bag X1, Matieland 7602, Stellenbosch, South Africa
T: +27 21 808 4436, E: gvanzijl@sun.ac.za
INTRODUCTION The NHBRC stipulates general home build-
The South African government has pro- ing technical requirements and guidelines
PROF BILLY BOSHOFF (Pr Eng) is a professor in
vided nearly 3 million subsidised housing in the form of the Home Building Manual civil engineering at the University of Pretoria.
units since 1994 (Department of Human and Guide, which is based on SANS 10400. He was head of the Structural Engineering
Settlements 2017). However, a backlog of Both SANS 10400 and the Home Division at Stellenbosch University for seven
years, and served as the president of the
over 2 million units persists (Sisulu 2016). Building Manual rely on normative refer- Concrete Society of Southern Africa from
Government-subsidised housing units are ence standards for the material-specific 2012 to 2014. He completed his PhD at
typically a stand-alone dwelling of 40 m2 design aspects, which have typically been Stellenbosch University in the field of
fibre-reinforced concrete in 2007, and has published over 100 papers in
(Laubscher 2014), containing a kitchen, prescriptive in nature. For masonry, this
journals and conference proceedings. His research interest is in the field of
living area, two bedrooms and a bathroom reference standard is SANS 10164 The struc- construction materials, but more specifically fibre-reinforced concrete,
(see Figure 1). tural use of masonry (prescriptive-based), concrete cracking in the fresh state, as well as unconventional and
The structural design of housing as well as the recently adopted SANS 51996 eco-friendly construction materials.

in South Africa is regulated by the
Application of the National Building
Regulations, based on the National
Building Regulations and Standards Act of
1977 (RSA 1977). The standard was first
published in 1985 but has since been updat-
ed several times to the current edition
SANS 10400 (SANS 2010a). On a practical
level, all housing construction in South
Africa is regulated by the National Home
Builders Registration Council (NHBRC),
the establishment of which is enshrined
in The Housing Consumers Protection
Measures Act (RSA 1995; NHBRC 2015).
Design of masonry structures (performance-
based). SANS 10400 is performance-based
in nature but contains extensive deemed-to-
satisfy solutions, the typical mixed approach
taken in transitioning from prescriptive to
performance-based regulation.
A number of developments over the
past decade or two warrant a reconsidera-
tion of these deemed-to-satisfy solutions
in SANS 10400, specifically with regard to
masonry walling solutions:
QQ Loading: The South African loading
code SANS 10160 2011 (SANS 2011) has
been revised, in the form of an adaption
Contact details:
Address at the time of preparing this paper:
Department of Civil Engineering, Stellenbosch University
Private Bag X1, Matieland 7602, Stellenbosch, South Africa
Current address:
Faculty of Engineering, Built Environment and Information Technology
University of Pretoria
Private Bag X20, Hatfield 0028, South Africa
T: +27 12 420 2746, E: billy.boshoff@up.ac.za
Keywords: low-income housing, National Building Regulations,
SANS 10400, South African loading code, concrete
masonry, simplified micro-model

De Villiers WI, Van Zijl GPAG, Boshoff WP. Review of compatibility between SANS 10400 deemed-to-satisfy masonry wall provisions and loading code.
J. S. Afr. Inst. Civ. Eng. 2021:63(1), Art. #1062, 16 pages. http://dx.doi.org/10.17159/2309-8775/2021/v63n1a5 45

(a)
Figure 1 T ypical 40 m2 government subsidised concrete masonry house: (a) plan (CMA 2011,
reproduced with the kind permission of the CMA) and (b) under construction
of EN 1990 (EN 1990) and EN 1991 (EN
1991), with notable changes and addi-
tions. The design of single-storey mason-
ry structures must take seismic loading
into account more comprehensively in
certain areas of the country, and signifi-
cant improvements have been made to
the South African wind data map.
QQ Category 1 buildings: The Application
of the National Building Regulations
(SANS 2010b) has been revised with
significant changes, including the intro-
duction of Category 1 buildings, specifi-
cally aimed at drawing more low-income
structures into a regulatory framework.
QQ Adoption of Eurocode 6: The South
African masonry industry is in the
process of updating its suite of masonry
standards to the EN approach, Eurocode
6 Design of Masonry Structures, mark-
ing a transition from prescriptive to
performance-based standards.
QQ Advances in numerical analyses: A
significant amount of research has been
conducted internationally, using finite
element modelling, to better understand
the discontinuous behaviour of masonry
structures, and these advances need to
46
be taken into consideration in the speci-
fications for masonry.
QQ Outdated mechanical limits: Current
mechanical limits set in the South
African prescriptive standards of
conventional masonry units are largely
based on yield line theory analysis (JSD
1995), taken from the withdrawn British
Standard BS 5628-1 (BS 1978).
This paper therefore investigates the
response of conventional concrete masonry
walls in the context of South African low-
income housing (LIH) by means of finite
element (FE) analysis. The analyses are
performed on single-storey, unreinforced,
single-leaf, external masonry walls, which
conform to the deemed-to-satisfy solutions
of SANS 10400. Two critical wall layouts
are identified (W1 and W2), modelled in
DIANA FE analysis software and subjected
to three load conditions as required by SANS
10160: the serviceability limit state (SLS) and
the ultimate limit state for wind (ULS-W)
and seismic (ULS-S) actions. The results of
the analyses provide insight into the in-plane
and out-of-plane structural behaviour of
conventional concrete masonry walls of LIH
housing, relative to the expected loading.
Volume 63 Number 1 March 2021
(b)
FINITE ELEMENT MODEL
Modelling approach
Significant advances in numerical methods
and computational capabilities in recent
decades have altered the way in which
masonry is analysed. For masonry finite
element modelling, two main approaches
have been established, namely macro- and
micro-modelling, with the level of abstrac-
tion directly related to the complexity
and size of the problem to be analysed,
(Giambanco et al 2001; Reyes et al 2008;
Roca et al 2010; Abdulla et al 2017).
Macro-modelling assumes a smeared
continuum approach, where the unit, mortar
and unit-mortar interface behaviours are
combined in a representative continuous
material. In contrast, micro-modelling repre-
sents a high degree of detail where the unit,
mortar and unit-mortar interface are mod-
elled distinctly. Simplified micro-modelling
(SMM) is a subset of micro-modelling as its
name implies, wherein the units are modelled
as expanded elements, with solely elastic
material properties, to encompass the volume
of the unit and the mortar in order to main-
tain the overall geometry. The relatively weak
mortar joint and unit-mortar interface are
combined into a single zero-thickness inter-
face element in which the nonlinear material
behaviour is concentrated (Lourenço 1996).
The SMM approach is considered the most
appropriate for this study in terms of the
wall scale to be analysed and the computing
requirements (Figure 2 refers).
Constituent material model
The constitutive material model chosen is
the Combined Cracking-Shearing-Crushing
(CCSC) model, implemented in DIANA. The
plasticity-based model is defined by a multi-
surface yield function, shown in Figure 3,
consisting of a tension cut-off, a Coulomb
Journal of the South African Institution of Civil Engineering
 hhmm
hhu
u 16-noded
hhm 16-noded
m interface
interface
element
elementfor for
mortar
mortarjoints
joints
Interface
interface Potentialcrack
elements potential crack
elements (joints)
(joints) ininthe
theunit
unit

Zerothickness
zero thickness
+h
h u+h
u mm

20-noded solid
20-noded solid
element for
element for bricks
bricks
Continuum elements (units)
continuum
(a) elements (units) (b)

Figure 2 S MM approach: (a) in 2D with expanded unit elements (adapted from Lourenço & Rots 1997) and (b) in 3D with solid brick elements and
2D interface elements (adapted from Macorini & Izzuddin 2011)

friction criterion and an elliptical compres-

sion cap. The depiction of the composite yield
coulomb
Coulomb
surface in 3D in Figure 3(b) does not include frictionmode
friction mode ||τ|τ |
the elliptical compression cap, but it has been
implemented in DIANA (2017). cap
Capmode
mode
intermediate
Intermediate
yield
yieldsurface
surface
WALL CONFIGURATIONS
In an effort to make buildings that meet tension
Tension mode
regulatory requirements more affordable mode
σσ
to low-income communities and to reduce
the immense health and safety risk that
(a) initial
Initial yield surface residual
Residual yield surface
yield surface yield surface
unregulated informal structures present
(Watermeyer 2004), the Joint Structural ττs τ
s s
Division (JSD) of the South African
Institution of Civil Engineering (SAICE) cco
o
developed a new category of buildings in
2000 (Watermeyer & Milford 2003). This ff t
cco t σσ
Category 1 building type was introduced o
τt
in The Application of the National Building t
Regulations: SANS 10400 in 2004.
These buildings are restricted in size and (b)
ττt
occupancy class. The floor area is limited t
to 80 m2, wall lengths to less than 6 m Figure 3 C
 ombined Cracking-Shearing-Crushing yield criterion: (a) in 2D (adapted from Lourenço
between lateral supports, and the structure 1996) and (b) in 3D (adapted from Van Zijl 2000)
to a single storey with no basement. The
wall thickness can be as little as 90 mm, deemed-to-satisfy masonry wall solutions and openings. These two wall configurations
compared to 140 mm in non-Category 1 in SANS 10400-K (SANS 2011b). Therefore, are detailed in Figure 4, where the hatched
buildings (SANS 2011b). Occupancy classes these solutions are used to identify appropri- areas designate positions of return walls
are restricted to places of instruction or ate wall configurations for the LIH context. which provide lateral support.
worship, small shops, offices, dormitories, Additional recommendations and limitations The selection criteria for these two
domestic residences and dwelling houses. regarding the geometry, roof configuration, representative single-leaf wall layouts are
Category 1 buildings also make allowance energy use and seismic loading, as set out in detailed in Table 1. Although only the walls
for lower performance levels with regard to SANS 10400-A (SANS 2010b), SANS 10400‑L are modelled, the walls are set within the
size and serviceability limits, but no distinc- (SANS 2011c), SANS 10400‑XA (SANS 2011d) context of representative 40 m2 Category 1
tion is made between the categories regard- and SANS 10160-4 (SANS 2017) respectively, houses to derive loading and support con-
ing seismic and wind loading. are also taken into consideration. Two differ- ditions. The roof construction is assumed
ent wall configurations, panel wall W1 and to be timber, with metal sheet covering.
Deemed-to-satisfy requirements gable wall W2, are derived by selecting the Typical reinforcement according to
Society’s expectation of a wall’s perfor- most extreme and critical combinations of SANS 10400-K (SANS 2011b) of 5.6 mm
mance is implicitly represented by the distance between lateral supports, wall height steel rods and 2.8 mm brickforce (Figure 5)

Journal of the South African Institution of Civil Engineering Volume 63 Number 1 March 2021 47
 6 000 mm

Secondary: 0 5.6 mm × 2

756 mm
Primary: 0 5.6 mm × 3
(equivalent of)

1 008 mm
2 646 mm

Brickforce: 0 2.8 mm × 2

Brickforce: 0 2.8 mm × 2

882 mm
Brickforce: 0 2.8 mm × 2
750 mm 900 mm 600 mm 3 000 mm 750 mm

4 950 mm Primary: 0 5.6 mm × 3

630 mm 630 mm
Secondary: 0 5.6 mm × 2
Primary: 0 5.6 mm × 2

1 010 mm
(equivalent of)
2 650 mm

Brickforce: 0 2.8 mm × 2
1 010 mm

Brickforce: 0 2.8 mm × 2

300 mm 900 mm 450 mm 3 000 mm 300 mm

Primary: 0 5.6 mm × 2
Figure 4 Wall W1 (top) and W2 (bottom) layout and dimensions

