Volume 62 Number 3

of the South African Institution of Civil Engineering September 2020

QQ  ater meter replacement and its impact on water

W
consumption, utility revenue and payback periods
QQ L aboratory performance and modelling behaviour
of hot-mix asphalt with recycled crushed glass
QQ  uantifying the economic benefits of gauge changes
Q
on the South African core railway network
QQ  hysical model study of bedrock scour downstream
P
of dams due to spillway plunging jets
QQ  he effect of lane blockage on signalised intersection
T
performance – analysing and modelling
Volume 62  No 3  September 2020 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
joint EditorS-in-chief 2 Water meter replacement and its impact on water
Prof Chris Clayton
University of Southampton consumption, utility revenue and payback periods
c.clayton@soton.ac.uk
Prof Gerhard Heymann
M Brinkley, A A Ilemobade, M Ncube
University of Pretoria
gerhard.heymann@up.ac.za
Editor 10 Laboratory performance and modelling behaviour
Verelene de Koker
Tel +27 11 805 5947, Mobile +27 83 378 3996
of hot-mix asphalt with recycled crushed glass
verelene@saice.org.za
T B George, J K Anochie-Boateng, K J Jenkins
journal editorial Panel
Prof Y Ballim – University of the Witwatersrand
Prof W Burdzik – University of Pretoria
Prof CRI Clayton – University of Southampton 23 Quantifying the economic benefits of gauge
Dr P Day – Jones & Wagener (Pty) Ltd (chairman)
Prof J du Plessis – University of Stellenbosch changes on the South African core railway network
Prof GC Fanourakis – University of Johannesburg
Prof PJ Gräbe – University of Pretoria C P Demmerez de Charmoy, P J Gräbe
Dr C Herold – Umfula Wempilo Consulting
Prof G Heymann – University of Pretoria
Prof A Ilemobade – University of the Witwatersrand
Prof SW Jacobsz – University of Pretoria 36 Physical model study of bedrock scour downstream
Prof EP Kearsley – University of Pretoria
Prof C Roth – University of Pretoria
of dams due to spillway plunging jets
Prof W Steyn – University of Pretoria
Mr M van Dijk – University of Pretoria A Bosman, G R Basson
Prof C Venter – University of Pretoria
Prof A Visser – University of Pretoria
Dr E Vorster – Zutari
Prof J Wium – University of Stellenbosch
53 The effect of lane blockage on signalised intersection
Prof A Zingoni – University of Cape Town
Prof M Zuidgeest – University of Cape Town
performance – analysing and modelling
Peer reviewing E Doğan, E Korkmaz, A P Akgüngör
The Journal of the South African Institution of
Civil Engineering is a peer-reviewed journal
that is distributed internationally
61 Guidelines for the preparation of
Design and reproduction
Marketing Support Services, Ashlea Gardens, Pretoria papers and technical notes
Printing
Fishwicks, Pretoria
Papers for consideration should be submitted
online at: http://journal.saice.org.za
The South African Institution of Civil Engineering accepts no
responsibility for any statement made or opinion expressed
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 Water meter replacement
Journal of the South African
Institution of Civil Engineering and its impact on water
ISSN 1021-2019
Vol 62 No 3, September 2020, Pages 2–9, Paper 0855 consumption, utility revenue
MEGHAN BRINKLEY (MSc, PrEng, PMP,
MSAICE) currently works for SMEC South
and payback periods
Africa (Pty) Ltd and has over 14 years of
experience in the Engineering and Project
Management fields. She is currently M Brinkley, A A Ilemobade, M Ncube
involved with a number of water demand
management and water metering projects.
In these projects, her focus is on data
capturing and project management to ensure accurate records and trend Mechanical water meter accuracy degrades over time due to wear and tear. Based on this
analysis for clients. This article emanated from her MSc research project at premise, many water utilities around the world undertake meter replacement projects.
the School of Civil and Environmental Engineering, University of the In many instances, these projects neither assess the potential or actual impacts of meter
Witwatersrand, Johannesburg.
replacement on consumption and utility revenue, nor undertake meter replacement based
Contact details: on a best practice, customised metering policy. From 2013 to 2015, Johannesburg Water
SMEC South Africa (Pty) Ltd
(SOC) Limited embarked on a project to replace more than 50 000 mechanical water meters
267 Kent Avenue
Ferndale that had not been read over a period of 180 days due to technical faults. This study obtained
Johannesburg 2194 and analysed two databases of these meters and arrived at a sorted and validated database
South Africa comprising 3 679 meter replacements. For each meter replacement, twelve months of water
T: +27 11 369 0647
E: meghan.brinkley@smec.com
consumption data was obtained for the period prior to and after replacement. This study then
investigated the impact of replacement on water consumption, utility revenue and payback
PROF ADESHOLA ILEMOBADE (PhD, MAP, periods. Some high-level highlights from the study include the following: (i) a metering
AMSAICE, MWISA) is currently an Associate policy that addresses such issues as efficient meter database management, and the optimum
Professor at the School of Civil and meter replacement age and/or volumetric reading, is critical for any water utility if it is to
Environmental Engineering, University of
obtain the best results from meter replacement; (ii) in order to guarantee the highest returns
the Witwatersrand, Johannesburg, where
he teaches infrastructure planning and on investment, bulk meters (40 mm and larger) should be prioritised for meter replacement
water engineering subjects at over single-stand residential meters; (iii) the replacement of 20 mm and 25 mm meters with
undergraduate and postgraduate levels. His 15 mm meters resulted in more accurate metering of consumption by virtue of increased
research interests are in the areas of water reuse, water demand, water
conservation and water supply. He has published extensively, received
consumption; and (iv) payback of the 15 mm replacement mechanical meters was achieved
awards for research conducted and supervised, and led several within 12 months, with the replacement of 20 mm and 25 mm meters resulting in average
multi-disciplinary research and consulting teams. monthly revenue increases per property of R266.70 and R203.08 respectively.
Contact details:
School of Civil and Environmental Engineering
University of the Witwatersrand
INTRODUCTION make up for the R300 million reduction
Private Bag 3
Wits 2050 Water meters are considered a water util- in the local government equitable share
South Africa ity’s “cash register” (van Zyl 2011), and in 2016/17 (National Treasury 2016).
T: +27 11 717 7153 in many instances they are the primary Large-scale water meter replacement
E: adesola.ilemobade@wits.ac.za
source of income for many utilities and or installation projects are thus widely
municipalities around the world. In South recognised as an effective means of reduc-
DR MTHOKOZISI NCUBE (PhD, PrEng, PMP,
CP3P, MSAICE, FWISA) is currently a Water Africa, revenue from water sales is one of ing water wastage and usage (Masia &
Specialist with the Development Bank of the primary sources of income for Water Erasmus 2013). Water metering provides
Southern Africa where he advises, consults, Services Authorities, with water sales a fair and equitable basis for charging for
and supports the preparation and financing
being, on average, 20% of all tradable ser- services, and thereby facilitates the means
of water and sanitation projects in
sub-Saharan Africa. His research interests vices and as high as 100% in some munici- by which cost recovery can be achieved
include water metering, apparent water palities (National Treasury 2017; Ncube as required by the Water Services Act 108
losses, water demand and data science in water. He has published several & Taigbenu 2019). Average non-revenue of 1997 (RSA 1997). Only a few studies
articles and actively participates in research and consulting activities.
water in South Africa is, however, estimat- have, however, articulated the impacts of
Contact details: ed to be as high as 1.580 billion m3/‌a nnum meter replacement on a utility’s revenue
Development Bank of Southern Africa
PO Box 1234
(Bhagwan et al 2014), which is approxi- (Selsky 2001).
Halfway House 1685 mately one third of water being supplied. Mechanical water meters, which
South Africa This translates to a loss of over R7 billion constitute the majority of meters used
T: +27 11 313 3287
per year based on an average bulk water in South Africa and many jurisdictions
E: mthokozisin@dbsa.org
tariff of approximately R5/m3. It was around the world, experience a degrada-
estimated that, reducing these losses by tion of accuracy over time (Stoker et al
Keywords: water meter replacement, consumption, revenue, about 4.6% through technical interventions 2012), mainly due to wear and tear of
payback (such as water meter management), could the moving parts. This degradation is a

Brinkley M, Ilemobade AA, Ncube M. Water meter replacement and its impact on water consumption, utility revenue and payback periods.
2 J. S. Afr. Inst. Civ. Eng. 2020:62(3), Art. #0855, 8 pages. http://dx.doi.org/10.17159/2309-8775/2020/v62n3a1
function of several factors, such as poor
water quality, abnormal water veloci-
ties, excessive throughput volumes or
flow rates, sub-optimal installation and
poor meter handling. The assumption
that meter ageing, on its own, makes
water meters less accurate, leads to the
erroneous hypothesis that revenues are
lost because the consumption of water
is not accurately recorded by ageing
meters (Arregui et al 2006; Shields et al
2012; Couvelis & Van Zyl 2015; Arregui
et al 2016).
Based on the degradation premise
above, several water utilities have, in more
recent times, implemented large-scale
meter replacement projects. Replacing
water meters that are still providing rea-
sonably accurate recordings is, however,
wasteful and futile. According to several
authors (including Allender 1996; Davis
2005; Arregui et al 2006; Fantozzi 2009;
Viyakesparan 2010; Shields et al 2012;
Couvelis & Van Zyl 2015; Fontanazza et
al 2015; Moahloli et al 2019), there is a
point that economically justifies the cost
of meter replacement. Prior to that point,
meter replacement is unwarranted and
would qualify as wasteful expenditure.
In many of the above studies the age of
meters and the total volume of water
through the meters were parameters that
determined meter replacement. Table 1
summarises some of the studies in lit-
erature and their recommendations on
optimum meter replacement periods and/
or volumes.
The above studies recommend that
water utilities undertake investigations to
determine the parameter(s) most influen-
tial to optimum meter replacement, based
on prevailing site conditions and water
tariff structures. Taking into considera-
tion studies related specifically to revenue
generation and meter replacements, it
should be noted that Davis (2005) further
quantified the optimum replacement point
by utilising revenue generation and replace-
ment cost (i.e. payback period) as the basis
for comparison (i.e. when the lost revenue
exceeds the replacement cost), but this is
utility specific.
In Moahloli et al (2019), the average
volume of apparent losses due to meter
inaccuracies was calculated to be 1.8 kℓ
per meter per month. As shown in Table 1,
Moahloli et al (2019) also found that,
by using the Net Present Value Chain
(NPVCn) method, the optimum period for
meter replacement was 9, 12 and 16 years
Journal of the South African Institution of Civil Engineering  Volume 62  Number 3  September 2020
Table 1 Meter age and/or volume of water through the meter as basis for meter replacement
Proposed basis for meter
Reference Area of study
replacement
Bold (2013) Age – 10 years South Africa
Susanville Water
Age – 20 to 25 years California
Department (2010)
Bold (2013) Age – 6 years Germany
McNally (2008) Age – 9 to 13 years City of London
Ncube & Taigbenu
Volume – 8 to 10 Mℓ South Africa
(2015)
Mutikanga (2012) Volume – 8 Mℓ Uganda
Davis (2005) Volume –1.42 million gallons (5.3 Mℓ) Arizona
Fantozzi (2009) Age – 9 to 19 years dependent on inputs Four Italian Utilities
Viyakesparan (2010) Volume – 3.9 Mℓ Regional Council of Australia
Allender (1996) Age – 16 to 30 years Anne Arundel Country (Maryland)
The Region of Peel
Age – 20 years Canada
(2018)
9, 12 and 16 years of age and total
registered volumes of 3.971 Mℓ, 5.162 Mℓ
Moahloli et al (2019) South Africa
and 6.750 Mℓ at 10%, 8% and 6%
discounted rates respectively
at the total registered volumes of 3.971 Mℓ, LARGE-SCALE WATER METER
5.162 Mℓ and 6.750 Mℓ, and at 10%, 8% REPLACEMENT PROJECT
and 6% real discount rates, respectively. JW is a City of Johannesburg Municipality
The above is based on the premise that Owned Entity with the mandate to supply
meters are read accurately from the date of potable water and sanitation services. In
installation. accordance with the City of Johannesburg
Mukheibir et al (2012) was able to Metropolitan Municipality Bylaws (City of
quantify the volume of non-registered flow Johannesburg 2018b), all water consumers
to be 1% of the average daily registered on a Level of Service 3 (i.e. municipal piped
consumption for a new meter and 3.5% water supply to a stand) must be metered.
of the average daily registered consump- Furthermore, all consumer meters within
tion for an aged meter. Mukheibir et al the jurisdiction of JW must always be fully
(2012) further computed the average operational and accessible for the recording
under-registration to be between -0.4% and of consumers’ consumption. The Bylaws,
+0.1%, changing from over-registration however, allow for the estimation of water
to under-registration as the meter aged. consumption for a period of up to 180 days.
From their study, Mukheibir et al (2012) In keeping with the above requirements,
concluded that, based on the unregistered JW implemented a project that replaced
percentages, it was not cost-effective to more than 50 000 meters throughout its
replace water meters if they registered less area of jurisdiction between 2013 and 2015.
than 250 kℓ/yr. These meter replacements were aimed at
In tandem with many of the above addressing the backlog of water meter esti-
highlighted studies that investigated the mations (for billing purposes) in excesses
parameters impacting on optimum meter of the required maximum of 180 days. The
replacement, this study investigated a replaced meters included:
large-scale water meter replacement pro- QQ meters that could no longer be located
ject undertaken by Johannesburg Water QQ meters with faded or frosted dials
(SOC) Limited (JW). The impact that QQ inaccessible meters
water meter replacement has had on water QQ meters requiring maintenance, includ-
consumption, with a particular focus on ing damaged meter boxes and lids
revenue and the payback periods for the QQ meters suspected to have stopped.
investment, was also evaluated. Based on Two databases formed the basis for the
results from the study, recommendations analysis undertaken in this study. The
regarding optimum meter replacement first database (D1) contained a list of
are proffered. residential water meters obtained from
3
Table 2 Summary details of meter changes
1 000
Number of
Old New
Group