Table 1 Selection criteria for representative houses and wall layouts

Layout Selection Clause Standard Figure 5 B
 ed joint reinforcement above
openings for wall W1 (top) and W2
Wall effective
W1 140 mm* B.3.3 (a) SANS 10160-4 (2017) (bottom)
thickness
3.6 (c) SANS 10400-A (2010b)
W1 6.0 m is modelled in the bed joints above the open-
Table 1, Panel C SANS 10400-K (2011b)
Wall length ings, instead of concrete lintels. The rod
3.6 (c) SANS 10400-A (2010b) reinforcement yield strength is taken as the
W2 5.0 m
Tables 5 & 6 SANS 10400-K (2011b) required proof stress of rod reinforcement
Table 1, Panel C SANS 10400-K (2011b) by SANS 10400-K (SANS 2011b), namely
W1 2.7 m** 485 N/‌mm2. Whilst brickforce proof stress
B.3.3 (b) SANS 10160-4 (2017)
Wall height is not specified in SANS 10400, tensile tests
Fig 4 SANS 10400-K (2011b) conducted by Talocchino (2005) on typical
W2 2.6 m**
B.3.3 (b) SANS 10160-4 (2017) South African brickforce found a proof yield
Roof slope 15° 4.2.2.1 SANS 10400-L (2011c) stress of 500 N/mm2. Therefore, the brick-
Truss spacing 1.2 m Table 4 SANS 10400-L (2011c)
force yield strength is taken as 485 N/mm2 as
well. The elastic modulus of both reinforce-
Fig 6(a), Table 7 SANS 10400-K (2011b)
ment types is taken as 200 000 N/‌mm2.
Openings Various 6.2.2 SANS 10160-4 (2017)
4.4.4 SANS 10400-XA (2011d) Boundary conditions
5.6 mm rods Tables 20, 21 & 23, Fig 27 SANS 10400-K (2011b) In modelling the support conditions of the
Reinforcement wall models, a number of assumptions need
2.8 mm brickforce B.3.3 (d), (f) SANS 10160-4 (2017)
to be made. The foundations are modelled
Vertical control joint none Table 19 SANS 10400-K (2011b)
as fixed, as depicted in Figure 6, hence fully
* shear wall teff requirement of 190 mm is not met (SANS 10160-4 (2017) B.3.3 (a)) supported with no potential for differential
** shear wall heff /teff < 17 requirement is not met (SANS 10160-4 (2017) B.3.3 (b))
settlement. Pin supports on the short return

48 Volume 63 Number 1 March 2021 Journal of the South African Institution of Civil Engineering
Figure 6 Boundary conditions for wall W1 (left) and W2 (right), inner perspective

walls provide lateral support to the modelled Table 2 Input parameters for wall numerical analyses
walls, whilst allowing for some rotation of Parameter Symbol DIANA Method Value Unit
these joints. It is assumed that the roof truss
system does not provide substantial lateral Unit

load transfer, based on the type of connec- Density ρu – EXP 2 090 kg/m3
tion and the typical poor quality of the con- E-modulus Eu YOUNG EXP 17 700 N/mm2
nection between the roof and walling in LIH.
Therefore the roof line is modelled as unsup- Poisson’s ratio νu POISSON LIT 0.16 –

ported laterally for both wall configurations. Crack interface

Tensile strength f t,c TENSTR FEA 0.66 N/mm2

MATERIAL INPUT PARAMETERS Mode I fracture energy GIf,c GF EXP 0.047 N/mm

Cohesion cc COHESI LIT 1.0 N/mm2

Material characterisation Friction angle φc PHI LIT 37 °
Solid conventional concrete masonry units
form the basis of the masonry analysed, Dilatancy coefficient ψc PSI LIT 0 °

together with 10 mm mortar joints. The Mode II fracture energy GIIf,c MO2VAL LIT 0.47 N/mm
blocks have a length of 290 mm, width of
Compressive strength fc,c COMSTR EXP 12.1 N/mm2
140 mm and height of 116 mm. Extensive
material parameter characterisation is Shear traction contrib Css,c CS LIT 1.0x10 -3 –
required for the selected constituent mate- Compr fracture energy Gc,c GC LIT 10.0 N/mm
rial model, CCSC, that was selected in
Eq plastic relative displ κp,c DUPEAK LIT 0.030 mm/mm
DIANA. Table 2 contains the material input
parameters that were used to define the Tangential stiffness ks,c DSSX/Y LIT 763x10 -3 N/mm3
concrete masonry, as well as the method by Normal stiffness kn,c DSNZ LIT 1770x10 -3 N/mm3
which they were determined. Experiments
Joint interface
were conducted by Fourie (2017) to deter-
mine several of the unit, crack and joint Tensile strength f t,j TENSTR LIT 0.12 N/mm2
interface parameters, indicated by EXP Mode I fracture energy GIf,j GF LIT 0.005 N/mm
in Table 2 under Method. Finite element
Cohesion cj COHESI EXP 0.17 N/mm2
analysis (FEA) was used to determine the
tensile strength of the crack interface and Friction angle φj PHI EXP 49.5 °
the compressive fracture energy and equiva- Dilatancy coefficient ψj PSI LIT 0 °
lent plastic relative displacement of the
Mode II fracture energy GII f,j MO2VAL LIT 0.05-0.08σ N/mm
joint interface by numerically fitting data to
experimental data. Suitable literary sources Compressive strength fc,j COMSTR EXP 5.5 N/mm2
(LIT) were used to determine the remain- Shear traction contrib Css,j CS LIT 0.7 –
ing parameters. For further details on the
Compr fracture energy Gc,j GC FEA 18.0 N/mm
material input parameters or the process to
obtain them see De Villiers et al (2018). Eq plastic relative displ κp,j DUPEAK FEA 0.030 mm/mm
Characteristic values of the current Tangential stiffness ks,j DSSX/Y LIT 214 N/mm3
material input data cannot be established
Normal stiffness kn,j DSNZ LIT 520 N/mm3
since the data is statistically insufficient.

Journal of the South African Institution of Civil Engineering Volume 63 Number 1 March 2021 49
The use of nominal values is allowed in Table 3 Critical design loads for SLS, ULS-W and ULS-S to SANS 10160
such instances according to SANS 10160-1 Roof OP
(SANS 2018a), and it requires the use of (N/mm2) IP
Self-Weight Wind Zone A Zone B
mean values for the parameters govern-
ing the structural stiffness. The choice of SLS W1–44 –10.1 × 10 –3 43.9 × 10 –3 1.4 × 10 –3 1.1 × 10 –3 24.4 × 10 –3
input parameters, as well as the modelling
SLS W1–40 –10.1 × 10 –3 36.3 × 10 –3 1.2 × 10 –3 0.9 × 10 –3 20.2 × 10 –3
approach, was validated experimentally
in a separate process for both the in-plane SLS W1–36 –10.1 × 10 –3 29.4 × 10 –3 0.9 × 10 –3 0.7 × 10 –3 16.3 × 10 –3

and out-of-plane behaviour by comparing SLS W2–44 – – 2.3 × 10 –3 1.0 × 10 –3 17.6 × 10 –3

numerical results to experimental data
SLS W2–40 – – 1.9 × 10 –3 0.8 × 10 –3 14.5 × 10 –3
of dedicated large- and medium-scale
masonry wall tests (see De Villiers 2019). SLS W2–36 – – 1.5 × 10 –3 0.7 × 10 –3 11.8 × 10 –3

ULS–W W1–44 –9.0 × 10 –3 117.1 × 10 –3 3.7 × 10 –3 2.8 × 10 –3 65.1 × 10 –3

Material partial factors
ULS–W W1–40 –9.0 × 10 –3 96.8 × 10 –3 3.1 × 10 –3 2.3 × 10 –3 53.8 × 10 –3
Material partial factors make a significant
contribution to the design process in limit ULS–W W1–36 –9.0 × 10 –3 78.4 × 10 –3 2.5 × 10 –3 1.9 × 10 –3 43.6 × 10 –3
states design. The main considerations in ULS–W W2–44 – – 6.0 × 10 –3 2.7 × 10 –3 46.9 × 10 –3
both SANS 51996-1-1 (SANS 2018c) and
ULS–W W2–40 – – 5.0 × 10 –3 2.2 × 10 –3 38.7 × 10 –3
SANS 10164-1 (SANS 1989) in determining
partial factors for materials are manufactur- ULS–W W2–36 – – 4.0 × 10 –3 1.8 × 10 –3 31.4 × 10 –3
ing and construction or execution control. ULS–S W1 – – 0.8 × 10 –3 0.8 × 10 –3 53.9 × 10 –3
The greater the certainty regarding the unit
ULS–S W2 – – 1.0 × 10 –3 1.0 × 10 –3 56.4 × 10 –3
and mortar strength and the manner in
which they are combined on site, the greater
the reward is in terms of the load magnitude Table 4 Load combination partial factors according to SANS 10160-1 (SANS 2018a)
that the masonry is allowed to bear. Load case Self-weight Roof imposed Wind Seismic
Given the inadequate guidance for
the execution control classification in SLS 1.0 0.0 0.6 –

Eurocode 6 (Sýkora & Holický 2010), and the
disparity in classifications, a direct compari-
son of the material partial safety factors is
not reasonable. However, in broad terms, the
SANS 51996-1-1 (least conservative) ranges
from 1.5 to 3.0, the UK National Annex to
Eurocode 6 of 2005 (BSI 2005) from 2.3 to
3.0, and SANS 10164-1 (most conservative)
from 2.9 to 3.5.
Despite their importance, material par-
tial factors are not included in this study.
Admittedly, this may result in a less con-
clusive evaluation of the concrete masonry
walls and the design loads applied to them,
which are adjusted with partial factors.
However, including the material partial fac-
tors renders the finite element output less
clear, impacting in particular the fracture
behaviour of the concrete masonry as well
as the investigation into the relative impor-
tance of the parameters.
DESIGN LOADS
SANS 10400-B (SANS 2012) requires
the structural strength and stability of
the structure to be assessed by loading it
with the relevant actions as determined
according to the South African load-
ing code, SANS 10160 (SANS 2011). All
relevant design situations are taken into
consideration to arrive at the most critical
ULS-W 0.9
ULS-S 1.0
combinations of permanent, imposed, wind
and seismic actions for the two wall con-
figurations under consideration. The three
load cases considered are the serviceability
limit state (SLS) and the ultimate limit
states under wind (ULS-W), based on basic
fundamental wind speeds of 44, 40 and
36 m/s respectively, and seismic (ULS-S)
actions. The latter are determined by means
of the equivalent lateral static force method.
Table 3 summarises the factored loads
applied in the numerical analyses for the
vertical loads on the roof and the horizontal
out-of-plane (OP) and in-plane (IP) loads on
the walls, whereas Table 4 details the partial
load factors for the three load cases, accord-
ing to SANS 10160-1 (SANS 2018a).
Load assumptions
The assumptions made in determining
these critical load cases are detailed in the
following three sub-sections according to
self-weight and imposed load, wind load
and seismic load.
Self-weight and imposed load
The self-weight of the walls is based on
the density determined experimentally,
0.0 1.6 –
0.0 – 1.0
as detailed in Table 2. The roof assembly
consists of six bay Howe type trusses,
assuming a timber density of 5 000 N/m 3
according to Table A.4 of SANS 10160-2
(SANS 2011a) for the structural pine,
and 0.5 mm metal sheeting with a self-
weight of 39.5 N/m 2 according to Table
A.5 of SANS 10160-2 (SANS 2011a). The
roof is classified as an inaccessible roof
according to Table 5 of SANS 10160-2
(SANS 2011a) and loads for normal main-
tenance and repair of 400 N/m 2 would be
included. However, since an additional
compressive load on the walls is favour-
able, the load combination nullifies the
roof-imposed load.
Wind load
The loads due to wind actions are deter-
mined according to SANS 10160-3 (SANS
2018b). The pertinent parameters are
summarised in Table 5 and assumptions
discussed thereafter.
In most instances, the parameter result-
ing in the most critical load is selected.
The basic fundamental wind speed is taken
as the highest value for any area in South
Africa of 44 m/s; however, loads based on

50 Volume 63 Number 1 March 2021 Journal of the South African Institution of Civil Engineering
Table 5 Wind load parameters to SANS 10160-3 (SANS 2018b)
Parameter
Fundamental value of basic wind speed
Terrain category
Terrain roughness factor
Topography factor
Air density
Peak wind pressure
Table 6 Seismic load parameters to SANS 10160-4 (SANS 2017)
Parameter
Peak ground acceleration
Ground type
Building importance factor
Reliability redundancy factor
Behaviour factor
Fundamental period of vibration factor
basic fundamental wind speeds of 40 and
36 m/s are also included in the results for
comparative purposes. The terrain cat-
egory is chosen as the most likely scenario
for single-storey residential structures
in a suburban or peri-urban setting. The
default topography factor is chosen, on
the assumption that it is unlikely that low-
income housing is developed on extreme
terrain, which is costly to construct on.
The highest air density value is chosen, to
result in the highest critical load.
Additionally, SANS 10400-B (SANS 2012)
specifies minimum wind pressures to be
applied to housing structural systems of
370 N/m2 and to housing structural elements
of 450 N/m2. The peak wind pressures
determined according to SANS 10160-3
(1 213 N/‌m2) (2018) exceed these minimum
load requirements.
IP
OP Zone A
Figure 7 Load applications for wall W1 (left) and W2 (right) for SLS and ULS-W (N/mm2)
Journal of the South African Institution of Civil Engineering Volume 63 Number 1 March 2021
Building Code (ICBO 1997) and introduced
Symbol Value Clause
in the South African loading code to com-
pensate for a lower nominal peak ground
νb 44 m/s 7.2.2 acceleration of 0.1 g (Wium 2010). However,
- C Table 2 the UBC permits a reliability redundancy
factor range of 1.0 to 1.5, compared to
cr(z) 0.73 7.3.2, Table 3
a range of 1.2 to 1.5 in SANS 10160-4
c0(z) 1 7.3.3 (SANS 2017). The lower limit of 1.2 was
ρ 1.2 kg/m3 Table 4 set to compensate for the higher behaviour
factors for reinforced concrete shear walls
qp(z) 1213 N/m2 7.4, Equation 6
used in the UBC (ICBO 1997) compared to
Eurocode 8 (EN 1998; Wium 2010).
For determining the seismic design
Symbol Value Clause
load, the reliability redundancy factor is
chosen as the lower limit of the allowable
ag 0.15 g 5.2, Figure 1 range (1.2 to 1.5), hence less conservative,
– 4 5.1.2, Figure 2, Table 2 for two reasons. First, a higher peak ground
acceleration of 0.15 g was selected for the
γ1 1.0 7.3, Table 3
analyses, not 0.1 g. Second, the lower limit
ρ 1.2 7.3, Equation 6 of 1.2 in SANS 10160-4 (SANS 2017) was
q 1.5 8.2, Table 4 introduced to compensate for the higher
behaviour factors of reinforced concrete
CT 0.05 8.5.2.1
shear walls. This discrepancy in behaviour
factors is less relevant for this study, given
Seismic load that a consistent behaviour factor for unre-
The loads due to seismic actions are inforced masonry of 1.5 is used. It would
determined according to SANS 10160-4 hence be justifiable to use a reliability
(SANS 2017). The pertinent parameters are redundancy factor of 1.0. However, compli-
summarised in Table 6 and assumptions ance with SANS 10160-4 (SANS 2017)
discussed thereafter. is considered salient and a factor of 1.2
The highest peak ground acceleration is used.
for natural seismicity in South Africa is
selected. The most unfavourable of ground Load application
types is chosen and the selected build- The critical design loads, detailed in
ing importance factor is commensurate Table 3, are applied to the two wall con-
with a typical residential structure. The figurations, as shown in Figure 7 for the
behaviour factor for unreinforced masonry SLS and ULS-W, and in Figure 8 for the
is used, given that minimum detailing and ULS-S. Loads that act out-of-plane (OP) of
reinforcement requirements are adhered the wall, are applied as a uniform distrib-
to. The fundamental period of vibration is uted load over the entire wall, including
chosen based on structural system type. the wind or seismic load, as applicable.
The reliability redundancy factor is not Under wind action, the most critical
present in the parent standard, Eurocode 8 load case is the modelled wall acting as
(EN 1998), but was taken from the Uniform side wall in the context of a 40 m 2 house
Roof wind IP
OP Zone B
OP Zone B
OP Zone A
51
 IP
OP
Figure 8 Load applications for wall W1 (left) and W2 (right) for ULS-S (N/mm2)
structure. The OP wind loads are there-
fore differentiated into Zones A and B,
according to SANS 10160-3 (SANS 2018b).
The total OP force (N) is applied uniform-
ly over the masonry portions of the model
walls to account for the lack of surface
area over the wall openings.
The horizontal in-plane (IP) load is dis-
tributed over the full height and thickness
of the wall. The load arises from the lateral
loads on the walls adjacent to the wall
modelled, and includes either the wind or
seismic load, as applicable. The IP load var-
ies linearly for the wind load case, with the
maximum (presented in Table 3) applied
at the top of the wall. Using the principle
that lateral loads are applied at the location
of the mass, the IP load for the seismic
load case is distributed uniformly over the
height of the wall.
The vertical loads transferred from the
roof to the wall are applied at each truss
support point and distributed over the width
of masonry block to prevent stress concen-
trations at these points. The roof self-weight
and the wind load, if applicable, are included
in this load. Vertical uplift is indicated by a
positive value, whereas a compressive force
is indicated by a negative value. Since W2 is
IP displacement OP deflection
Door
Pier 1 Lintel
Pier 2
Base 2
Figure 9 Crack position and deflection/displacement measurement legend (W1 left, W2 right)
52
IP
a gable wall configuration, which does not
support trusses, the roof load is only applied
to the W1 configuration.
In the past, numerical and experimental
studies on unreinforced masonry walls have
been focussed on the effect of either OP or
IP loading. In more recent research, the sig-
nificance of the interaction of these two load
conditions has gained prominence, but still
focused on masonry infill walls. Few numer-
ical investigations have taken into account
the combined effect of IP and OP actions
on load-bearing unreinforced masonry, and
even fewer experimental studies (Milani
2008; Agnihotri et al 2013; Najafgholipour
et al 2013; Dolatshahi et al 2015). Typically
the findings are that the IP load may have a
crucial effect on the OP capacity of the wall.
The wall slenderness and aspect ratios large-
ly determine the interaction level. Therefore,
this study applies the simultaneous action of
OP and IP loading.
RESULTS
Results overview
Figure 9 provides an interpretation
key for results discussed later. The OP
IP displacement
Pier 3
Door
Pier 4
Pier 2
Base 1 Base 2
Volume 63 Number 1 March 2021
OP
deflections provided in Figures 12 and 17
are measured at the top midspan position,
as indicated in Figure 9. The IP displace-
ments provided in Figures 13 and 18 are
measured at the top left corner of each
respective wall. Typical crack positions
found in the numerical analyses are also
identified in Figure 9, for the later discus-
sion on crack damage classification and
crack width in Figure 14.
Typical failure modes are presented in
Figure 10 for the SLS and ULS-W and in
Figure 11 for the ULS-S. Compressive fail-
ure or crushing is not identified in either of
the wall configurations for both OP and IP
failures. This is not remarkable given the
small structure size and low vertical loads.
The contours in Figure 10 indicate the OP
deflections and OP failure dominates for
both the SLS and ULS-W, since most of the
total load applied is in the lateral direction.
The ‘Base 1’ crack, located in the lowest
bed joint of the wall, indicates tensile fail-
ure, together with shear failure in columns
adjacent to the door openings (‘Door’) for
wall configurations W1 and W2 for both
load cases. Subsequent tensile cracks also
form in the pier adjacent to the window
opening (‘Pier 1’ and ‘Pier 2’).
OP deflection
Pier 3
Pier 1 Lintel
Pier 4
Base 1
Journal of the South African Institution of Civil Engineering
Figure 10 Typical failure for SLS and ULS-W for walls W1 (left) and W2 (right)

The contours in Figure 11 indicate the Table 7 Summary of OP SLS and ULS-W loads and IP ULS-S loads
IP displacements. IP failure dominates for SANS 10160 Design load /
the ULS-S load case, due to shear action Numerical resistance
design load numerical resistance
in the walls bearing most of the seismic OP/
SLS/ULS
IP [N] [N[ [N] [N] [-] [-]
load. Tensile/flexural cracks dominate,
starting with ‘Base 2’ next to the door W1 W2 W1 W2 W1 W2
and ‘Pier 1’ to ‘Pier 3’ surrounding the SLS-44 OP 12 482 11 530 16 086 21 430 0.8 0.5
window opening. Crack onset at ‘Base 2’
SLS-40 OP 10 315 9 529 16 086 21 430 0.6 0.4
is tensile, and progresses to sliding shear,
whereas ‘Pier 4’ is a combination of shear SLS-36 OP 8 356 7 718 16 086 21 430 0.5 0.4
and tensile stepped cracks. The W2 wall
ULS-W-44 OP 33 285 30 746 15 597 16 543 2.1 1.9
configuration is made particularly vulner-
able by the slender column to the left of ULS-W-40 OP 27 508 25 410 15 597 16 543 1.8 1.5
the door opening. ULS-W-36 OP 22 281 20 582 15 597 16 543 1.4 1.2
As an overview, the design loads and
ULS-S IP 19 613 20 539 25 964 11 295 0.8 1.8
numerical wall resistance determined
through the numerical analyses are pre-
sented in Table 7 for the three load cases, three load cases. The South African load- load-carrying capacity of the walls, namely
namely SLS, ULS-W (for three basic funda- ing code (SANS 10160 2011) OP design in W1 ULS-W, W2 ULS-W (both for all
mental wind speeds of 44, 40 and 36 m/s) loads for each of the three load cases are three basic fundamental wind speeds) and
and ULS-S. For each instance of the critical also included to contextualise the results. W2 ULS-S. It is important to note that,
load direction, the ratio of the design In Table 3 the design loads were provided in line with recent findings in literature
load to the numerical wall resistance is in the form of pressures (N/mm2) but are (Vaculik 2012; Derakhshan et al 2018), the
included. Failure is therefore indicated by a converted to forces (N) to facilitate the OP response for the ULS-S is inadequate
ratio of greater than 1.0. comparison of the design loads and the and that OP behaviour of unreinforced
resistance capacities of the walls. masonry cannot be disregarded under
Out-of-plane response Of the six analyses presented in seismic action. SANS 10400-B (SANS 2012)
The OP load/deflection responses for W1 Figure 12, in half of them the OP design imposes a 1:175 deflection limit on such
and W2 are depicted in Figure 12 for the load significantly exceeds the OP building walls, which is well above the