properties
meter size meter size
with meter
(mm) (mm)
changes

Number of replaced meters

1 20 15 2 531 100
2 25 15 1 071

3 20 40 2

4 25 80 1 10
5 40 15 2

6 40 40 15

7 40 50 5
1

0–1 000

1 001–2 000

2 001–3 000

3 001–4 000

4 001–5 000

5 001–6 000

6 001–7 000

7 001–8 000

8 001–9 000

9 001–10 000

10 001–11 000
8 50 40 1

9 50 50 22

10 80 80 9
Meter reading range (kℓ)
11 100 100 20 20 mm 25 mm

Total 3 679
Figure 1 Number of meters changed per meter reading range

the JW Meter Information System (Ncube
2019). The consumption records in D1
were from July 2003 to July 2015. The
second database (D2) was collated by the
contractors who undertook the JW large-
scale meter replacement project between
2013 and 2015.
D1 and D2 were sorted and validated
(i.e. the records used in the analysis
were those where the old and new meter
serial numbers were exact matches on
both databases). From the 50 000 meter
replacements, the sorting and validation
exercise produced a third database (D3)
comprising 3 679 meter changes (see
Table 2). This low percentage (7.4%) of
validated meter replacements highlights
a significant challenge experienced by
JW and many water utilities in the man-
agement of meter information to allow
rigorous analysis and the drawing of
relevant conclusions.
The following parametric relationships
were investigated using data in D3:
QQ Replaced meter sizes and their impact
on consumption and utility revenue
QQ Change in average monthly consump-
tion against meter age
QQ Payback periods of replacement meters.
While it was the initial intention of JW
to replace meters size-for-size (i.e. a new
50 mm meter replaces an old 50 mm
meter), a subsequent decision was taken
to replace all 20 mm and 25 mm meters
with 15 mm meters, which the city had
recently adopted. It should be noted that
the majority of the meters replaced were 1 000 kℓ comprised 28% (1 042) of D3.
below-ground installations, which were In total therefore, 14% of the meters in
changed for above-ground installations, for D3 were 10 years old and had a read-
ease of location by meter readers. While ing less than or equal to 1 000 kℓ. This
the change in design or layout was not translates to an average consumption of
taken into consideration when comparing 8.33 kℓ per month and differs significantly
pre- and post-intervention consumption, it from the average monthly consumption
is not expected to have any impact on the of 30 kℓ of Johannesburg residential
consumption figures. properties (Ncube 2019). It is likely that
Groups 1 and 2 of Table 2 comprise a these meters clocked over during their
significant percentage (98%) of the meters 10-year existence and thus introduced
changed, and are the focus of this study, certain biases into subsequent analyses.
as the numbers of the rest of the meters If the meters had clocked over, this would
were insignificant. Meter class could not be imply that the overall cumulative volume
considered, as meter class information was that had passed through the meter was
unknown for the majority of the meters. significantly higher than the last reading
For each of the 3 679 meter changes in recorded, potentially including an addi-
D3, twelve months of water consumption tional 10 000 kℓ onto the recorded volume,
data before and after the meter change was depending on meter type and number of
obtained from D1. For each group of meter dials on the meter. This would therefore
changes, the variation in consumption after significantly alter the average monthly
the meter change was determined together consumptions, based on the age of the
with the resulting change in revenue. meter. Due to the non-recording of meter
Revenue was calculated using the City of clock-over data, this study is thus unable
Johannesburg’s 2017/18 block water tariffs to validate and correct these in D3.
(City of Johannesburg 2018a).
Replaced meter sizes and
their impact on consumption
ANALYSES AND DISCUSSION and utility revenue
Figures 1 and 2 show the number of Table 3 summarises the overall impact
meters changed per meter reading range of meter replacements on consumption
and meter age respectively for Group 1 and revenue. The replacement of 20 mm
(20 mm) and Group 2 (25 mm) meters. meters with 15 mm meters resulted in a net
In Figure 2, meters aged 10 years com- increase in average monthly consumption
prised 51% (1 873) of D3, while meters (8.87 kℓ) and revenue (R266.70) per meter.
with volumetric readings between 0 and This is an outcome observed in Ncube and

4 Volume 62  Number 3  September 2020  Journal of the South African Institution of Civil Engineering
 10 000

1 000
Number of replaced meters

100

10

1
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 31 32 33
Meter age (years)
20 mm 25 mm

Figure 2 Number of meters changed per meter age

Taigbenu (2019) who published profiles to 11 had a statistically insignificant before that optimum would negatively
for the typical Johannesburg residential number of meter changes, it is important impact a water utility’s revenues and would
consumer showing that 15 mm meters to note that the replacement of meters not result in marked improvements in the
were ideal for this user category. The larger than 25 mm resulted in notably accuracy of recorded consumption.
replaced 20 mm meters were, therefore, larger changes in average consumption per Based on mechanical meter degradation
over-sized, and hence under-registered meter change per month (and therefore over time, and after meter replacement,
consumption. Similarly, the replacement of revenue generated or lost) in comparison it is expected that the change in average
25 mm meters with 15 mm meters resulted to meters in Groups 1 and 2. It there- monthly consumption would be larger for
in a net increase in average monthly con- fore makes sense for water utilities to older meters or meters with larger volu-
sumption (6.01 kℓ) and revenue (R203.08) prioritise bulk meter replacement, when metric readings. This trend should occur
per meter, indicating that the 25 mm appropriate, over meters servicing single up to a turning point where no appreciable
meters were also over-sized, and hence residential properties. increase is evident. The turning point indi-
under-registered consumption. cates the optimum point at which a meter
Table 4 presents the impact of replaced Change in average monthly should be replaced.
meter sizes on consumption and utility consumption against meter age Figure 3 shows the change in average
revenue for each group within D3. The As mechanical meters degrade with time, monthly consumption after meter replace-
table shows that there is generally a net it is financially wise for a water utility to ment for meters aged 2 to 33 years. Figure 3
increase in consumption, and conse- determine the optimum age and/or volu- may have provided the answer to the
quently increased revenue across all meter metric reading when mechanical meters question: At what age should residential
groups except Group 8. While Groups 3 should be replaced. Replacing meters meters be replaced? Sadly, it does not, due

Table 3 Overall impact of meter replacements on consumption and revenue

Old average monthly New average monthly Change in average Change in average % of meters with
Group consumption consumption monthly consumption monthly revenue increase in monthly
(kℓ) (kℓ) (kℓ) (rand) consumption

1 29.55 38.42 8.87 ± 1.82 266.70 ± 66.98 65

2 43.74 49.75 6.01 ± 1.93 203.08 ± 125.56 59

Journal of the South African Institution of Civil Engineering  Volume 62  Number 3  September 2020 5
Table 4 Impact of replaced meter size on monthly consumption and utility revenue per group
Change in Change in

consumption
Number average average Total revenue Total revenue

Change in
Revenue Revenue
Group

Meter change of meter monthly monthly per month per month

difference difference
description changes or consumption revenue per (old meters) (new meters)
(rand) (%)
properties per property property (rand) (rand)
(kℓ) (rand)

Increase 1 654 24.16 753.81

1 20 mm to 15 mm 1 601 685.56 2 276 713.24 675 027.69 42
Decrease 877 –19.95 –651.96
Increase 629 25.14 862.27
2 25 mm to 15 mm 1 171 264.87 1 386 554.87 215 290.00 18
Decrease 442 –21.22 –740.00
3 20 mm to 40 mm Increase 2 131.76 5 097.17 12 368.18 22 562.52 10 194.34 82
4 25 mm to 80 mm Increase 1 65.19 2 413.83 509.20 2 923.02 2 413.83 474
5 40 mm to 15 mm Increase 2 7.31 127.32 311.89 566.53 254.64 82
Increase 9 160.84 6 201.62
6 40 mm to 40 mm 400 633.32 450 317.57 49 684.26 12
Decrease 6 –26.39 1 021.72
Increase 3 101.01 3 911.08
7 40 mm to 50 mm 70 382.50 76 857.21 6 474.71 9
Decrease 2 –67.9 –2 629.27
8 50 mm to 40 mm Decrease 1 –101.91 –3 945.92 15 264.14 11 318.22 3 945.92 –26
Increase 13 148.02 5 725.38
9 50 mm to 50 mm 319 640.08 358 987.28 39 347.19 12
Decrease 9 –101.07 –3 898.09
Increase 6 38.83 1 423.36
10 80 mm to 80 mm 127 990.84 131 240.36 3 249.53 3
Decrease 3 –45.62 –1 763.54
Increase 15 226.42 8 737.21
11 100 mm to 100 mm 396 747.77 520 216.09 123 468.32 31
Decrease 5 –39.84 –1 517.95
Total 3 679 4 116 798.34 5 238 256.92 1 121 458.57 27