Figure 11 Typical failure for ULS-S for walls W1 (left) and W2 (right)

Journal of the South African Institution of Civil Engineering Volume 63 Number 1 March 2021 53
 35 000
SANS 10160 ULS-W 44 m/s
30 000
SANS 10160 ULS-W 40 m/s
25 000
SANS 10160 ULS-W 36 m/s
W1 OP load (N)
20 000
15 000
SANS 10160 SLS 44 m/s
NUM ULS-S
10 000
SANS 10160 SLS 40 m/s
SANS 10160 SLS 36 m/s
5 000
0 0
0 1
W1 OP deflection (mm)
Figure 12 Out-of-plane response of walls W1 (left) and W2 (right) for SLS, ULS-W and ULS-S
OP deflection range encountered in these
analyses of between 2.5 mm and 7 mm.
The gable of the W2 configuration was
not buttressed, contrary to the specifica-
tions of SANS 10160-4 (SANS 2017). It was
presumed that this lack of lateral support
to the gable would cause instabilities in the
analyses, but this element was noncriti-
cal in the OP loading conditions under
consideration. This is most probably due
to the presence of more vulnerable, slender
elements in the wall in other locations. A
wall with better-proportioned openings
may well cause the gable to become critical,
requiring buttressing.
In-plane response
The IP load/displacement responses for W1
and W2 are depicted in Figure 13 for the
three load cases. The South African load-
ing code (SANS 10160 2011) IP design loads
for each of the three load cases are also
included to contextualise the results.
Reflective of the OP response, in three
of the six analyses presented in Figure 13,
the IP design load significantly exceeds
the IP load-carrying capacity of the walls,
namely in W1 ULS-W, W2 ULS-W (both
for all three basic fundamental wind
speeds) and W2 ULS-S. Notably, the
seismic IP capacity of W1 is 2.3 times
greater than W2. This is explained in part
54
W2 OP load (N)
NUM SLS
NUM ULS-W
SANS 10160 ULS-S
2 3
by the particularly weak geometry derived
for W2.
The IP displacement of both of the
wall configurations is not significant (less
than 1 mm). Arguably, the IP displace-
ment would be greater if the load/dis-
placement path is continued numerically,
but the laborious work of overcoming
the post-peak divergence prevents this
pursuit. There are no limitations specified
for IP displacement in SANS 10400 as
they are for IP deflections. However, the
pronounced reduction in load-carrying
capacity, together with the negligible IP
displacement, is typical of the extremely
brittle behaviour of masonry.
Crack damage classification
The crack damage is classified and pre-
sented in Figure 14 for W1 and W2. Only
the dominant crack for each load case
combination is included for clarity. The
frame of reference for the crack widths
is taken from the damage categories and
maximum crack widths in SANS 10400-B
(SANS 2012) and the South African Home
Building Manual (NHBRC 2015). The
damage categories vary from less than
0.25 mm, classified as negligible, to greater
than 25 mm, classified as very severe.
These classifications were developed by
Watermeyer and Tromp (1992) to serve
Volume 63 Number 1 March 2021
35 000
30 000 SANS 10160 ULS-W 44 m/s
25 000 SANS 10160 ULS-W 40 m/s
NUM SLS
20 000 SANS 10160 ULS-W 36 m/s
NUM ULS-W
15 000
SANS 10160 SLS 44 m/s
10 000 SANS 10160 SLS 40 m/s
SANS 10160 ULS-S
SANS 10160 SLS 36 m/s
5 000
NUM ULS-S
0 1 2 3 4 5 6 7
W2 OP deflection (mm)
as serviceability performance criteria for
masonry walls and were subsequently
included in SANS 10400.
Several typical crack positions were
identified in Figures 9, 10 and 11 for the
SLS and ULS-W and ULS-S load cases
respectively. The most dominant crack
is identified for each of the analyses per-
formed and plotted against the OP loads for
W1 and W2 in Figure 14. Cracks occurred
in a number of typical positions for the
IP and OP-dominant loading conditions,
as illustrated in Figure 10, as well as in
Figure 11 for the ULS-W and in Figure 12
for the ULS-S. For each analysis performed,
the most dominant crack is identified and
plotted against the OP load in Figure 14
(left) for W1 and (right) for W2. For both
W1 and W2, the ‘Base’ cracks as well as
cracks around the window openings in the
‘Piers’ are prolific. Most cracks measured
in the numerical analyses of the concrete
masonry walls fall below the ‘negligible’
(less than 0.25 mm) classification. With
further development of the walls’ post-peak
responses, the cracks would undoubtedly
widen. However, in the numerical analyses
performed, these initial cracks suffice to
demonstrate crack development and to
produce a significant reduction in the load-
carrying capacity, which is typical of the
brittle nature of masonry.
Journal of the South African Institution of Civil Engineering
 30 000 25 000

NUM ULS-S
25 000 SANS 10160 ULS-S
20 000

20 000 SANS 10160 ULS-S

15 000

W2 IP load (N)
W1 IP load (N)

15 000

SANS 10160 ULS-W 44 m/s 10 000 NUM ULS-S

10 000 SANS 10160 ULS-W 44 m/s

SANS 10160 ULS-W 40 m/s

SANS 10160 ULS-W 36 m/s NUM SLS SANS 10160 ULS-W 40 m/s
NUM SLS SANS 10160 ULS-W 36 m/s
5 000
5 000 NUM ULS-W SANS 10160 SLS 44 m/s NUM ULS-W
SANS 10160 SLS 40 m/s SANS 10160 SLS 44 m/s
SANS 10160 SLS 40 m/s
SANS 10160 SLS 36 m/s
SANS 10160 SLS 36 m/s

0 0
0 0.05 0.10 0.15 0.20 0 0.2 0.4 0.6 0.8
W1 IP displacement (mm) W2 IP displacement (mm)

Figure 13 In-plane response of walls W1 (left) and W2 (right) for SLS, ULS-W and ULS-S

Summary points of diversion, such as adjusting the method was used throughout. However,
The progression of most of the analyses step size, increasing the tolerance of the convergence was rarely achieved. It is clear,
was limited by divergence. Several steps convergence criteria, employing other though, from the load-displacement or
were taken to move the analyses past these iterative procedures, etc, and the arc length load-deflection trace, that the linear-elastic

35 000 35 000

SANS 10160 ULS-W 44 m/s

30 000 30 000 SANS 10160 ULS-W 44 m/s

SANS 10160 ULS-W 40 m/s

25 000 NUM ULS-S Pier 4 25 000 SANS 10160 ULS-W 40 m/s

NUM SLS Base 1

SANS 10160 ULS-W 36 m/s SANS 10160 ULS-W 36 m/s
W2 OP load (N)
W1 OP load (N)

20 000 20 000
SANS 10160 ULS-S
Negligible
0.25 mm

NUM SLS Base 1

NUM ULS-W Pier 1
15 000 NUM ULS-W Base 1 15 000
SANS 10160 SLS 44 m/s
SANS 10160 SLS 44 m/s
SANS 10160 SLS 40 m/s
10 000 10 000 SANS 10160 SLS 40 m/s
SANS 10160 ULS-S
SANS 10160 SLS 36 m/s
SANS 10160 SLS 36 m/s
5 000 5 000
Negligible
0.25 mm

NUM ULS-S Pier 2

0 0
0 0.1 0.2 0.3 0 0.1 0.2 0.3
W1 crack width (mm) W2 crack width (mm)

Figure 14 Crack damage classification for walls W1 (left) and W2 (right) for SLS, ULS-W and ULS-S

Journal of the South African Institution of Civil Engineering Volume 63 Number 1 March 2021 55
region has been surpassed and that post- Table 8 CON adjusted joint parameters
peak global stiffening is improbable. Joint interface parameters Symbol Baseline Adjusted Unit
Given the conventional strength and
stiffness range of the concrete masonry Tensile strength f t,j TENSTR 0.12 0.84 N/mm2

studied, as well as the typically conservative
nature of deemed-to-satisfy solutions, it was
anticipated prior to the study that the resis-
tance of these conventional masonry walls
would exceed the design loads in both ULS
load cases. Contrary to this expectation,
both wall configurations failed to resist the
ULS-W design loads for basic fundamental
wind speeds of 44, 40 and 36 m/s, as did W2
for the ULS-S design loads, and by a large
margin. Plausible origins of these failure
are the applied design load, the material
input parameters, the derived geometry of
the wall configurations and the assumed
boundary conditions. These aspects are
discussed in the following section.
DISCUSSION
Design load
Most of the assumptions or selections
detailed for determining the wind load
were made to achieve the most critical
wind loading, not the most likely. However,
even design wind loads based on 40 and
Mode I fracture energy GIf,j
Cohesion
Material
To investigate the potential increase in OP
load-carrying capacity due to improved
material properties, reasonable maximum
values for the three most influential OP
parameters (joint tensile strength, joint
cohesion and joint mode I fracture energy)
were sought in literature. Thereafter,
ULS-W load analyses were performed on
W1 and W2 with these three adjusted
joint parameters. Experimental data on
the joint tensile properties is scarce, but
reasonable maximum values for mode I
fracture energy and cohesion were found
in literature for normal density concrete
blocks with general purpose mortar joints,
conducted by Van Der Pluijm (1999),
and are detailed in Table 8. All other
GF 0.005 0.011 N/mm
cj COHESI 0.17 1.17 N/mm2
parameters, as provided in Table 2, are
kept constant.
The outcome of these analyses (Table 9
and Figure 17) shows an increase in the
load-carrying capacity of 53% for W1 and
11% for W2, due to the improved material
properties. This reduces the discrepancy
to the most critical design load by 47%
and 13% for W1 and W2, respectively.
Increasing the critical joint parameters
to reasonable maximum values does not
increase W2’s resistance sufficiently to
withstand any of the wind design loads.
In the case of W1, the wall’s resistance is
increased such that the ULS-W load case
based on the 36 m/s basic fundamental
wind speed can be resisted, but the 40 and
44 m/s not.
6 000 mm

756 mm
36 m/s basic fundamental wind speeds,
which account for the majority of areas in
South Africa, exceed the walls’ capacities in

1 008 mm
most instances.
2 646 mm

It is also noteworthy that the wind pres-

sure determined in this study (1 213 N/m2,
Table 5) is over three times the minimum

882 mm
wind pressure specified in SANS 10400-B
(SANS 2012) (370 N/m2). The substantially
higher design load for the ULS-W case is in
part due to the recent revision of the wind 1 200 mm 900 mm 1 200 mm 1 500 mm 1 200 mm
loading code, SANS 10160-3 (SANS 2018b).
The wind load partial factor has
4 950 mm
increased from 1.3 to 1.6 and the highest
fundamental basic wind speed from 36 m/s
630 mm 630 mm

to 44 m/s. These two changes combined

result in a 1.5 times higher load than
would have been the case before these revi-
sions. However, the ratios of design load
to resistance are 2.1 and 1.9 for original
1 010 mm

configurations of W1 and W2, respectively.