35

30
Change in average monthly consumption (%)

25 y = 0.0344x – 0.0735
R2 = 0.4619
20

15 y = –0.0005x2 – 0.0097x – 0.0807

R2 = 0.1348
10

–5

–10

–15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33
Age of meters
2 to 10 year old meters 2 to 33 year old meters Increasing number of meters in each age group

Figure 3 Change in average monthly consumption against meter age

6 Volume 62  Number 3  September 2020  Journal of the South African Institution of Civil Engineering
 40

35

30

25
Change in average monthly consumption (%)

y = –0.0005x2 + 0.0097x + 0.0807

20 R2 = 0.1348

15

10

y = 0.0363x0.8991
5 R2 = 0.7859

–5

–10

–15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33
Age of meters
4 to 10 year old meters 4 to 33 year old meters Increasing number of meters in each age group

Figure 4 Change in average monthly consumption against meter age (excluding groups below 20 meters and outliers)

to the poor R2 value of 0.1348, which may time (in this instance, in months) required NCFt = net cash flow (i.e. inflows minus
be influenced by the very likely possibility for the investment in meter replacement outflows) during a single period
of meter clock-overs after 10 years of age. to be recovered, and was calculated using of time, t.
As can be seen in Figure 3, after 10 years Equation 1 (Khan 1999):
of meter age, the data for average monthly Table 5 shows the payback period results
consumption is disordered. There is, how- n NCFt against meter age for all the 15 mm meters
NPV = ∑t=0 (1)
ever, a notable increase in average monthly (1 + r)t installed. All calculated negative payback
consumption for replaced meters aged 2 to period values (specifically for 3 and 6-year
10 years with an R2 value of 0.4619. This is Where: old meters) were converted to zero, as it
further confirmed when meter age groups n = number of months/periods is impossible to have negative payback
with less than 20 meters and outliers (i.e. r = discount rate (taken as 5.3%) periods, but rather implies that there was
groups with negative changes in average
monthly consumption) are excluded from Table 5 Payback periods for 2 to 10-year old 15 mm meters
the graph. The result is an R2 value of Number of 15 mm
0.7859 as can be seen in Figure 4. While Total number of Payback period
Meter age replacement meters
replaced meters in D3 (months)
inconclusive, due to the negative change (purchase cost = R2 708)
in average monthly consumption values at 2 10 10 66.26
3 and 6 years of meter age, the best fit line
3 12 10 0
for replaced 2 to 10-year old meters lends
itself to the earlier explanation of how the 4 46 42 11.81
accuracy of meters degrade over time.
5 54 49 8.49

Payback periods of 6 70 66 0
replacement meters 7 96 87 7.48
Payback periods (using 2018 meter replace-
ment costs) were calculated for 15 mm 8 128 119 8.76

meters, which comprised 98% of meter 9 141 135 7.41

replacements within D3, aged 2 to 10 years.
10 1 884 1 875 8.35
Payback period represents the length of

Journal of the South African Institution of Civil Engineering  Volume 62  Number 3  September 2020 7
immediate recouping of meter replacement
investments.
Excluding the zero payback period
values and plotting payback period (Y)
against meter age (X) results in a best fit
power graph of Y = 98.723 X–1.231 with
an R2 value = 0.78. The relatively high
R2 value implies that, as meters aged (from
2 to 10 years), higher meter reading inac-
curacies resulted, and this led to the under-
registration of consumption. As a result of
the new replacement meters being more
accurate in registering consumption, there
would be increased revenues, and conse-
quently shorter payback periods. Excluding
2-year old meters in Table 5 and meters
with immediate payback, payback for
15 mm replacement meters was achieved
within 12 months.
Payback periods were also calculated
and plotted for volumetric ranges as shown
in Table 6. The best fit line shows a weak
exponential (Y = 7.6696e–0.1239X with an
R 2 value = 0.41) increase in payback period
with increasing volumetric range. This is
counter-intuitive, i.e. replaced meters with
readings of less than 1 000 kℓ resulted
in lower payback periods than higher
volumetric ranges immediately adjacent.
As was explained previously, this is likely
attributed to meter clock-overs, where the
recorded meter readings were not actual,
but rather readings subsequent to the
meter registering its maximum value.
The longest payback period was
expected to occur in the higher volumetric
regions (volumetric readings >5 200 kℓ).
However, a comparison was done between
chronological age and volumetric age, and
a large number of the meters which were
10 years and older recorded low volumetric
readings (<1 000 kℓ). This indicates that
the chronological age of the meter may be
contributing to the increase in total con-
sumption, rather than the volumetric age of
the meter. This is contrary to the results of
Table 6 Payback periods for 15 mm meters based on volumetric ranges
Total number
Volumetric
of replaced
ranges (kℓ)
meters in D3
0 – 1 000 711
1 000 – 2 000 629
2 000 – 4 000 899
4 000 – 5 200 348
5 200 – 6 750 326
>6 750 766
8 Volume 62  Number 3  September 2020  Journal of the South African Institution of Civil Engineering
other studies and may be explained by the
meter clock-over and should therefore be
used with caution.
HIGHLIGHTS AND
RECOMMENDATIONS
This paper evaluated a Johannesburg
Water (SOC) Limited large-scale meter
replacement project that was implemented
between 2013 and 2015. While over 50 000
meters were replaced during the two-year
project, the final database employed in
this study only contained 3 679 validated
meter replacements. Some distinct findings
and conclusions from this study are noted
below and can assist in meter management:
QQ While in general the meter replace-
ment project yielded positive benefits
in terms of accurate consumption
readings, and therefore increased
revenue, there is a need for JW, and
indeed for all water utilities, to develop
or evaluate their metering policies, as
there were several meter replacements
in this study which resulted in negative
revenues. Replacement of meters at
the optimum replacement age and/or
volumetric reading is beneficial in the
long run. Also, the storing of validated
meter information (e.g. identity, class,
clock-over, installation and/or replace-
ment details) by water utilities enables
critical evaluation and research that
would produce invaluable strategic and
operational information. Unfortunately,
due to suspected meter clock-overs,
conclusive answers could not be
obtained regarding the parameter(s)
best suited to determine optimal meter
replacement.
QQ Calculated revenue due to the replace-
ment of bulk meters (meters larger
than 25 mm) was higher, on average,
than for residential meters on a single
stand. Therefore, bulk meters should be
Number of 15 mm
replacement Payback period
meters (purchase (months)
cost = R2 708)
711 8.41
629 11.26
894 11.45
342 8.11
320 21.37
707 14.60
prioritised for meter replacement over
single stand residential meters.
QQ Meter-sizing impacts the overall perfor-
mance of the meter. The replacement
of 20 mm and 25 mm meters with
15 mm meters in this study, in general,
resulted in more accurate metering of
consumption by virtue of increased
consumption. The better suitability of
15 mm meters for Johannesburg resi-
dential stands is confirmed by Ncube
and Taigbenu (2019).
QQ There was a general increase in revenue
from the replacement of older meters.
Therefore, by replacing meters of
which the ages are close to the deter-
mined optimum, higher revenues are
envisaged.
QQ In general, for 15 mm meter replace-
ment, payback was achieved within
12 months.
ACKNOWLEDGEMENT
The authors are grateful to Johannesburg
Water (SOC) Limited for permission to use
their data in this study.
REFERENCES
Allender, H D 1996. Determining the economical
optimum life of residential water meters. Water
Engineering and Management, 143(9): 20–24.
Arregui, F, Cabrera, E, Cobacho, R & García-Serra, J
2006. Reducing apparent losses caused by meters
inaccuracies. Water Practice and Technology, 1(4).
Arregui, F, Balaguer, M, Soriano, J & García-Serra, J 2016.
Quantifying measuring errors of new residential water
meters considering different customer consumption
patterns. Urban Water Journal, 13(5): 463–475.
Bhagwan, J, Wegelin, W, Mckenzie, R & Wensley, A
2014. Counting the lost drops: South African’s
study into non-revenue water. Water Practice and
Technology, 9(502): 52–508.
Bold, B 2013. Metering specialist offers local
municipalities water loss solution. Engineering
News, 8 November 2013. Available at: http://
www.engineeringnews.co.za/article/metering-
specialist-offers-local-municipalities-water-loss-
solution-2013-11-08 (accessed on 27 March 2019).
City of Johannesburg 2018a. Approved tariffs. Available
at: https://www.joburg.org.za/documents_/
Documents/Other/traffifs/Approved%20tarrifs.
compressed.pdf (accessed on 29 September 2019).
City of Johannesburg 2018b. City of Johannesburg
Metropolitan Municipality Water Services
By-laws. Available at: https://www.joburg.org.
za/documents_/By-Laws/Documents/bylaws/
water%20services%20bylwas.pdf (accessed on
9 March 2019).
Couvelis, F & Van Zyl, J 2015. Apparent losses due to
domestic water meter under-registration in South
Africa. Water SA, 41(5): 698–703.
Davis, S 2005. Residential water meter replacement
economics. Available at: http://rash.apanela.com/
tf/leakage/Residential%20Water%20Meter%20
Replacement%20Economics.pdf (accessed on
8 February 2018).
Fantozzi, M 2009. Reduction of customer meters
under-registration by optimal economic
replacement based on meter accuracy testing
programme and unmeasured flow reducers.
Proceedings, 5th IWA Water Loss Reduction
Specialist Conference, Cape Town, pp 233–239.
Fontanazza, C, Notaro, V, Puleo, V & Freni, G 2015.
The apparent losses due to metering errors: A
proactive approach to predict losses and schedule
maintenance. Urban Water Journal, 12(3): 229–239.
Khan, M Y 1999. Theory and Problems in Financial
Management. Boston, MA: McGraw Hill Higher
Education.
Masia, O A & Erasmus, L D 2013. Smart metering
implementation for enabling water conservation
and water demand management: An Investigation
in Gauteng, South Africa. Proceedings,
AFRICON 2013: IEEE African Conference on
Sustainable Engineering for a Better Future,
Mauritius, pp 1–5.
McNally P 2008. Water meter and meter reading
strategy 2008 recommendation report. Report of the
Environment and Transportation Committee 2008.
London.
Moahloli, A, Marnewick, A & Pretorius J H C 2019.
Domestic water meter optimal replacement period
Journal of the South African Institution of Civil Engineering  Volume 62  Number 3  September 2020
to minimize water revenue loss. Water SA, 45(2):
165–173. Available at: https://doi.org/10.4314/wsa.
v45i2.02 (accessed on 21 May 2019).
Mukheibir, P, Stewart R, Giurco, D, O’Halloran, K
2012. Understanding non-registration in domestic
water meters: Implications for meter replacement
strategies. Water (December): 95–100.
Mutikanga, H 2012. Water loss management: Tools
and methods for developing countries. PhD Thesis.
Leiden, The Netherlands.
National Treasury (South Africa) 2016. Budget Review
2016. Division of revenue and spending by provinces
and municipalities. Chapter 6. Available at: http://
www.treasury.gov.za/documents/national%20
budget/2016/review/Chapter%206.pdf (accessed
on 21 June 2016).
National Treasury (South Africa) 2017. Local
government revenue and expenditure: First
Quarter Local Government Section 71 report for
the period 1 July 2017–30 September 2017. Press
Release. Available at: http://www.treasury.gov.
za/comm_media/press/2017/2017120801%20
Press%20Statement%20-%201st%20Q%20S71%20
Publication%20-%2005%20December%202017.
pdf (accessed on 16 May 2018).
Ncube, M 2019. Decision enhancement for the
management of apparent water losses. PhD Thesis,
Johannesburg, University of the Witwatersrand.
Ncube, M & Taigbenu, A 2015. Meter accuracy
degradation and failure probability based on meter
tests and meter change data. Proceedings, 4th YWP-
ZA Biennial Conference and 1st African YWP
(Young Water Professionals) Conference, 16–18
November, Pretoria.
Ncube, M & Taigbenu, A 2019. Assessment of apparent
losses due to meter inaccuracy: A comparative
approach. Water SA, 45(2): 174–182.
RSA (Republic of South Africa) 1997. Water Services
Act, 108 of 1997.
Selsky, P 2001. Meter retrofit yields water savings.
Journal of the American Water Works Association,
93(3): 36–120.
Shields, D, Barfuss, S & Johnson, M 2012. Revenue
recovery through meter replacement. Journal of
the American Water Works Association, 104(4):
E252–E259.
Stoker, D, Barfuss, S & Johnson, M 2012. Flow
measurement accuracies of in-service residential
water meters. Journal of the American Water Works
Association, 104(12): E637–E642.
Susanville (California) Water Department 2010.
Available at: http://susanvillepublicworks.org/
files/65865553.pdf (accessed on 15 Jul. 2015).
The Region of Peel (Canada) 2018. Peel’s 2018 Water
meter replacement program starts today. Available
at: http://www.peelregion.ca/news/archiveitem.
asp?year=2018&month=3&day=10&file=2018310.
xml (accessed on 16 May 2019).
Van Zyl, J 2011. Introduction to integrated water meter
management, 1st edition. Report TT 490/11, Water
Research Commission, Pretoria.
Viyakesparan, J 2010. Maximizing revenue for meter
replacement strategy. Hamilton, New Zealand.
Available at: https://www.waternz.org.nz/Attach
ment?Action=Download&Attachment_id=1778
(accessed on 27 March 2019).
9
TECHNICAL PAPER Laboratory performance
Journal of the South African
Institution of Civil Engineering and modelling behaviour
ISSN 1021-2019
Vol 62 No 3, September 2020, Pages 10–22, Paper 0917 of hot-mix asphalt with
THERESA GEORGE is a researcher in the
Transport Infrastructure Engineering Division
recycled crushed glass
at the Council for Scientific and Industrial
Research (CSIR). She holds an MEng in Civil
Engineering from the University of T B George, J K Anochie-Boateng, K J Jenkins
Stellenbosch. Her field of speciality is in
pavement materials, focusing particularly on
bituminous materials and pavement design.  
In South Africa research is currently under way to determine the suitability of using locally
Contact details: available recycled crushed glass as a partial fine aggregate substitute in the production of
Built Environment
Council for Scientific and Industrial Research (CSIR)
asphalt mixes. This paper characterises the laboratory performance of a dense-graded asphalt
P O Box 395, Pretoria 0001 South Africa wearing course mix consisting of 15% recycled crushed glass. The influence of selected
T: +27 12 841 2924 antistripping additives on moisture susceptibility was specifically assessed as a variable in
E: tgeorge@csir.co.za
the performance evaluation of the glass-asphalt mix as follows: (a) the effect of 1% hydrated
lime, (b) the effect of 0.5% liquid antistripping additive, and (c) the effect without the addition
DR JOSEPH ANOCHIE-BOATENG, who holds a
PhD in Civil Engineering from the University
of antistripping additive. The effect of these variables on the moisture susceptibility of the
of Illinois at Urbana-Champaign, USA, glass-asphalt mix was evaluated using the tensile strength ratio parameter supported with a
joined the Council for Scientific and microscopic imaging analysis. Additionally, the stiffness and permanent deformation properties
Industrial Research (CSIR) in 2008. He
of the glass-asphalt mix that demonstrated optimum resistance to moisture damage was
pioneered and leads the investigation into
the potential utilisation of crushed glass as a compared to the same mix without crushed glass. The performance properties were evaluated
substitute for natural aggregate in asphalt using the Huet-Sayegh model and a polynomial model respectively, which were used particularly
mixes in South Africa. He also led the development of various manuals, to develop performance characterisation models for the glass-asphalt mix. The findings of
guidelines and test protocols to support the pavement and asphalt
this study revealed that an anti-stripping additive is essential to meet moisture susceptibility
industries in South Africa and Tanzania. Through his research work he
developed research topics for postgraduate students from the University criteria and alleviate moisture damage in dense-graded glass-asphalt mixes. In particular,
of Pretoria, Tshwane University of Technology and University of moisture susceptibility was improved using hydrated lime rather than the liquid antistripping
Stellenbosch. He works closely with the South African Road Federation additive. Furthermore, the selected constitutive models were able to effectively characterise
(SARF) on an asphalt mix design course to develop the necessary skills for
southern Africa.  
the laboratory performance of both mixes, with the glass-asphalt mix demonstrating improved
resistance to permanent deformation when compared with the conventional asphalt mix.
Contact details:
Built Environment
Council for Scientific and Industrial Research (CSIR)
P O Box 395, Pretoria 0001, South Africa INTRODUCTION stockpiled and earmarked for disposal to
T: +27 12 841 2947
In South Africa, the National landfill, thereby contributing to the waste
E: janochieboateng@csir.co.za
Environmental Management Waste Act glass that is currently being landfilled.
PROF KIM JENKINS (PrEng, MSAICE, FSAAE)
(Act 59 of 2008) commits the govern- This adds more undue pressure to rapidly
leads the SANRAL (South African National ment to, amongst others, promote “waste depleting landfill space and has led to the
Roads Agency) Chair in Pavement minimisation, reuse, recycling and recov- necessity for adopting sustainable prac-
Engineering within the Geotechnical and
ery of waste”. However, national waste tices. Waste glass that is recovered to be
Transportation Division of the Department
of Civil Engineering at Stellenbosch information indicates that an estimated recycled or re-used is a key component in
University. Under his lead the pavement 77% (approximately 1 000 000 tons) of this approach.
research group also collaborates waste glass generated in South Africa is The road pavement industry in the
internationally, amongst others with the Technische Universiteit Delft
being landfilled, while only 23% is recycled international community has provided a
(Netherlands) and Wuhan University (China). Prof Jenkins serves on
several international scientific committees and participates in (Department of Environmental Affairs number of alternative uses for recovered
international conferences, having published widely in academic journals 2017). This data highlights that a substan- waste glass. The use of crushed glass in
and industry-related publications. He also chairs an international tial amount of waste glass could therefore hot-mix asphalt (HMA) paving applications
committee on recycling of road building materials linked to
sustainable practice.
potentially be recovered or diverted from has been widely implemented in the United
landfill to be recycled or reused. States and Canada since the early 1970s.
Contact details:
Department of Civil Engineering
Additionally, considerable quantities Other countries that have reported using
Pavement Engineering of recycled crushed glass fines (less than crushed glass in asphalt paving applications
Stellenbosch University 5 mm in size) accumulate as stockpiles at include the United Kingdom, Australia,
Private Bag X1, Matieland 7602, South Africa
glass-packaging manufacturing plants in New Zealand, Japan and Taiwan (Yamanaka
T: +27 21 808 4379
E: kjenkins@sun.ac.za Gauteng and the Western Cape provinces et al 2001; Su & Chen 2002; Dane County
of South Africa. These processed glass Department of Public Works 2003; Arnold
Keywords: glass-asphalt, moisture susceptibility, dynamic modulus, fines, which are unusable in the glass- et al 2008; Australian Government 2011;
permanent deformation, constitutive modelling packaging manufacturing process, are Andela & Sorge n.d.).

George TB, Anochie-Boateng JK, Jenkins KJ, Laboratory performance and modelling behaviour of hot-mix asphalt with recycled crushed glass.
10 J. S. Afr. Inst. Civ. Eng. 2020:62(3), Art. #0917, 13 pages. http://dx.doi.org/10.17159/2309-8775/2020/v62n3a2
 Early applications of crushed glass in
hot-mix asphalt pavements in the United
States incorporated glass particles bigger
than 12.5 mm, with quantities in excess
of 25% (Federal Highway Administration
1998). The application of coarse glass par-
ticles (bigger than 5 mm) in large quantities
was considered to be a major contribut-
ing factor to the stripping and ravelling
problems reported in early glass-asphalt
pavement applications. The increased
susceptibility to stripping has also been
attributed to the comparatively smooth sur-
face texture of glass material, which reduces
adhesion at the bitumen-glass interface
(Federal Highway Administration 1998). To
avoid these concerns, various studies have
limited the quantity of crushed glass in the
surface course, with control limits set on
the grading of the glass particles used. More
recently, 10 to 15% crushed glass has been
specified for use in asphalt wearing courses
in the United States, while some countries,
e.g. New Zealand, utilise as little as 5% glass
content (Federal Highway Administration
1998). Hughes (1990) demonstrated that the
addition of 15% crushed glass, with a maxi-
mum particle size (MPS) of 10 mm, had a
minor effect on the moisture susceptibility
of a medium dense-graded glass-asphalt
mix incorporating 1% hydrated lime. The
Virginia Transportation Research Council
(VTRC 1998) also reported on the superior
stripping resistance of a fine dense-graded
asphalt mix incorporating hydrated lime
and consisting of up to 12% crushed glass
(with an MPS of 10 mm), whereas the addi-
tion of a liquid antistripping additive dem-
onstrated a reduction in stripping resistance
with an increase in crushed glass content
(VTRC 1998).
Su & Chen (2002) reported that glass-
asphalt pavement test sections in Taiwan,
incorporating 10% of crushed glass (with an
MPS of 5 mm) indicated no stripping of the
glass particles, and also no rutting of the
glass-asphalt pavement after one year of ser-
vice. Furthermore, improved skid resistance
along the longitudinal and transverse profile
of the glass-asphalt pavement test sections
was reported. Wu et al (2003) showed that
the permanent deformation behaviour
the glass-asphalt mix incorporating 15%
crushed glass indicated higher stiffness
moduli than a traditional dense-graded
asphalt wearing course mix at elevated tem-
peratures. Similar to the observations made
by Su and Chen (2002) and Wu et al (2003),
Lachance-Tremblay et al (2014) concluded
that the addition of 10% crushed glass can
be utilised in asphalt mixes without com-
promising the overall rutting performance
of the asphalt mix. In South Africa, however,
minimal research has been conducted
on the viability of using crushed glass in
asphalt pavement applications.
OBJECTIVE
The objective of this study was to determine
the influence of 15% recycled crushed
glass on the laboratory performance of a
10 mm nominal maximum particle size
(NMPS) medium dense-graded asphalt
wearing course mix. The effect of selected
antistripping additives on the moisture
susceptibility of three such glass-asphalt
mixes (referred to as GA Mix 1, GA Mix 2
and GA Mix 3) was assessed in this study
using the standard tensile strength ratio
(TSR) parameter supported with a micro-
scopic imaging method. The mentioned
moisture susceptibility methods were used
to select the optimum glass-asphalt mix that
demonstrates least resistance to moisture
damage. The stiffness and permanent
deformation properties of the optimum
glass-asphalt mix was then compared to a
conventional medium dense-graded asphalt
Table 1 Asphalt mix proportions and design aggregate grading
Aggregate type and size
Glass-asphalt mix
Andesite 10 mm
Andesite 7.1 mm
Andesite crusher dust (<5 mm)
Granite crusher sand (<5 mm)
Mine sand (<0.6 mm)
Crushed glass (<5 mm)
Filler: hydrated lime
Design aggregate grading
wearing course mix that is typically used
for road construction in South Africa. The
Huet-Sayegh model was used to characterise
the stiffness properties of both mixes and
establish an empirical stiffness characterisa-
tion model for the glass-asphalt mix tested.
Additionally, an alternative mathematical
model, which makes use of a polynomial
modelling technique, was used to character-
ise the permanent deformation resistance of
both mixes, which was also validated with
the Francken model.
MATERIALS AND MIX DESIGN
Aggregate material and
design aggregate grading
The same aggregates (i.e. andesite, granite
and mine sand) that were used in the con-
ventional mix were utilised in the production
of the glass-asphalt mix. The granite crusher
sand was partially substituted with 15% of
recycled crushed glass due to the similar
particle size distribution of both materials.
The composition of recycled crushed glass
was selected based on previous studies
(mentioned above) indicating an optimum
crushed glass content of 10 to 15%. The com-
bined grading of the individual aggregates,
crushed glass material and mineral filler
was optimised to represent a similar design
aggregate grading to the conventional mix by
using a non-linear least-square optimisation
technique in Microsoft Excel 2013.
Table 1 indicates the percentage
composition of each aggregate type and
Asphalt mix proportions (%)
Conventional asphalt mix
28 28.5
17 12.5
22 26
11 26
6 6
15 –
1 1