2 650 mm

These revisions alone do not account for the

discrepancy, and they were implemented for
good reason. The reliability performance
1 010 mm

of the original wind load partial factor of

1.3 was found to be inadequate (Botha et
al 2018) and the South African wind map
has improved due to, in part, a seven-fold 900 mm 900 mm 750 mm 1 500 mm 900 mm
increase in the historical extreme wind data
available in South Africa (Kruger et al 2017). Figure 15 Wall layouts and dimensions for reduced window openings for W1 (top) and W2 (bottom)

56 Volume 63 Number 1 March 2021 Journal of the South African Institution of Civil Engineering
Figure 16 Baseline (left) and adjusted (right) boundary conditions for return walls
Table 9 P
 eak OP resistance to ULS-W and IP to ULS-S for adjusted material parameters, geometry
and boundary conditions
Peak resistance [N]
Baseline
Adjusted material parameters
Adjusted geometry
Adjusted boundary conditions
It is important to recall that in all the
analyses of W1 and W2, mean material
parameter values are used, and the material
resistance has not been reduced by means of
material partial safety factors. Applying this
necessary reduction for ULS-based design
would further widen the discrepancy.
Geometry
The limitations on wall panel sizes and
openings set out in the SANS 10400-K
deemed-to-satisfy solutions are taken from
the JSD Code of Practice: Foundations and
Superstructures for Single Storey Residential
Buildings of Masonry Construction (JSD
1995). Different wall panel configurations
were analysed using the yield line approach
to derive the panel sizes, and the then
current South African masonry structural
design code SABS 0164-1 (SABS 1980) was
applied to the respective elements to derive
the opening limitations (Watermeyer 1996).
The total area of openings for both
W1 and W2 falls within the specifications
of seismic design principles set out in the
loading code, SANS 10160-4 (SANS 2017)
Clause 6.2.2, of being less than one third
of the overall wall area. The openings
are positioned “as uniformly as possible”,
but given the large opening length of 3 m
permitted in the deemed-to-satisfy solu-
tions of SANS 10400-K (SANS 2011b), it
does result in large openings at both wall
ends, which is undesirable according to
Journal of the South African Institution of Civil Engineering Volume 63 Number 1 March 2021
OP W1 OP W2 IP W2
15 597 16 543 11 295
23 835 18 373 –
18 687 18 946 29 534
18 011 18 197 –
the seismic design principles of the loading
code.
To investigate the potential improve-
ment in OP resistance of both walls under
ULS-W loading and IP resistance for W2
under ULS-S loading due to more robust
geometry, the original window opening
length is halved to 1 500 mm, and the door
and window openings are positioned in less
extreme positions in the wall, as illustrated
in Figure 15 for W1 and W2. All other origi-
nal dimensions of the walls are maintained.
The outcome of the OP ULS-W
analyses (Table 9 and Figure 17) shows
an increase in the load-carrying capacity
of 20% for W1 and 15% for W2, due to
the reduced window opening and the less
extreme positions of the openings. This
reduces the discrepancy to the most critical
design load by 17% for both W1 and W2.
The outcome of the IP ULS-S analyses
on W2 (Table 9 and Figure 18) shows an
increase in the load-carrying capacity of
160% for W2, due to the reduced window
opening and the less extreme positions
of the openings. The IP resistance of W2
now exceeds the seismic design load by
40%. Reducing the window opening by half
does not increase the walls’ OP resistances
sufficiently to withstand any of the wind
design loads. However, this mitigation
strategy significantly increases W2’s IP
load-carrying capacity to successfully resist
the full seismic design load.
Boundary conditions
The conservative assumption was made
that the timber truss system provides
negligible lateral support to the top of the
walls. The effect of this assumption could
be meaningful, but its validity is sustained
given the similarly weak OP resistance
of the opposite wall, which is meant to
provide the additional lateral resistance, as
well as the typically poor quality of con-
nection between truss and wall in LIH.
A potential source of error could be
excessive rotation of the short return walls,
which provide lateral restraint to the walls.
The pinned modelling of the walls could
underestimate the rotational restraint
that a full-length return wall would offer,
thereby allowing greater OP deflection.
To investigate the effect of this, the trans-
lational restraint on the return walls is
applied to all nodes in the boundary plane,
as opposed to just the central row of nodes,
as shown in Figure 16.
The outcome of the OP ULS-W
analyses (Table 9 and Figure 17) shows
an increase in the load-carrying capacity
of 15% for W1 and 10% for W2, due to
the adjusted boundary conditions and
increased rotational restraint. This reduces
the discrepancy to the most critical design
load by 14% for W1 and 12% for W2.
Increasing the rotational restraint provided
by the return walls does not increase the
walls’ resistances sufficiently to withstand
any of the wind design loads.
CONCLUSIONS
The FE analyses were executed success-
fully for the two wall configurations under
three load cases. The failure modes can
be broadly classified as tensile for the
OP-dominant cases and a combination of
tensile/shear failure for the IP-dominant
cases. The analyses revealed the wall
57
 35 000 35 000

SANS 10160 ULS-W 44 m/s

SANS 10160 ULS-W 44 m/s
30 000 30 000
SANS 10160 ULS-W 40 m/s
SANS 10160 ULS-W 40 m/s
Adjusted material
25 000 parameters 25 000

SANS 10160 ULS-W 36 m/s

SANS 10160 ULS-W 36 m/s

W2 OP load (N)
W1 OP load (N)

20 000 Adjusted boundary 20 000 Adjusted material

conditions Adjusted boundary parameters Adjusted
conditions geometry
Adjusted
15 000 geometry 15 000
Baseline Baseline

10 000 10 000

5 000 5 000

0 0
0 1 2 3 4 0 1 2 3
W1 OP deflection (mm) W2 OP deflection (mm)

Figure 17 OP ULS-W response for adjusted material parameters, geometry and boundary conditions for W1 (left) and W2 (right)

configurations’ failure to resist the design

35 000 loads in most instances, and significantly
so in the OP response to the ULS-W load
case. This is in part, but not exclusively,
30 000 Adjusted due to recent increases to two important
geometry
parameters in the wind loading code.
Three mitigation strategies were
employed, namely improving the tensile
25 000 performance by increasing three critical
joint material parameters, improving the
W2 IP load (N)

distribution of openings, and increas-

SANS 10160 ULS-S
ing the rotational restraint of the return
20 000 walls. None of these strategies improved
the resistance of the walls to the point of
successfully resisting the full design wind
load. However, reducing the length and
15 000
improving the distribution of openings
significantly increased the IP resistance of
W2 to seismic loading.
Additionally, it is important to recall
10 000
that these analyses were performed without
Baseline the use of material partial safety factors.
Current standardised partial factors for
conventional masonry materials range
5 000
between 1.5 and 3.5. Reducing the material
strength by these factors, as is required by
limit states design, would further signifi-
cantly increase the discrepancy between
0
0 0.1 0.2 0.4 0.6 0.8 the walls’ resistances and the design loads.
W2 IP displacement (mm) This raises the issue of possible incon-
sistencies between the deemed-to-satisfy
Figure 18 IP ULS-S response for reduced window opening for W2 provisions in SANS 10400-K for wall

58 Volume 63 Number 1 March 2021 Journal of the South African Institution of Civil Engineering
panel and opening sizes and the current
EC-based loading code, both for wind, i.e.
SANS 10160-3 (SANS 2018b) and seismic,
i.e. SANS 10160-4 (SANS 2017) loads.
Based on a case study of extreme wind
loads on an inland housing development in
South Africa, Mahachi et al (2018) came to
the conclusion that a review of the techni-
cal standards in housing development is
necessary, specifically the NHBRC Home
Building Manual. Griffith (2000) reports
on a similar case of discrepancies between
the capacity of the ‘deemed-to-comply’ wall
provisions of the South Australian Housing
Code and the Australian masonry stan-
dard’s design load. It has long been found
that, within the field of masonry buildings,
low-rise, unreinforced ones with light
roofs (such as LIH), experience the most
wind damage (Sparks et al 1989), especially
non-engineered ones, relying on empirical
design procedures. This, coupled with the
significant changes in the required wind
and seismic design loads with the revision
of the South African loading code, war-
rants a review of the SANS 10400 deemed-
to-satisfy wall provisions.
It is recommended that this is done with
the preferential housing solution in mind, by
first specifying desired wall configurations,
based on constructability, typical South
African building practice and skill level, fen-
estration requirements for building energy
usage, natural lighting and ventilation,
and fire safety. Numerous such specifica-
tions are well documented in the relevant
parts of SANS 10400 and can be used as a
basis to determine the desired geometry.
Subsequently, the derived wall configura-
tions can be structurally analysed using the
simplified micro-modelling approach.
ACKNOWLEDGEMENTS
Funding: This work was supported by the
National Research Foundation of South
Africa (grant numbers 87961, 106965).
REFERENCES
Abdulla, K, Cunningham, L & Gillie, M 2017.
Simulating masonry wall behaviour using a
simplified micro-model approach. Engineering
Structures, 151: 349–365.
Agnihotri, P, Singhal, V & Rai, D 2013. Effect of
in-plane damage on out-of-plane strength of
unreinforced masonry walls. Engineering Structures,
57: 1–11.
Botha, J, Retief, J & Viljoen, C 2018. Reliability
assessment of the South African wind load design
formulation. Journal of the South African Institution
of Civil Engineering, 60(3): 30–40.
BS (British Standard) 1978. BS 5628-1:1978. Code of
Practice for the Use of Masonry. Structural Use of
Unreinforced Masonry. London: British Standards
Institution.
BSI (British Standards Institution) 2005. NA to BS EN
1996-1-1. UK National Annex to Eurocode 6. Design
of Masonry Structures. General Rules for Reinforced
and Unreinforced Masonry Structures. London: BSI.
CMA (Concrete Manufacturers Association) 2011.
Initiatives. http://www.cma.org.za/Initiatives/
The-CMA-House.
De Villiers, W I 2019. Computational and experimental
modelling of masonry walling towards performance-
based standardisation of alternative masonry units
for low-income housing. PhD Thesis. Stellenbosch
University.
De Villiers, W I, Fourie, J & Boshoff, W P 2018.
Numerical modelling of alternative masonry units.
Proceedings, International RILEM Symposium on
Concrete Modelling (CONMOD18), 26–29 August
2018, Delft, The Netherlands.
Department of Human Settlements (DHS) 2017.
Delivery of Serviced Sites and Houses/Units from
HSDG. Pretoria: DHS.
Derakhshan, H, Lucas, W, Visintin, P & Griffith, M
2018. Out-of-plane strength of existing two-way
spanning solid and cavity unreinforced masonry
walls. Structures, 13: 88–101.
DIANA 2017. Finite Element Analysis User’s Manual
10.2. Delft: DIANA FEA BV.
Dolatshahi, K, Aref, A & Whittaker, A 2015.
Interaction curves for in-plane and out-of-plane
behaviours of unreinforced masonry walls. Journal
of Earthquake Engineering, 19: 60–84.
EN (European Standard) 1990. EN 1990 (2002).
Eurocode: Basis of Structural Design. Brussels:
European Committee for Standardization (CEN).
EN (European Standard) 1991. EN 1991 (2002+).
Eurocode 1: Actions on Structures. Brussels: CEN.
EN (European Standard) 1998. EN 1998-1 (2004).
Eurocode 8: Design of Structures for Earthquake
Resistance. Part 1: General Seismic Actions and
Rules for Buildings. Brussels: CEN.
Fourie, J 2017. Characterisation and evaluation of the
mechanical properties of alternative masonry units.
MEng Dissertation. Stellenbosch University.
Giambanco, G, Rizzo, S & Spallino, R 2001. Numerical
analysis of masonry structures via interface models.
Computer Methods in Applied Mechanics and
Engineering, 190: 6493–6511.
Griffith, M 2000. Experimental Study of the Flexural
Strength of URM (Brick) Walls. Adelaide, Australia:
University of Adelaide.
ICBO (International Conference of Building Officials)
1997. Uniform Building Code (UBC)™. Whittier, CA:
ICBO.
JSD (Joint Structural Division) 1995. Code of Practice:
Foundations and Superstructures for Single Storey
Residential Buildings of Masonry Construction.
Halfway House: SAICE and IStructE Joint
Structural Division.
Kruger, A, Retief, J & Goliger, A 2017. Development
of an updated fundamental basic wind speed map
for SANS 10160-3. Journal of the South African
Institution of Civil Engineering, 59(4): 12–25.
Laubscher, J 2014. Reviewing challenges between the
need for government-subsidised housing in South
Africa and the sustainability requirements of the
National Building Regulations. Proceedings, World
SB14, 28–30 October 2014, Barcelona, Spain,
pp 18–24.
Lourenço, P B 1996. Computational Strategies for
Masonry Structures. Delft, The Netherlands: Delft
University Press.
Lourenço, P B & Rots, J 1997. Multisurface interface
model for analysis of masonry structures. Journal of
Engineering Mechanics, 123: 660–669.
Macorini, L & Izzuddin, B 2011. A non-linear interface
element for 3D mesoscale analysis of brick-masonry
structures. International Journal for Numerical
Methods in Engineering, 85: 1584–1608.
Mahachi, J, Bradley, R & Goliger, A 2018. Wind
storm damage to houses: Planning and design
consideration. Proceedings, Out-Of-The-Box
Conference, 24–25 October 2018, Pretoria,
pp 82–92.
Milani, G 2008. 3D upper bound limit analysis of
multi-leaf masonry walls. International Journal of
Mechanical Sciences, 50: 817–836.
Najafgholipour, M, Maheri, M & Lourenço, P B 2013.
Capacity interaction in brick masonry under
simultaneous in-plane and out-of-plane loads.
Construction and Building Materials, 38: 619–626.
NHBRC (National Home Builders Registration
Council) 2015. Home Building Manual and Guide.
Johannesburg: NHBRC.
Reyes, E, Casati, M J & Gàlvez, J C 2008. Cohesive
crack model for mixed mode fracture of brick
masonry. International Journal of Fracture,
151: 29–55.
Roca, P, Cervera, M, Gariup, G & Pela, L 2010. Structural
analysis of masonry historical constructions:
Classical and advanced approaches. Archives of
Computer Methods in Engineering, 17: 299–325.
RSA (Republic of South Africa) 1977. National Building
Regulations and Building Standards Act, 103 1977,
Government Gazette, No 145 (5640), 6 July 1977,
Pretoria: Government Printer.
RSA (Republic of South Africa) 1995. Housing
Consumers Protection Measures Act, 9 of 1995.
Government Gazette, 401(19418), 2 November 1995,
Pretoria: Government Printer.
SABS 1980. SABS 0164-1:1980. Code of Practice. The
Structural Use of Masonry, Part 1: Unreinforced
Masonry Walling. Pretoria: South African Bureau of
Standards.
SANS (South African National Standard) 1989.
SANS 10164-1 1989. The Structural Use of Masonry,