of a glass-asphalt mix incorporating 10% Sieve size (mm) 14 10 7.1 5 2 1 0.6 0.3 0.15 .075
crushed glass (with an MPS of 5 mm) was Percentage by mass passing (%)
comparable to a traditional dense-graded
Glass-asphalt mix 100 99 78 60 40 27 21 15 9 5.4
asphalt wearing course mix. Arabani (2010)
observed an increase in stiffness modulus Conventional asphalt mix 100 97 75 59 42 30 21 14 9 5.8
with an increase in crushed glass content Grading specification minimum 100 80 32 4
up to 15%, above which reduced stiff-
Grading specification maximum 100 67 10
ness modulus was reported. Moreover,

Journal of the South African Institution of Civil Engineering  Volume 62  Number 3  September 2020 11
Table 2 Particle size distribution of recycled crushed glass
Sieve size (mm) 5 2
Recycled crushed glass (%) 100 82
the mineral filler required to achieve the
conventional mix design aggregate grad-
ing. It can also be observed that the design
grading meets the grading control points
specified for a 10 mm NMPS dense-graded
asphalt mix as per standard design practices
for traditional asphalt mixes in South Africa
(Sabita 2016). NMPS is defined in South
Africa as “one sieve size larger than the larg-
est sieve to retain a minimum of 15% of the
aggregate particles” (Sabita 2016).
Crushed glass material
The recycled crushed glass was procured
from a glass manufacturing plant located in
the Gauteng Province of South Africa. The
type of glass that was used in this study is
soda-lime glass. The general composition
of soda-lime glass is as follows: 70–74% sili-
con dioxide (SiO2), 10–14% calcium oxide
(CaO) and 13-16% sodium oxide (Na2O).
The chemical composition of the glass
material used in this study was obtained
by means of X-ray fluorescence (XRF).
The analysis on a sample of the material
revealed a chemical composition of 72%
SiO2, 10% CaO and 13% Na2O. Other
compounds were also present in smaller
amounts and included 2.5% aluminium
oxide (Al2O3), 0.1% chromium oxide
(Cr2O3) and 0.7% iron oxide (Fe2O3).
It is known that aluminium and chro-
mium compounds are present in higher
levels of amber and green-coloured glass
respectively. Additionally, iron compounds,
carbon and sulphur are also used as dif-
ferent colouring agents. The effect of these
(a)
200 μm Glass Consol EHT = 15.00 kV Mag = 50 X
Figure 1 S EM image showing (a) angularity and (b) surface texture of recycled crushed glass particles – Scale: (a) 200 µm, Mag 50× (b) 20 µm, Mag 200×
12
1 0.6 0.3 0.15 0.075
46 27 15 8 4.3
elements on long-term asphalt performance
characteristics, as well as the long-term
environmental impact, requires future
investigation.
The particle size distribution of the
crushed glass is reported in Table 2 and
represents a fine continuous grading with
an MPS of 5 mm.
The microscopic morphology of the
crushed glass material was examined by
scanning electron microscopy (SEM).
The SEM examinations indicate that the
crushed glass consists mainly of fine angu-
lar particles with fine-to-coarse textured
features present on the surfaces of the glass
particles, as shown in Figure 1. Also, due
to the fine grading of the crushed glass
material, reduced quantities of elongated
particles and very few particles with sharp
edges can be observed.
The observed angularity is also con-
sistent with the fine aggregate angularity
(FAA) test results obtained from the FAA
test, which was performed on the crushed
glass material and the traditional fine
aggregates in accordance with ASTM C
1252 (2003). The FAA test measures the
loose uncompacted void content of the fine
aggregates and assumes a higher angularity
with a higher uncompacted void content.
The results are compared in Table 3. It can
be observed that the crushed glass demon-
strates increased angularity (50.9%) in com-
parison with the fine andesite (44.6%) and
fine granite (45.0%) aggregates. This can
be desirable in glass-asphalt mixes because
angular particles interlock better with each
(b)
Signal A = BSD1 20 μm Glass Consol
Volume 62  Number 3  September 2020  Journal of the South African Institution of Civil Engineering
Table 3 U
 ncompacted void content of fine
aggregates and crushed glass
Uncompacted
Aggregate
void content (%)
Andesite crusher dust 44.6
Granite crusher sand 45.0
Recycled crushed glass 50.9
other, thereby providing adequate inter-
particle friction necessary to resist perma-
nent deformation of the mix. Furthermore,
the crushed glass material and fine
aggregates meet the minimum criteria of
45% for a design traffic level of 30 million
equivalent standard axle loads (ESALs), as
specified in AASHTO M323 (2013).
An X-ray diffraction (XRD) analysis was
also conducted on the entire grading of the
recycled crushed glass material to identify
the percentage composition of amorphous
(non-crystalline) silica, i.e. glass, present in
the sample. This provides an indication of
the relative degree of “cleanliness or purity”
of the recycled crushed glass obtained from
the glass provider in South Africa.
Table 4 indicates the percentage com-
position of amorphous silica versus crystal-
line silica present in the sample of recycled
crushed glass. The form of crystalline
silica identified in the recycled crushed
glass sample is quartz, which is the most
common form of crystalline silica and is a
mineral commonly found in most rocks,
sands and soils.
It can be observed that on average more
than 90% of the recycled crushed glass
sample comprises amorphous silica, while
less than 10% comprises crystalline silica.
The locally available source of recycled
crushed glass therefore demonstrates a
EHT = 15.00 kV Mag = 200 X Signal A = BSD1
Table 4 X-ray diffraction analysis of crushed glass
Crushed glass particle size
Passing 0.075 mm
0.075 mm
0.15 mm
0.3 mm
0.6 mm
1 mm
2 mm
5 mm
high degree of purity and may hence be
considered favourably for application in
glass-asphalt mixes.
Antistripping additives
and mineral filler
Hydrated lime (1% by mass of dry aggre-
gate) constituted the filler component and
served as an antistripping additive in GA
Mix 1. The incorporation of 1% hydrated
lime was selected based on the maximum
specified amount allowed for inclusion
in dense-graded asphalt mixes in South
Africa (Sabita 2016). The hydrated lime was
introduced into the mix by firstly mixing it
with the cold aggregates and crushed glass,
and then reheating to 165°C before mixing
with bitumen.
A liquid antistripping additive (0.5%
by volume of binder) was applied to GA
Mix 2. The liquid additive, namely WETFIX
BE, is an amine surfactant and is specially
designed to improve the adhesion between
the bitumen and aggregate surfaces in
hot-mix asphalt. Additionally, its heat
stability allows it to be stored in hot bitu-
men for up to five days at temperatures
up to 170°C without losing its properties.
This additive is often used as an adhesion
promoter in hot-mix asphalt production in
South Africa. The liquid additive was added
to the binder according to the supplier’s
Amorphous silica (%) Crystalline silica (%)
83.67 14.36
86.75 12.57
89.05 9.47
91.66 6.69
95.82 2.75
97.56 1.37
97.19 1.27
97.04 1.05
recommendations, and involved heating
the original binder to approximately 150°C
(mixing temperature), maintaining the same
temperature while blending in the liquid
additive, using a high-shear blender, for one
hour. The blended binder was thereafter
maintained at the same temperature and
immediately incorporated into the mix.
To investigate the effect of the liquid
antistripping additive on the moisture sus-
ceptibility of GA Mix 2, as well as similar
effects on the moisture sensitivity of GA
Mix 3 without an antistripping additive,
the mineral filler in GA Mixes 2 and 3
consisted of 1% (by mass of dry aggregate)
of baghouse fines instead of hydrated lime.
Although the type of mineral filler varied,
the design grading of all three mixes was
not affected, due to similar particle size
distribution of both filler components.
Mix design and optimum
binder content
The South African mix design guideline for
dense-graded asphalt mixes (Sabita 2016) was
followed to determine the optimum binder
content of GA Mixes 1, 2 and 3, as well as
the conventional asphalt mix. In total, 80
samples were prepared for the mix design.
The same binder, i.e. 50–70 penetration
grade binder with penetration (25°C, 100 g,
0.1 mm) of 65 and softening point after
rolling thin film oven (RTFO) aging at 47°C,
that was utilised in the conventional mix
was used to prepare the glass-asphalt mixes.
The constituent aggregates (as per the design
aggregate grading), respective mineral fillers
and the binder were mixed at a temperature
of 150°C and the prepared mixtures were
then short-term-aged at 135°C for four hours.
The mixtures were thereafter compacted at
135°C using a Superpave Servopac gyratory
compactor in accordance with AASHTO
T312 (2015a). Compaction was conducted
at four trial binder contents (i.e. 4.0, 4.5, 5.0
and 5.5%) to dimensions of approximately
150 mm diameter by 115 mm height at 100
gyrations. The South African mix design
method specifies a laboratory compaction
requirement of 100 gyrations for a design
traffic level of 3 to 30 million ESALs.
The bulk relative density (BRD) of three
replicate compacted samples at each trial
binder content was determined in accor-
dance with SANS 3001-AS10 (SANS 2011a).
Additionally, the maximum theoretical
relative density (MTRD) for each asphalt
mix was determined on two loose samples
at each trial binder content in accordance
with SANS 3001-AS11 (SANS 2011b). The
average laboratory measured MTRD values
and BRD values were used to determine
the voids of the compacted samples, as
presented in Table 5.
The South African mix design method
specifies the optimum bitumen binder
content to be established at 4% air voids in
the mix. As per this criterion, an optimum
binder content of 5.4% for GA Mixes 1, 2
and 3, and 5.2% for the conventional mix
was determined. Although the particle size
distribution of the crushed glass and the
granite crusher sand is similar, the crushed
glass is on average 15% finer between 1 mm
and 2 mm particle size. It is suspected that
the incorporation of finer crushed glass may
have contributed to higher optimum binder
content in comparison with the conventional
mix, owing to a larger surface area.