Journal of the South African Institution of Civil Engineering Volume 63 Number 1 March 2021 59
 Part 1: Unreinforced Masonry Walling. Pretoria:
SABS Standards Division.
SANS 2010a. SANS 10400 2010. The Application of
the National Building Regulations, 3rd ed. Pretoria:
SABS Standards Division.
SANS 2010b. SANS 10400-A 2010. The Application of
the National Building Regulations. Part A: General
Principles and Requirements, 3rd ed. Pretoria: SABS
Standards Division.
SANS 2011. SANS 10160 2011. Basis of Structural Design
and Actions for Buildings and Industrial Structures,
1.1 ed. Pretoria: SABS Standards Division.
SANS 2011a. SANS 10160-2 2011. Basis of Structural
Design and Actions for Buildings and Industrial
Structures. Part 2: Self-weight and Imposed Loads.
Pretoria: SABS Standards Division.
SANS 2011b. SANS 10400-K 2011. The Application of
the National Building Regulations. Part K: Walls,
3rd ed. Pretoria: SABS Standards Division.
SANS 2011c. SANS 10400-L 2011. The Application of
the National Building Regulations. Part L: Roofs,
3rd ed. Pretoria: SABS Standards Division.
SANS 2011d. SANS 10400-XA 2011. The Application
of the National Building Regulations. Part X:
Environmental Sustainability; Part XA: Energy
Usage in Buildings. Pretoria: SABS Standards
Division.
SANS 2012. SANS 10400-B 2012. The Application
of the National Building Regulations. Part B:
Structural Design, 3rd ed. Pretoria: SABS Standards
Division.
60
SANS 2017. SANS 10160-4 2017. Basis of Structural
Design and Actions for Buildings and Industrial
Structures. Part 4: Seismic Actions and General
Requirements for Buildings. Pretoria: SABS
Standards Division.
SANS 2018a. SANS 10160-1 2018. Basis of Structural
Design and Actions for Buildings and Industrial
Structures. Part 1: Basis of Structural Design.
Pretoria: SABS Standards Division.
SANS 2018b. SANS 10160-3 2018. Basis of Structural
Design and actions for Buildings and Industrial
Structures. Part 3: Wind. Pretoria: SABS Standards
Division.
SANS 2018c. SANS 51996-1-1 2018. Eurocode 6: Design
of Masonry Structures. Part 1-1: General Rules for
Reinforced and Unreinforced Masonry Structures,
1st ed. Pretoria: SABS Standards Division.
Sisulu, L 2016. Housing Backlog at 2.1 million, says
Minister Sisulu. Engineering News, 22 April 2016.
Sparks, P, Liu, H & Saffir, H 1989. Wind damage
to masonry buildings. Journal of Aerospace
Engineering, 2(4): 186–198.
Sýkora, M & Holický, M 2010. Probabilistic model for
masonry strength of existing structures. Engineering
Mechanics, 174(1): 61–70.
Talocchino, G 2005. Design and construction criteria
for domes in low-cost housing. MEng Dissertation,
Johannesburg: University of the Witwatersrand.
Vaculik, J 2012. Unreinforced masonry walls subjected
to out-of-plane seismic actions. PhD Thesis.
Adelaide, Australia: University of Adelaide.
Volume 63 Number 1 March 2021
Van Der Pluijm, R 1999. Out-of-plane bending of
masonry: Behaviour of bed joints. PhD Thesis.
Eindhoven, The Netherlands, Technische
Universiteit Eindhoven, pp 1849–1862. doi.
org/10.6100/IR528212.
Van Zijl, G P A G 2000. Computational modelling of
masonry creep and shrinkage. PhD Thesis. Delft,
The Netherlands: Delft University of Technology.
Watermeyer, R 1996. Recent developments in providing
houses of masonry construction in South Africa.
The Structural Engineer, 74(19): 325–331.
Watermeyer, R 2004. The impact of structural
engineering on the sustainability of human
settlements in developing countries. Proceedings, 2nd
International Conference on Structural Engineering,
Mechanics and Computation, Cape Town.
Watermeyer, R & Tromp, B 1992. A systematic
approach to the design and construction of single-
storey residential masonry structures on problem
soils. The Civil Engineer in South Africa, March:
83–96.
Watermeyer, R & Milford, R 2003. The use of
performance-based building codes to attain
sustainable housing objective: The South African
approach. Presented at the Global Policy Summit on
the Role of Performance-Based Building Regulations
in Addressing Societal Expectations, International
Policy, and Local Needs. Washington, DC, US.
Wium, J A 2010. Background to draft SANS 10160
(2009) Part 4. Seismic loading. Journal of the South
African Institution of Civil Engineering, 52(1): 20–27.
Journal of the South African Institution of Civil Engineering
Design and construction TECHNICAL NOTE
of laboratory-scale Journal of the South African
Institution of Civil Engineering

activated carbon, gravel ISSN 1021-2019

Vol 63 No 1, March 2021, Pages 61–66, Technical Note 1158

and rice husk filter columns ENGINEER CLEMENT OGUCHE ATAGUBA is a

for the treatment of

lecturer at the Federal Polytechnic, Idah, in
Nigeria, where he teaches civil and water
resources engineering courses at both

stormwater runoff from

National and Higher National Diploma
levels. His research interests are in water
quality/treatment, municipal waste

automobile workshops
management, water sanitation and hygiene
(WASH), and the application of computer technology in water resources
and sanitation facilities management. He is currently a PhD student at
Stellenbosch University.

Contact details:
C O Ataguba, I Brink Department of Civil Engineering
Stellenbosch University
Private Bag X1, Matieland,
Stellenbosch 7602
The design and construction of low-cost laboratory-scale filter columns using locally available South Africa
Nigerian filter materials – granular activated carbon (GAC), gravel (GR) and rice husk (RH) – were T: +234 803 567 1452
E: clematrix2008@gmail.com
carried out and reported. The filter materials and columns were designed, constructed and used
for the treatment of stormwater runoff from selected automobile workshops in Nigeria over
DR ISOBEL BRINK is a senior lecturer in water
a period of three rainy months. The combined granular activated carbon and rice husk filter quality and environmental engineering at
systems performed best with pollutant removal efficiency of 58%. It was shown that the materials, the Department of Civil Engineering,
considered as waste, could be recycled and used as filter materials in the treatment of stormwater Stellenbosch University. She is interested in
new innovations in water quality
from automobile workshops. This low-cost technology for stormwater runoff treatment, especially
improvement methods and undertakes
for automobile workshops at large scale and in-situ, can be further explored. projects with wide application. Recent
research includes simple point-of-use
systems for potable water treatment, river water quality modelling, and
the use of green infrastructure and LID technologies for surface runoff
INTRODUCTION and non-availability of appropriate labour water quality improvement.
Stormwater runoff from automobile to operate the technologies (Bahgat et al
Contact details:
workshops washes debris, waste, oil com- 1999). These challenges have created Department of Civil Engineering
binations, grease, etc, through the drainage opportunities for the exploration of locally Stellenbosch University
infrastructure within the urban catchment available materials/technologies. Private Bag X1, Matieland,
Stellenbosch 7602
and discharges into receiving water bodies. In this research, the potential of com-
South Africa
This has resulted in a high degree of water bined (a) gravel – activated carbon, (b) acti- T: +27 21 808 4195
pollution in Nigeria (Idu 2015; Ekiye & vated carbon – rice husk, and (c) single rice E: icbrink@sun.ac.za
Zejiao 2010). husk as filter materials in the treatment of
Studies have shown that land use plays a selected automobile workshop stormwater
critical role in the concentration and com- runoff was investigated. These materials
position of pollutants that are discharged are readily available and affordable in
into receiving water bodies (Khatun et al Nigeria. Five automobile workshops were
2014; Shrestha 2017; Wang et al 2013). selected from the two study towns of Idah
In a study carried out by Pitt et al (1995), and Lokoja in Nigeria for this stormwater
it was reported that stormwater runoff sampling and treatment. These selected
from vehicle service and parking lots was workshops were named Automobile
found to have the highest levels of metals, Workshops 1–5.
petroleum hydrocarbon compounds and This technical note reports the design,
organics pollutants when compared with construction and preliminary use of the
other urban land uses. combined (a) gravel – granular activated
Conventional wastewater/stormwater carbon, (b) granular activated carbon – rice
treatment methods, such as reverse osmo- husk, and (c) single rice husk filter materi-
sis, chemical precipitation, electrodialysis, als as vertical columns for the treatment of
ion exchange, ultrafiltration, etc, have automobile workshop stormwater runoff.
been reported to be unsuitable for adop- Results of the laboratory investigation
tion in developing countries due to high of the pre-treatment and post-treatment Keywords: automobile workshop, filter media, pollution,
operating costs, high energy requirement quality of the stormwater samples collected stormwater runoff, treatment