Table 5 BRD, MTRD and voids after 100 gyrations

Bitumen Voids Voids Voids Voids
BRD MTRD BRD MTRD BRD MTRD BRD MTRD
binder (%) (%) (%) (%)
content
(%) GA Mix 1 GA Mix 2 GA Mix 3 Conventional mix

4.0 2.378 2.616 9.1 2.385 2.620 9.0 2.380 2.617 9.1 2.446 2.645 7.5

4.5 2.410 2.592 7.0 2.414 2.595 7.0 2.410 2.598 7.2 2.479 2.618 5.3

5.0 2.438 2.576 5.4 2.443 2.581 5.4 2.439 2.578 5.4 2.508 2.604 3.7

5.5 2.462 2.557 3.7 2.467 2.561 3.7 2.462 2.559 3.8 2.528 2.583 2.1

Journal of the South African Institution of Civil Engineering  Volume 62  Number 3  September 2020 13
LABORATORY PERFORMANCE
TESTING PROGRAM
Mix sample preparation
The performance testing samples were man-
ufactured at a mixing temperature of 150°C
and a compaction temperature of 135°C.
Prior to compaction, the asphalt mixtures
were shot-term-aged (to simulate the ageing
process that occurs during the asphalt pro-
duction phase and the transportation phase)
for four hours at the compaction tempera-
ture. The moisture susceptibility test samples
were compacted to dimensions of 150 mm
diameter by 60 mm height, and the stiffness
and permanent deformation test samples
were compacted to dimensions of 100 mm
diameter by 150 mm height. The test samples
were compacted at the determined optimum
binder content to a target air void content
of approximately 7% (representative of field
voids). Compaction of the samples to the tar-
get height and air voids was achieved after 90
to 120 gyrations using a Superpave Servopac
gyratory compactor in accordance with
AASHTO T312 (2015a). In total, 30 samples
were prepared to characterise and compare
the performance of the glass-asphalt mix
with the conventional mix.
Modified Lottman testing
The Modified Lottman test, conducted in
accordance with ASTM D4867M (2009),
was performed to assess the moisture sus-
ceptibility of GA Mixes 1, 2 and 3. Six gyra-
tory compacted test samples for each glass-
asphalt mix were prepared and divided into
two subsets. One subset was maintained dry
while the other underwent a partial satura-
tion (55% – 80% saturation) and freeze-thaw
conditioning process. Although the freeze
cycle is optional in ASTM D4867M (2009),
this conditioning process was included since
the asphalt mix contains glass particles,
which makes it more prone to moisture
damage than traditional aggregates (Federal
Highway Administration 1998). In addition,
the glass particles require additional treat-
ment with anti-stripping additives. The aim
was to therefore simulate conditioning that
is more severe than conditioning expected
in the field.
The indirect tensile strength (ITS) of
each subset was determined by the indirect
tensile splitting test. During testing, a dia-
metrical load at 50 mm/min at 25°C was
applied until the maximum load required
to fracture the samples was reached. The
maximum load was used to determine the
ITS for each subset.
14
Additionally, ASTM D4867M (2009)
specifies a visual assessment to be con-
ducted on the fractured (tested) samples.
This assessment entails breaking the tested
samples open and reporting on the visu-
ally estimated degree of moisture damage.
To eliminate visual judgement and biased
interpretation associated with the current
standard of visual inspection and reporting,
a microscopic imaging analysis was imple-
mented in this study to assess the degree
of stripping that had occurred in each mix.
One fractured sample from the wet subset
of each mix was examined under a stereo
microscope at a 6× zoom magnification.
To obtain a representative area, several
sections of dimension 22.05 mm (width) by
14.68 mm (height), each spanning over the
cross-sectional area of the fractured sample,
were examined under the microscope.
The microscopic sections were captured
by a built-in integrated digital camera and
the resulting images were imported into
ImageJ 1.52a, which is the public domain
Java Image processing program, inspired by
the National Institute of Health (NIH). In
ImageJ, the captured images were convert-
ed to 8-bit greyscale images which consist
of 256 levels of grey intensity per pixel that
range from 0 (black) to 255 (white). The
variation in grey intensity levels is depen-
dent on the densities of each component
material in the mix. Dense materials are
represented by the brighter regions, while
low density materials are represented by
the darker regions. As such, in this analysis
the exposed aggregates correspond to the
brighter regions and the bituminous-coated
aggregates, mastic and voids correspond
to the darker regions. Since the pixel grey
values of the exposed aggregates and the
remaining material phases (i.e. bituminous-
coated aggregates, mastic and voids) were
distinctly different from each other, a
threshold value of 65 was easily selected
to distinguish between the two regions.
This threshold value was considered to be
most accurate in identifying the areas of
stripping and was consistently applied in
all three mixes for realistic comparison.
The applied threshold level converts the
image to a binary image, where all pixels
in the greyscale image greater than the
threshold value are replaced with the value
255 (white) and the remaining pixels with
the value 0 (black). The binary images were
used to quantify the degree of stripping by
measuring the area of white pixels (area of
exposed aggregates) as a ratio to the area
of the sum of white and black pixels (total
Volume 62  Number 3  September 2020  Journal of the South African Institution of Civil Engineering
sample surface area) for each mix. The area
measurement of white and black pixels was
automatically computed in the software.
Light reflection off the coated crushed
glass fines was noted in particular during
microscopic examinations of the glass-
asphalt samples. The reflection was indi-
cated by an even and consistent distribution
of bright speckles in the captured images of
all three mixes and were thus identified as
white pixels in the binary image. It should
be noted that similar reflection was also
noted in the unconditioned samples, when
examined under the microscope. To avoid
processing and analysis of these reflections,
particles up to and including 0.15 mm for all
three mixes were removed from the binary
image prior to stripping area measurement.
This was performed in ImageJ using the
‘particle analyser’ function, where only par-
ticles bigger than 0.15 mm were processed
and analysed.
The steps implemented during the imag-
ing analysis are summarised in Figure 2.
One section captured per sample of each
glass-asphalt mix (GA Mixes 1, 2 and 3) is
illustrated, while the same steps were fol-
lowed to obtain the percentage of stripping
across several sections of the sample.
Dynamic modulus testing
The dynamic modulus |E*| is a perfor-
mance-related parameter that is used to
characterise the resilient response (stiff-
ness) of hot-mix asphalt. The dynamic
modulus is defined by the absolute value of
the complex modulus, E*.
The dynamic modulus test was conduct-
ed on the optimum glass-asphalt mix and
the conventional mix in accordance with the
CSIR (Council for Scientific and Industrial
research) protocols developed for the asphalt
industry in South Africa (Anochie-Boateng
et al 2010). During the test, a haversine
compressive load pulse was applied on three
replicate gyratory compacted samples at five
test temperatures (-5, 5, 20, 40, 55°C) and six
loading frequencies (25, 10, 5, 1, 0.5, 0.1 Hz)
with no confining pressure. The vertical
deformation of the samples was determined
by recording the average measurements
of three axial linear variable displacement
transducers (LVDTs). The dynamic modulus
of the samples tested was computed by
recording the axial stresses and the resulting
axial resilient strains for the last five load
cycles for each test. A strain-controlled type
of dynamic modulus testing was followed
such that the measured strain was limited to
approximately 100 micro-strain.
 Glass-asphalt mix

GA Mix 1 GA Mix 2 GA Mix 3

Original image

8-bit greyscale image conversion

Binary image conversion (with reflecting glass particles)

Elimination of reflecting particles ≤ 0.15 mm

Stripping and sample surface area determination

Precentage stripping computation

1.1 % 1.6 % 11.5 %

Figure 2 M
 icroscopic imaging analysis methodology used in this study

Journal of the South African Institution of Civil Engineering  Volume 62  Number 3  September 2020 15
 contributes towards the enhanced dry
1 200 0.89 0.90 strength in comparison with Mixes 2 and
1 106
3. It is also apparent from the percentage
1 000 982 reduction in strength of the conditioned
950
0.85
subset of Mix 1 (i.e. 11%) in comparison with
Indirect tensile strength (kPa)

867

Tensile strength ratio (TSR)

Mix 2 (i.e. 20%) that the role of the hydrated
800
0.80 722 lime as an antistripping additive in the
692 0.80
presence of water is considerably more effec-
600
0.76 tive than the liquid antistripping additive.
0.75 Although a larger reduction in dry strength
400 is noted for Mix 2 in comparison with Mix 3,
the percentage reduction in strength of the
0.70 conditioned subset of Mix 2 (i.e. 20%) is not
200
as pronounced as in the case of Mix 3 (i.e.
24%), which is expected due to the absence
0 0.65
GA Mix 1 GA Mix 2 GA Mix 3 of an antistripping additive in Mix 3.
Glass-asphalt mix It is interesting to note that Mix 2, which
Wet strength Dry strength TSR contains a liquid antistripping additive, indi-
cates the lowest dry strength in comparison
Figure 3 M
 oisture susceptibility evaluation of glass-asphalt Mixes 1, 2 and 3 with Mixes 1 and 3. The liquid additive may
thus have an adverse effect on the strength
Flow number testing the mastic (i.e. binder and mineral filler) and of asphalt mixes, although the moisture
In this study, the flow number test was the aggregates in the mix, making the mix susceptibility may improve. This observation
conducted to evaluate the permanent more susceptible to moisture damage during is in line with the general notion that liquid
deformation resistance of the glass-asphalt cyclic loading. The moisture susceptibility additives like these cause early pavement
mix and the conventional mix. The asphalt of the mix was evaluated by the tensile failures such as rutting, even although they
mixture performance tester (AMPT) strength ratio (TSR), which was determined are known to improve moisture susceptibility
permanent deformation test procedure, by the ratio of the average ITS of the wet (Iowa Department of Transportation 2012).
specified in AASHTO TP 79 (2015b), was subset to the average ITS of the dry subset. In South Africa, a minimum tensile
used to conduct the flow number test on The results are presented in Figure 3. strength ratio of 0.8 for asphalt wearing
three replicate gyratory compacted samples. It can be observed from the results that courses is specified (Sabita 2016). It can be
The flow number test entails the application the incorporation of hydrated lime in Mix 1 seen in Figure 3 that GA Mixes 1 and 2 meet
of a repeated compressive haversine load at
1 Hz (i.e. one cycle with a loading time of Table 6 Microscopic imaging analysis of glass-asphalt mix
0.1 seconds and a rest period of 0.9 seconds) Analysed Area of white Total pixels Area of Total area of
and a measure of the corresponding cumu- sample sections pixels (mm2) (mm2) white pixels (%) white pixels (%)
lative axial permanent strain as a function GA Mix 1
of load cycles. In this study, a deviator stress
1 3.6 323.3 1.1
level of 276 kPa with a confining pressure of
2 1.9 323.3 0.6
69 kPa was applied on the test samples and
3 5.1 323.3 1.6
conducted at a test temperature of 50°C. 6.2
The cumulative permanent axial strain 4 0.6 323.3 0.2
response is typically divided into three main 5 4.9 323.3 1.5
phases, i.e. primary, secondary and tertiary. 6 3.9 323.3 1.2
The cycle number at which tertiary flow GA Mix 2
(i.e. tertiary stage) of the mix commences 1 2.1 323.3 0.6
is defined as the flow number. Thus, it is
2 5.2 323.3 1.6
considered that the permanent deformation
3 13.3 323.3 4.1
failure of the asphalt mix occurs at the onset 9.9
4 1.4 323.3 0.4
of tertiary flow and is indicative of the resis-
tance of the mix to permanent deformation. 5 4.9 323.3 1.5
6 5.4 323.3 1.7
GA Mix 3
RESULTS AND DISCUSSIONS 1 10.4 323.3 3.2
2 18.3 323.3 5.7
Moisture susceptibility evaluation 3 30.9 323.3 9.6
Moisture damage may occur due to the 46.4
4 37.2 323.3 11.5
loss of strength and durability in an asphalt
5 33.2 323.3 10.3
mix in the presence of moisture. Moisture
6 19.6 323.3 6.1
permeates and weakens the bond between

16 Volume 62  Number 3  September 2020  Journal of the South African Institution of Civil Engineering
Table 7 Moisture susceptibility ranking for GA Mixes 1, 2 and 3
Glass-asphalt mix Performance ranking
GA Mix 1
GA Mix 2
GA Mix 3
* Minimum TSR criteria (South Africa) = 0.8
the minimum tensile strength ratio criteria,
whereas GA Mix 3 fails to comply with the
specified requirement. Based on the TSR as a
reliable indicator of moisture susceptibility, it
is anticipated that GA Mix 1 will be less sus-
ceptible to moisture damage than GA Mix 2,
while GA Mix 3 is expected to demonstrate
the least resistance to moisture damage.
From the microscopic image analysis
conducted on several sections of the frac-
tured conditioned sample, the ratio of the
area of white pixels (exposed aggregates) to
the ratio of the total area of white and black
pixels (exposed aggregates and bituminous-
coated aggregates) was determined from
the processed binary images to quantify
the degree of stripping for GA Mixes 1, 2
and 3, according to the method described
earlier. The results of this analysis are
reported in Table 6.
The results show a clear variation in
the degree of stripping with the addition
of both antistripping additives. A signifi-
cant reduction in the area of white pixels
(stripped areas) can be observed for GA
Mixes 1 and 2 compared to GA Mix 3.
The total area of white pixels reduces from
approximately 47% (GA Mix 3) to less than
10% (GA Mixes 1 and 2). These results
confirm the effectiveness of antistripping
additives in significantly improving the
moisture susceptibility of glass-asphalt
mixes. Moreover, the addition of hydrated
lime appears to reduce the stripped areas
slightly more than the liquid antistripping
additive, although a major distinction was
not apparent from the microscopic images.
It is evident from the above evalua-
tion that both methods appear to rank
the moisture susceptibility of GA Mixes
1, 2 and 3 in the same order, as reported
in Table 7. The ranking of 1 to 3 indicates
least to most moisture susceptible. Based
on these rankings, GA Mix 1 was selected
as the optimum glass-asphalt mix on which
further investigation regarding the stiffness
and permanent deformation properties were
conducted. It is also recommended that
microscopic imaging techniques be used as
a tool to validate the TSR results of mixes
that are susceptible to moisture damage.
Journal of the South African Institution of Civil Engineering  Volume 62  Number 3  September 2020
Microscopic image
TSR*
analysis (%)
1 0.89 (1) 6.2 (1)
2 0.80 (2) 9.9 (2)
3 0.76 (3) 46.4 (3)
E0
E∞ – E0 k h
Figure 4 R
 epresentation of Huet-Sayegh
model (Nilsson et al 2002)
Stiffness evaluation
Dynamic modulus modelling
The dynamic modulus behaviour of the GA
Mix 1 and conventional mix was evaluated
using the Huet-Sayegh model. The Huet-
Sayegh model uses physical elements to
describe the dynamic modulus behaviour
of HMA materials and is represented by
a combination of two units connected in
parallel, i.e. a spring, E∞ – E 0, and two
bi-parabolic dashpots, h and k, connected
in series with a spring, E0, connected in
parallel, as shown in Figure 4.
The associated mathematical expression
of dynamic modulus for the Huet-Sayegh
model is determined by Equation 1
(Nilsson et al 2002).
E∞ – E 0
|E*(iωτ)| = E0 +
1 + δ(iωτ)–k + (iωτ)–h
Where:
|E*| = dynamic modulus
i2 = complex number, defined by i2 = –1
ω = angular frequency (Hz), ω = 2πf
E 0, E∞, δ, k, h, τ = model constants
In Equation 2, E∞ represents the purely elas-
tic component (E* for the high frequency
Table 8 Determination of Huet-Sayegh model parameters
Asphalt mix E0
Glass-asphalt 53
Conventional 55
and low temperature domain) of the com-
plex modulus, while E0 represents the long-
term elastic modulus (residual E* for the low
frequency and high temperature domain).
δ, k, and h represent the parameters of the
parabolic elements of the Huet-Sayegh
model and τ is the retardation time regard-
ing the effect of temperature on the complex
modulus, which is defined as per Equation 2.
2
τ = e a + bT + cT (2)
Where:
τ = retardation time (s)
T = temperature (°C)
a, b, c = regression constants
In this paper, the measured dynamic modu-
lus data of GA Mix 1 and the conventional
mix is represented by the Huet Sayegh model
in the Black Space diagram shown in Figure 5
(p 18). The Black Space diagram presents a
plot of the phase angles and the correspond-
ing dynamic moduli at the tested tempera-
ture and frequency. For bituminous materials
the viscous or elastic properties are indicted
by the phase angle. The phase angle for a
purely elastic material is 0°, while for a purely
viscous material the phase angle is 90°.
The purely elastic modulus is obtained
when the phase angle approaches zero,
resulting in the storage modulus approach-
ing the elastic modulus, and the loss modu-
lus approaching zero. The purely elastic
parameter E∞ was obtained from the Black
diagram by extrapolating the curve to zero
phase angle, as shown in Figure 5. The deter-
mination of the remaining five parameters,
E0, δ, k, h and τ was performed graphically
such that the measured data is most accu-
rately represented by the model in the Black
diagram. This was obtained by a non-linear
least-squares regression of the set of mea-
 (1)
sured dynamic moduli and corresponding
phase angles at the respective test frequen-
cies and temperatures. The regression was
carried out using the Solver function in
Microsoft Excel 2013. It can be observed that
the use of the Huet-Sayegh model provides
an excellent fit for the measured dynamic
modulus data. The obtained Huet-Sayegh
model parameters, at a reference tempera-
ture of 20°C, are provided in Table 8.
E∞ δ k h τ
31000 2.3699 0.1932 0.6188 0.01863
32000 2.5014 0.2047 0.6533 0.02106
17
 transferred to the aggregate skeleton, while
45 at low temperatures it is the binder that
40 determines the load-carrying capacity of
the mix. In this regard, a stable aggregate
35
skeleton with optimal aggregate interlock
Phase angle (degree)

30 contributes to increased effective stiffness

25 at elevated temperatures.
As previously mentioned, the glass
20
particles utilised in this study demonstrate
15 higher angularity than the traditional
10 aggregates. The higher angularity in turn
increases the interlock between the
5
crushed glass particles and constituent
0
1.00E+00 1.00E+02 1.00E+04 1.00E+06 1.00E+08
aggregates in the mix, and could be a con-
Log E* (MPa) tributory factor towards the increased stiff-
(a) –5°C 5°C 20°C 40°C 55°C Huyet-Sayegh model ness observed with the glass-asphalt mix at
40°C and 55°C. It can also be observed that
the phase angles for both mixes at 20°C are
50
comparable, reflecting comparable stiffness
45 at intermediate temperatures.
40
Permanent deformation evaluation
Phase angle (degree)

35
30
Permanent deformation modelling
25
The flow number is an indicator of asphalt
20 mix rutting resistance and was used in this
15 study to compare the permanent deforma-
10 tion resistance of GA Mix 1 and the con-
5 ventional mix. The flow number is defined
by the number of cycles at which the rate
0
1.00E+00 1.00E+02 1.00E+04 1.00E+06 1.00E+08 of change of permanent strain is minimum
Log E* (MPa) and corresponds with the onset of tertiary
(b) –5°C 5°C 20°C 40°C 55°C Huyet-Sayegh model flow. Inaccuracies in flow number deter-
mination may arise when the minimum
Figure 5 H
 uet-Sayegh model representation of Black Space diagram for (a) glass-asphalt mix and permanent strain rate is merely obtained
(b) conventional mix from the laboratory measured results, due to
existing variation/noise in the experimental
The obtained Huet-Sayegh model in the mix. At elevated temperatures, the data. Additionally, there is an infinite
parameters were used to establish an empir- binder in the mix becomes more viscous number of cycles in the secondary stage
ical model for the prediction of dynamic and the load carrying capacity is gradually where the constant permanent strain slope
modulus of the glass-asphalt mix tested,
as described in Equation 3. Validation of
the model parameters will, however, be 50
conducted when additonal laboratory and 45
actual field testing data becomes available.
40

35
Phase angle (degree)

31 000 – 53
|E*(iω)| = 53 + (3)
1 + 2.37(iω0.02)–0.19 30
+ (iω0.02)–0.62 25

20
Dynamic modulus results
A comparative plot of the Huet-Sayegh 15

model for both mixes at 20°C, 40°C and 10

55°C in the Black Space diagram is present- 5
ed in Figure 6. Lower phase angles for GA
0
Mix 1 in comparison with the conventional 1.00E+00 1.00E+02 1.00E+04
mix can be observed at 40°C and 55°C. Log E* (MPa)
The lower phase angles indicate a tendency H-S model: GA Mix 1 H-S model: Conventional Mix
towards more elastic behaviour due to
the aggregate skeleton taking precedence Figure 6 Black Space diagram for glass-asphalt mix and conventional mix

18 Volume 62  Number 3  September 2020  Journal of the South African Institution of Civil Engineering
is minimum. As a result, various algorithms
have been proposed for identifying the flow 5 0.03
Measured permanent strain (GA Mix 1)
number in the flow number test. Most of
Measured strain slope (GA Mix 1)

Cumulative permanent strain (%)

the proposed algorithms, however, are only 4 Polynomial model (GA Mix 1)

Permanent strain slope (%)

capable of modelling the primary and sec-
ondary phases of permanent strain and pose 0.02
a challenge in representing a good-fitting 3
three-phase permanent deformation model
where the tertiary phase is also encoun- 2 FN = 331
tered. AASHTO TP 79 (2015b) refers to the 0.01
Francken model as most accurate for flow
number computation. The Francken model 1
εp = a + bN + cN2
combines a power model, which describes
the primary and secondary phases of 0 0
permanent deformation and an exponential 0 200 400 600 800 1 000 1 200
model, which describes the tertiary phase. (a) Load cycles (N)
In this study, an alternative math-
ematical model, which makes use of a 5 0.03
polynomial modelling technique for flow Measured permanent strain (convential mix)
number determination, is used to compare Measured strain slope (convential mix)
Cumulative permanent strain (%)

4 Polynomial model (convential mix)

the permanent deformation resistance of

Permanent strain slope (%)

the glass-asphalt mix and the conventional 0.02
mix, and the results are validated with 3
the Francken model computation of flow
number. The analysis methodology of each
2 FN = 257
model used for identifying the flow number 0.01
is described next.
1 εp = a + bN + cN2
Polynomial model
During the secondary stage of permanent
0 0
deformation, the rate of change of strain 0 100 200 300 400 500 600 700 800 900
(strain slope) is considered to reach a (b) Load cycles (N)
minimum after a certain number of
loading cycles. The polynomial model is Figure 7 P
 olynomial model describing permanent deformation of (a) glass-asphalt mix and
implemented in this region of the perma- (b) conventional mix
nent deformation curve, whereby a second-
degree polynomial curve, in the form of from a non-linear least-squares regression of εp(N) = ANB + C(e DN – 1) (6)
Equation 4, is used to model the measured the set of measured permanent deformation
rate of change of permanent strain. data. The regression was carried out using Where:
the Solver function in Microsoft Excel 2013. εp(N) = permanent strain (%)
εp = a + bN + cN 2(4) The determined coefficients of the model are N = number of load cycles
provided in Table 9. A,B, C and D = regression coefficients
Where: The obtained regression coefficients
εp = permanent strain rate of the polynomial model were used to Figure 8 shows the Francken model
N = load cycle determine the flow number values using which was used to model the permanent
a, b and c = regression coefficients the approach mentioned above. The flow deformation behaviour of GA Mix 1 and
number (FN) obtained for GA Mix 1 and the conventional mix. The coefficients of
The flow number is determined by setting the conventional mix was determined at the Francken model were also obtained
the first derivative of Equation 4 to zero 331 and 257 load cycles, respectively. from a non-linear least-squares regression
and solving for the number of load cycles of the set of measured permanent
(N), as described in Equation 5. Francken model deformation data. The determined
The Francken model is described by coefficients of the model are provided
dεp Equation 6. in Table 10 and were used to determine
= b + 2cN = 0 (5)
dN
Table 9 Determination of polynomial regression coefficients
Figure 7 presents the polynomial model used
Asphalt mix a b c
to describe the average measured permanent
strain rate of GA Mix 1 and the conventional GA Mix 1 3.3020E–03 –6.1095E–06 9.2356E–09
mix as a function of load cycles. The coeffi-
Conventional 4.7541E–03 –9.5882E–06 1.8533E–08
cients of the polynomial model were obtained