Ataguba CO, Brink I. Design and construction of laboratory-scale activated carbon, gravel and rice husk filter columns for the treatment of stormwater
runoff from automobile workshops. J. S. Afr. Inst. Civ. Eng. 2021:63(1), Art. #1158, 6 pages. http://dx.doi.org/10.17159/2309-8775/2021/v63n1a6 61
 (a) (b) (c)
Figure 1 F ilter materials: (a) granular activated carbon, (b) gravel, (c) rice husk

over the period of nine weeks of a rainy

season in Nigeria have also been reported.
The use of these filter materials in
water treatment is not a novel concept,
as their uses, as documented by Halli et
Raw
stormwater Raw al (2018), Maebh (2016) and Lakhote et
stormwater
al (2016), have proved that they are good
low-cost water treatment materials. These
filter materials are ecologically safe, and
need little or no regeneration due to their
local availability and low production costs
(Cecen & Aktas 2011; Baker et al 1992;
Andersen 2016; Sukia 2020; Nurul Amin et
al 2006; Zunaira & Zhi, 2017; Xu et al 2013;
Chukwudebelu et al 2015).
GAC
filter Gravel
filter Raw
stormwater MATERIALS AND
METHODOLOGY
Figures 1(a – c) show the different filter
materials adopted for this research.
Similarly, Figure 2 is a schematic representa-
tion of the different filter media setups for
the filtration process. These configurations/
order of arrangement of the filter columns
in Figure 2 were basically the choice of the
researchers. The denser filter materials were
placed above the less dense ones. However,
GAC
filter further research work may consider chang-
ing the configurations of the filter columns
in order to study the quality of the effluents
for comparison with these results.
The smooth surface gravel was obtained
at minimal cost from the gravel mining
site at the bank of the River Niger in Idah,
Nigeria. Raw carbon was obtained from
local dealers of carbon. The raw carbon
was converted to granular activated carbon
Effluent Effluent Effluent using thermal activation (pyrolysis) as
collector collector collector
described in McDougall (1991). Rice husk
used for this research was obtained from
the rice mill in Idah, Nigeria. Commercially
available 100 mm diameter polyvinyl
(a) (b) (c) chloride (PVC) pipes were sourced from
the local market and used to fabricate the
Figure 2 S chematic representation of the filter media setups for the automobile stormwater filter columns. Other materials that were
runoff treatment: (a) GAC – RH filter setup, (b) GR – GAC filter setup, (c) RH filter setup obtained and used for the construction

62 Volume 63 Number 1 March 2021 Journal of the South African Institution of Civil Engineering
Table 1 Details of design equations, data and references for the filter designs
Filter Design equation Design data Reference

Vf ⎫⎪ ⎫⎪ = = Q ÷
Depth of column H = 0.40 m
πD 2
⎭h⎭ A
m Q Height of filter L = 0.3 m
Freeboard t = 0.10 m
4
Diameter of column D = 0.10 m
Wegelin (1996);
GR filter design Volume of gravel VG = 0.0024 m3
Density-Mass-Volume Relation
Density of gravel ρG = 1 400 kg/m3
Mass (m) Mass of gravel MG = 3.4 kg
Density ρ =
Volume (v) Filtration rate Vf = 1.0 m/h
Flow rate Q = 0.008 m3/h

Depth of column H = 0.40 m

V A×L Height of column L = 0.30 m
EBCT = =
Q Q Freeboard t = 0.10 m
Diameter of column D = 0.10 m USACE (2001);
GAC filter design
Volume of GAC VGAC = 0.0024 m3 Density-Mass-Volume Relation
Mass (m) Density of GAC ρGAC = 650 kg/m3
Density ρ = Mass of GAC MGAC = 1.56 kg
Volume (v)
Empty Bed Contact Time (EBCT) = 18 mins

Depth of column H = 0.40 m

Height of filter L = 0.30 m
Mass (m) Freeboard t = 0.10 m
RH filter design Density ρ = Diameter of column D = 0.10 m Density-Mass-Volume Relation
Volume (v) Volume of RH VRH = 0.0024 m3
Density of RH ρRH = 100 kg/m3
Mass of RH MRH = 0.24 kg

of the filter columns were PVC filter

mesh, Araldyte (sealant), flow regulator Influent
and discharge hose. The arrangements of
these materials is schematically shown in t(m) Free board
Figure 3. Filter column stands were also
fabricated to hold the individual columns
PVC filter column
as shown in Figures 2, 3 and 5.

L(m)
Design and construction of Filter material
granular activated carbon, gravel
and rice husk filter columns
The vertical downflow bed filter system PVC filter mesh
has been adopted from Wegelin (1996).
This design concept is based on its ease of Araldyte (sealant)
use of gravity flow in underdeveloped areas
(Diaper 1965; Pratap et al 2007). Flow regulator
The details of the design equations,
data and the relevant references used in Discharge hose
the design of the filters are presented in
Table 1 with the design equations from the
Filter stand
references, while the dimensions are the
authors’ work. The design data in Table 1 Effluent collector

was used to construct the columns and the

setup for the entire filter systems as shown
in Figures 3 and 4.
The PVC mesh was introduced in the
setup as shown in Figure 3 to retain and
hold back particles from the stormwater
runoff that might cause blockage of the
flow regulator. The PVC caps cover-
ing the PVC mesh at the bottom of the
columns were sealed to the column using
a commonly available water tight sealant
called Araldyte to eliminate leakages in
the system. Figure 3 C
 ross-section of filter system showing the different components

Journal of the South African Institution of Civil Engineering Volume 63 Number 1 March 2021 63
Table 2 Pollutant removal efficiency of the filters for Automobile Workshop 4 at week 1
Te Te Te
Untreated
Test parameters GAC-RH (GAC-RH) GR-GAC (GR-GAC) RH (RH)
sample
% % %

PH 8.24 35.875 10.46 18.599 9.95 22.568 12.85

Conductivity (µS/cm) 823.1 71.417 1 784.34 38.038 898.15 68.811 2 879.73

Turbidity (NTU) 101 63.139 266 2.920 114 58.394 274

Dissolved oxygen, DO (mg/L) 4.62 36.712 5.66 22.466 6.13 16.027 7.3

Total dissolved solids, TDS (mg/L) 604.08 68.776 1 181.01 38.956 646.56 66.581 1 934.69

Total solids, TS (mg/L) 786.08 82.162 1 297.01 70.567 787.56 82.128 4 406.69

Total suspended solids, TSS (mg/L) 182 92.638 116 95.307 141 94.296 2472

Oil and grease, O&G (mg/L) 3.43 46.238 4.88 23.511 4.24 33.542 6.38

Cadmium, Cd (mg/L) 0.012 47.826 0.014 39.130 0.017 26.087 0.023

Copper, Cu (mg/L) 0.016 46.667 0.018 40.000 0.016 46.667 0.03

Lead, Pb (mg/L) 1 41.860 1.22 29.070 1.35 21.512 1.72

Iron, Fe (mg/L) 20.5 45.767 28.65 24.206 32.35 14.418 37.8

Average pollutant removal efficiency (%) 56.590 36.898 50.343

Filter system setups, stormwater
sampling, treatment and
laboratory analyses
The different filter materials designed in
Table 1 were placed in the different filter
columns and fastened to the fabricated
stand as shown in Figure 4(b).
Stormwater sampling from each
automobile workshop spanning a period
of nine weeks during the rainy season was
carried out according to Lowe et al (2018).
The influents and effluents from the differ-
ent filter setups were analysed for quality
before (raw stormwater) and after treatment
with the filter media (filtered effluents).
Replacement of filter material was chosen
over regeneration for this study, as it is eco- filter system was computed, as presented in
nomical, and materials are locally available Tables 2, 3 and 4.
at little or no cost. The parameters analysed The pollutant removal efficiency Te of a
included: pH, conductivity, turbidity, oil particular filter system with respect to any
and grease (O&G), dissolved oxygen (DO), particular parameter is given as:
total dissolved solids (TDS), total solids
(TS), total suspended solids (TSS), cadmium Cd – C x
Te(x)% = × 100
(Cd), copper (Cu), lead (Pb) and iron (Fe). Cd
These parameters were selected based on
the characteristic pollutants associated with Where:
this land use (Pitt et al 1995). The param- Cd = concentration of a particular param-
eters were analysed in accordance with eter from its untreated sample
APHA (2017) at the Water Quality Control C x = concentration of the same parameter
Laboratory at the National Geosciences from the filtered sample x
Research Laboratories in Kaduna, Nigeria. (for x = filter systems GAC-RH, GR-GAC
The efficiency of pollutant removal of each or RH).

Table 3 Pollutant removal efficiency of the filters for Automobile Workshop 4 at week 2
Te Te Te
Untreated
Test parameters GAC-RH (GAC-RH) GR-GAC (GR-GAC) RH (RH)
sample
% % %

PH 8.91 33.108 10.8 18.919 10.23 23.198 13.32

Conductivity (µS/cm) 886 70.328 1 843 38.279 1 024 65.707 2 986

Turbidity (NTU) 98 63.838 192 29.151 106 60.886 271

Dissolved oxygen, DO (mg/L) 4.28 45.408 5.1 34.949 5.88 25.000 7.84

Total dissolved solids, TDS (mg/L) 641.46 68.547 1 164.78 42.887 770.05 62.242 2 039.44

Total solids, TS (mg/L) 854.46 80.200 1 357.78 68.537 971.05 77.498 4 315.44

Total suspended solids, TSS (mg/L) 213 90.641 193 91.520 201 91.169 2 276

Oil and grease, O&G (mg/L) 3.73 43.910 5.12 23.008 4.81 27.669 6.65

Cadmium, Cd (mg/L) 0.014 48.148 0.015 44.444 0.017 37.037 0.027

Copper, Cu (mg/L) 0.015 50.000 0.017 43.333 0.017 43.333 0.03

Lead, Pb (mg/L) 1.1 42.408 1.45 24.084 1.64 14.136 1.91

Iron, Fe (mg/L) 17.52 46.125 23.6 27.429 27.85 14.360 32.52

Average pollutant removal efficiency (%) 56.889 40.545 49.106

64 Volume 63 Number 1 March 2021 Journal of the South African Institution of Civil Engineering
Table 4 Pollutant removal efficiency of the filters for Automobile Workshop 4 at week 3
Te Te Te
Untreated
Test parameters GAC-RH (GAC-RH) GR-GAC (GR-GAC) RH (RH)
sample
% % %

PH 9.82 24.287 11.15 14.032 10.72 17.348 12.97

Conductivity (µS/cm) 916.27 69.561 1 973.02 34.456 1 011.11 66.411 3 010.2

Turbidity (NTU) 104 63.380 198 30.282 143 49.648 284

Dissolved oxygen, DO (mg/L) 4.11 47.975 5.65 28.481 6.4 18.987 7.9

Total dissolved solids, TDS (mg/L) 642.26 66.919 1 302.18 32.927 768.36 60.423 1 941.45

Total solids, TS (mg/L) 849.26 79.602 1 477.18 64.520 949.36 77.198 4 163.45

Total suspended solids, TSS (mg/L) 207 90.684 175 92.124 181 91.854 2 222

Oil and grease, O&G (mg/L) 3.3 48.031 4.92 22.520 4.76 25.039 6.35

Cadmium, Cd (mg/L) 0.012 62.500 0.016 50.000 0.015 53.125 0.032

Copper, Cu (mg/L) 0.013 56.667 0.017 43.333 0.019 36.667 0.03

Lead, Pb (mg/L) 0.7 57.576 0.85 48.485 1.2 27.273 1.65

Iron, Fe (mg/L) 15.2 58.402 22.42 38.643 24.6 32.677 36.54

Average pollutant removal efficiency (%) 60.465 41.650 46.387

RESULTS AND BRIEF DISCUSSION

The fabrication of these filter columns
shown in Figure 4 was done with ease as
the materials (100 mm diameter PVC pipe,
stand, flow regulator, flexible hose, etc)
were sourced locally in the study areas.
A unit of the treatment facility was made
up of three filter columns (Figure 4(a)).
A total of three units of treatment facili-
ties were constructed for each of the five
automobile workshops. The filter stands
were constructed with steel rods as shown
in Figure 4(b). The pollutant removal
efficiencies of the different low-cost filter
systems for Automobile Workshop 4 for
a preliminary period of three weeks with
respect to each parameter are presented in
Tables 2–4. It is suggested that spent rice

70

(a) (b) 60
Average pollutants removal (%)

50

40

30

20

10

0
GAC-RH RH GR-GAC
(c) GAC (c) RH (c) Gravel
Filter media
Figure 4 C
 onstructed / fabricated filter columns: (a) constructed filter column, (b) filter column
setup with stands, (c) filter column with different filter materials, from left to right: GAC, Figure 5 A
 verage pollutant removal efficiencies
RH and gravel of the different filter media

Journal of the South African Institution of Civil Engineering Volume 63 Number 1 March 2021 65
husk and GAC can be used as source of
heat energy for local industries after sun-
drying, while the gravel can be used for
concreting. The impact of the filters dis-
posal on the environment will be reduced.
Tables 2, 3, and 4 show the computed
average performance of these filter systems
in the treatment of the stormwater runoff
in terms of pollutant removal efficiency.
Figure 5 shows that the combined GAC–
RH filter system performed best with
average Te ≈ 58%. The RH filter system and
the combined Gravel–GAC filter systems
followed the combined GAC–RH filter
system with average Te ≈ 49% and average
Te ≈ 40% respectively.
CONCLUSION AND
RECOMMENDATION
The results obtained from the use of this
technology showed that the combined
GAC and RH filter system performed best
in the pollutant removal efficiency (58%).
The RH filter (49%) and the combined GR
and GAC (40%) filter systems respectively
followed in pollutant removal efficiency.
This technology can be upscaled for use
in other environments, with modifications
in the available filter materials and the
need for sustainable use of the technology.
Other cheap locally available agricultural
waste materials, such as sugarcane bagasse,
groundnut/melon shells, maize cobs, etc,
can be explored for use as filter materi-
als. It is highly recommended that other
low-cost technology for stormwater runoff
treatment, especially for automobile work-
shops at large scale and in-situ, be explored,
researched and prototypes rolled out.
REFERENCES
Andersen, J E H 2016. Hydraulic performance of
advanced treatment media to improve quality
of stormwater from airports exposed to de-icing
chemicals. MSc Dissertation. Trondheim, Norway:
Norwegian University of Science and Technology.
66
APHA (American Public Health Association) 2017.
Standard Methods for the Examination of Water
and Wastewater. 22nd ed. Washington, DC: APHA.
Bahgat, M, Dewedar, M A & Zayed, A 1999. Sand-
filters used for wastewater treatment: Buildup and
distribution of microorganisms. Water Research,
33(8): 1949–1955.
Baker, F S, Miller, C E, Repik, A J & Tolles, E D 1992.
Activated carbon. In Kirk Othmer Encyclopaedia of
Chemical Technology, Wiley Online Library.
Cecen, F & Aktas, Ö 2011. Activated Carbon for
Water and Wastewater Treatment: Integration of
Adsorption and Biological Treatment. New York:
Wiley.
Chukwudebelu, J A, Igwe, C C & Madukasi, E I
2015. Prospects of using whole rice husk for the
production of dense and hollow bricks. African
Journal of Environment, Science and Technology,
9(5): 493–501.
Diaper, E W J 1965. Upflow and downflow filtration
through graded media. Proceedings of the Institution
of Civil Engineers, 30(2): 437.
Ekiye, E & Zejiao, L 2010. Water quality monitoring
in Nigeria. Case study of Nigeria’s industrial cities.
Journal of American Science, 6, 22–28.
Idu, A J 2015. Threats to water resources development in
Nigeria. Journal of Geology and Geophysics, 4: 205.
Karthik, R G H, Abhishek, N P, Kishor, V R, Nikhil, E,
Poornima, K B & Shivakumara, B 2018. Comparative
analysis of locally available adsorbents for
purification of water. International Research Journal
of Engineering and Technology, 5(5): 1722–1726.
Khatun, A, Bhattacharyya, K G & Sarma, H P 2014.
Levels of pollutants in runoff water from different
land uses in Guwahati City, Assam, India. Archives
of Applied Science Research, 5: 96–100.
Lakhote, A, Ahire, P, Dabholkar, P, Dhadambe, V &
Gharat, S 2016. Comparative analysis of design of
water filter for rural areas. International Journal
of Innovative Research in Advanced Engineering,
3(12): 23–27.
Lowe, J, de Leon, D, Collins, J, Hoover, R & Book, S
2018. Standard Operating Procedure for Collecting
Grab Samples from Stormwater Discharges,
Vol 1.1. Publication No 18-10-023. Olympia,
WA: Washington State Department of Ecology.
https://fortress.wa.gov/ecy/publications/
summarypages/1810023.html.
Volume 63 Number 1 March 2021
Maebh, A G 2016. Development of filtration
technologies for effective, cost-efficient and robust
water treatment. PhD Thesis. Galway, Ireland:
National University of Ireland.
McDougall, G J 1991. The physical nature and
manufacture of activated carbon. Journal of
the Southern African Institute of Mining and
Metallurgy, 91(4): 109–120.
Nurul-Amin, M D, Kaneco, S, Kitagawa, T et al
2006. Removal of arsenic in aqueous solutions by
adsorption onto waste rice husk. Industrial and
Engineering Chemistry Research, 45: 8105–8110.
Pitt, R, Field, R, Lalor, M & Brown, M 1995. Urban
stormwater toxic pollutants: Assessment, sources
and treatability. Water Environment Research,
67(3): 260–275.
Pratap, M R, Khambhammettu, U, Clark, S E & Pitt, R
2007. Stormwater polishing: Upflow vs downflow
filters. Proceedings, World Environmental and
Water Resources Congress, American Society of
Civil Engineers, 15–19 May, Tampa, FL.
Shrestha, D 2017. Characterization and modelling
of stormwater for the City of Calgary. MSc
Dissertation. Canada: University of Calgary.
Sukia, S K 2020. Rice husk-derived adsorbents for
water purification, Chapter 6. In Naushad, M &
Lichtfouse, E (Eds.), Green Materials for Wastewater
Treatment, Cham, Switzerland: Springer.
USACE (United States Army Corps of Engineers)
2001. Adsorption Design Guide. Guide 1110-1-2. US
Department of the Army.
Wang, S, He, Q, Hainan, A, Wang, Z & Zhang, Q
2013. Pollutant concentrations and pollution loads
in stormwater runoff from different land uses in
Chongqing. Journal of Environmental Sciences,
25(3): 502–510.
Wegelin, M 1996. Surface Water Treatment by
Roughing Filters. A Design, Construction and
Operation Manual. St. Gallen, Switzerland: SKAT
Consulting Ltd.
Xu, M, Liu, X, Tang, Q, Qu, R & Xu, Q 2013.
Utilization of rice husks modified by
organomultiphosphonic acids as low-cost
biosorbents for enhanced adsorption of heavy metal
ions. Bioresource Technology, 149: 420–424.
Zunaira, A & Zhi, C 2017. Removal of arsenic from
drinking water using rice husk. Applied Water
Science, 7: 1449–1458.
Journal of the South African Institution of Civil Engineering
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 journal publishes research papers covering all the
disciplines of civil engineering (structural, geotechnical,
railway, coastal/marine, water, construction, environ-
mental, municipal, transportation) and associated topics
that are relevant to the civil engineering profession, and
that preferably have particular relevance to civil engi-
neering 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
QQ Technical papers are well-researched, in-depth,
fully referenced technical articles not exceeding
6 000 words in length (excluding tables, illustra-
tions 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 techni-
cal 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.
QQ 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.
QQ 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.
QQ 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
QQ 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
accompanied by an expanded abstract in English.
QQ Original work: Papers and technical notes must
be original contributions. Authors must affirm
that submitted material has not been published
previously, is not under consideration for publica-
tion 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 publication 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 obtaining permis-
sion from copyright owners when reproducing mate-
rial that had been published elsewhere. Proof of such
permission must be supplied.
SUBMISSION PROCEDURES
AND REQUIRED FORMAT
QQ Online submission: Manuscripts must be uploaded
as PDF files (http://journal.saice.org.za). Individual
file sizes may not exceed 10 MB. Should you experi-
ence problems uploading your paper, please contact
the editor (verelene@saice.org.za).
QQ 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.
QQ 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.
QQ Figures, tables, photos and illustrations: These
should preferably be submitted in colour, as the
journal is a full-colour publication.
QQ Their positions should be clearly marked in the
text as follows: [Insert Figure 1].
QQ Figures, tables, photos, illustrations and equa-
tions should be numbered consecutively and
should appear in the text directly after they have
been referred to for the first time.
QQ 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).
QQ Only those figures and photographs essential to
the understanding of the text should be included.
All illustrations should be referred to in the text.
QQ 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.
QQ 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.
QQ Mathematical expressions and presentation
of symbols:
QQ 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.
QQ Symbols should preferably reflect those used in
Microsoft Word Equation Editor or Mathtype,
or should be typed using the Times New Roman
symbol set.
QQ 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, symbols of metric measurement units
(m for metres, s for seconds, etc) and mathemati-
cal/trigonometrical 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.
QQ Metric measurement abbreviations/units should
conform to international usage – the SI system of
units should be used.
QQ Decimal commas may be used, but decimal
points are preferred.
QQ 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.
QQ 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
QQ 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:
QQ Donald, A W 1954. Pore pressure coefficients.
Géotechnique, 4(4):143–147.
QQ Terzaghi, K 1943. Practical soil mechanics. New
York: Wiley.
QQ 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.
QQ 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.
QQ 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
QQ 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.
QQ 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.
QQ 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.