Journal of the South African Institution of Civil Engineering  Volume 62  Number 3  September 2020 19
 changes from negative to positive, and
5.0 0.06 is considered to be the onset of tertiary
Measured permanent strain (GA Mix 1)
4.5 (plastic) flow in asphalt mixes. The flow
Francken model (permanent strain)
Cumulative permanent strain (%)

0.05 number for GA Mix 1 and the conventional

4.0 Francken model (permanent strain model)

Permanent strain slope (%)

mix was obtained at 353 and 271 load
3.5
0.04 cycles respectively.
3.0 A comparison of the flow number
2.5 0.03 results obtained using the polynomial
2.0 FN = 353 model and the Francken model revealed
0.02 the statistical information provided in
1.5
Table 11. Very similar flow number results
1.0 are obtained from both models, with a
0.01
0.5 coefficient of variance (CoV) less than 5%
0 0 for the glass-asphalt mix as well as for the
0 200 400 600 800 1 000 1 200 conventional mix. It is therefore reasonable
(a) Load cycles (N) to consider the polynomial model as accu-
rate and applicable for the determination
5.0 0.06 of flow number for both glass-asphalt and
Measured permanent strain (convential mix) conventional asphalt mixes.
4.5
Francken model (permanent strain)
Cumulative permanent strain (%)

0.05
4.0 Francken model (permanent strain model)
Permanent deformation results

Permanent strain slope (%)

3.5 From the determined flow number values,
0.04
3.0 it is apparent that the glass-asphalt mix
2.5 0.03 has an increased resistance to permanent
FN = 271 deformation than the conventional mix,
2.0
0.02
as represented by a higher flow number.
1.5 This observation is also consistent with the
1.0 improved stiffness behaviour indicated by
0.01
0.5 the glass-asphalt mix at elevated tempera-
tures (40°C and 55°C) described earlier in
0 0
0 100 200 300 400 500 600 700 800 900 this paper.
(a) Load cycles (N) The higher flow number may be attrib-
uted to the highly angular glass particles
Figure 8 F rancken model describing permanent deformation of (a) glass-asphalt mix and which provide adequate interlock to gener-
(b) conventional mix ate the inter-particle friction necessary to
resist permanent deformation. This was
Table 10 Francken model regression coefficients also observed through an SEM analysis
conducted by Anochie-Boateng and George
Asphalt mix A B C D
for the same source and grading of recycled
GA Mix 1 2.5759E-01 3.1425E-01 6.2991E-02 3.5671E-03 crushed glass in an asphalt mix (Anochie-
Boateng & George 2018). The SEM analysis
Conventional mix 2.0258E-01 3.60333+252E-01 1.7257E-01 3.3505E-03
demonstrated high angularity and in turn
increased interlock between the constituent
Table 11 Statistical analysis of modelled flow number (FN) results particles in comparison with a traditional
FN: FN:
asphalt mix incorporating less angular
Asphalt mix Mean STDEV CoV aggregates (see Figure 9).
Polynomial Francken

GA Mix 1 331 353 342 15.6 4.5%

Conventional mix 257 271 264 9.9 3.7%

CONCLUSIONS
The effectiveness of selected antistripping
additives in resisting moisture damage
the flow number of GA Mix 1 and the dεp was assessed as a variable in the labora-
= (ABN (B – 1)) + (CDeDN )(7)
conventional mix by equating the second dN tory performance evaluation of a 10 mm
derivative of the model, described in medium dense-graded asphalt wearing
Equation 8, to zero and solving for the load d2 εp course mix consisting of 15% recycled
= (AB(B – 1)N (B – 2)) + (CD 2 eDN )(8)
cycle number (N). The model coefficients dN2 crushed glass as follows: (1) effect of 1%
were also used to plot the permanent hydrated lime, (2) effect of 0.5% liquid
deformation rate (strain slope) and are Graphically this cycle number is repre- antistripping additive, and (3) effect
described by the first derivative of the sented at the point where the curvature of without the addition of an antistripping
model (Equation 7). the fitted permanent deformation curve additive. The effect of these variables on

20 Volume 62  Number 3  September 2020  Journal of the South African Institution of Civil Engineering
 (a) Glass particle
Glass particle
Glass particle
Mineral
aggregate stic
Ma
e n/
um
B it
Mineral aggregate
K Mg Al S Ca Si Na O C Electron
Figure 9 SEM of (a) glass-asphalt mix and (b) conventional mix (Anochie-Boateng & George 2018)
the moisture susceptibility of the glass-
asphalt mix was determined from the TSR
obtained from the Modified Lottman test,
supported with a microscopic imaging
analysis conducted on tested Modified
Lottman samples. The stiffness and
permanent deformation properties of the
glass-asphalt mix that demonstrated opti-
mum resistance to moisture damage were
thereafter compared to a conventional
asphalt mix (without crushed glass). The
Huet-Sayegh model and a polynomial
model were used to characterise the stiff-
ness and permanent deformation behav-
iour respectively. The following conclu-
sions can be drawn from this research:
QQ The recycled crushed glass material
evaluated in this study is potentially
capable of substituting traditional fine
aggregates that are typically used in
asphalt mixes in South Africa. This is
based on the improved rutting perfor-
mance demonstrated by the glass-asphalt
mix in comparison with the conventional
asphalt mix at elevated temperatures.
QQ The polynomial model is suitable
for determining the flow number of
the conventional and glass-asphalt
mixes, which was validated with the
Francken model. Similarly, the Huet-
Sayegh model is capable of accurately
characterising the resilient response
over the tested range of frequencies
and temperatures. The determined
model parameters are hence adequate to
establish a dynamic modulus predictive
model for the glass-asphalt mix, which
would, however, require validation
with additional laboratory and field
performance data.
QQ It was found that an antistripping
additive is required to meet moisture
Journal of the South African Institution of Civil Engineering  Volume 62  Number 3  September 2020
(b)
Mineral aggregate
Mineral aggregate
Mg. K S Na Al Ca Si O C Electron
100 μm
susceptibility criteria and alleviate
stripping in medium dense-graded
glass-asphalt mixes. In particular, it
can be concluded that hydrated lime is
more effective than the liquid antist-
ripping additive in alleviating stripping.
In this regard, the liquid antistripping
additive, which is known to be com-
monly used in HMA production in
South Africa, may not be as effective in
these non-conventional asphalt mixes.
Additionally, the microscopic image
analysis conducted to evaluate mix
moisture susceptibility is capable of
providing an accurate representation
of the degree of stripping and can be
used as a tool to quantitatively rank
moisture-sensitive mixes. Furthermore,
visual judgement and biased interpreta-
tion associated with the current stand-
ard of visual inspection and reporting
are eliminated.
ACKNOWLEDGEMENTS
The authors wish to acknowledge the
Council for Scientific and Industrial
Research (CSIR) through its R&D office
for funding this research through
Parliamentary Grant (PG) Funding. The
authors also wish to acknowledge Consol
Glass (Pty) Ltd. for providing the recycled
crushed glass material. Dr Morris de Beer
from the CSIR is also acknowledged for his
valuable contribution to this paper.
REFERENCES
AASHTO (American Association of State Highway
and Transportation Officials) 2013. Standard
specification for Superpave volumetric mix. Report
M323 2013. Washington, DC: ASSHTO.
Bitumen/Mastic
Mineral aggregate
AASHTO 2015a. Standard method for preparing and
determining the density of hot-mix asphalt (HMA)
specimens by means of the Superpave Gyratory
Compactor. Report R312. Washington, DC:
AASHTO.
AASHTO 2015b. Determining the dynamic modulus
and flow number for asphalt mixtures using the
Asphalt Mixture Performance Tester (AMPT).
Report TP 79. Washington, DC: AASHTO.
Andela, C & Sorge, E V (n.d.) Handbook of Alternative
Uses for Recycled Glass. Richfield Springs, NY:
Andela Products Ltd & and Glass Brokers, Inc.
Anochie-Boateng, J K, Denneman, E, O’Connell, J J,
Mturi, G & Ventura, D 2010. Hot-mix asphalt
testing for the South African pavement design
method. Proceedings, 29th Southern African
Transportation Conference (SATC), Pretoria.
Anochie-Boateng, J K & George, T B 2018.
Investigation of the use of waste crushed glass in
the production of asphalt mixes. Proceedings of
the Institution of Civil Engineers – Construction
Materials, 171(5): 187–194.
Arabani, M 2010. Effect of glass cullet on the
improvement of the dynamic behaviour of asphalt
concrete. Construction and Building Materials,
25: 1181–85.
Arnold, G, Werkemeister, S & Alabaster, D 2008. The
effect of adding recycled glass on the performance
of basecourse aggregate. Research Report 351.
Wellington, NZ: New Zealand Transport Agency.
ASTM (American Society for Testing and Materials)
2003. Standard Test Method for Uncompacted Void
Content of Fine Aggregate. Report C1252. West
Conshohocken, PA: ASTM International.
ASTM (American Society for Testing and Materials)
2009. Standard Test Method for Effect of Moisture
on Asphalt Concrete Paving Mixtures. Report
D4867M. West Conshohocken, PA: ASTM
International.
Australian Government 2011. National Waste Policy
Case Study. Canberra: Department of Sustainability,
Environment, Water, Population and Communities.
21
Dane County Department of Public Works (Wisconsin)
2003. Reuse/Recycling of glass cullet for non-container
uses. Madison, WI.
Department of Environmental Affairs (South Africa)
2017. South Africa State of Waste Report. First Draft
Report. Pretoria: DEA.
Federal Highway Administration 1998. User guidelines
for waste and by-product materials in pavement
construction. FHWA-RD-97-148, Report No 480017.
Washington, DC: FHA.
Hughes, C S 1990. Feasibility of using recycled glass
in asphalt. Richmond, VA: Virginia Department of
Transportation.
Iowa Department of Transportation 2012. Ranking
of HMA moisture sensitivity tests in Iowa. Final
Report. Ames, IA.
Lachance-Tremblay, E, Vaillancourt, M & Perraton, D
2014. Evaluation on the performance of an asphalt
mix with crushed glass. Proceedings, 59th Annual
22
Conference of the Canadian Technical Asphalt
Association, Winnipeg, Canada.
National Research Council (US) 1994. Permanent
deformation response of asphalt aggregate mixes.
Report SHRP-A-415. Washington, DC: NRC.
Nilsson, R N, Hopman, C P & Isacsson, U 2002.
Influence of different rheological models
on predicted pavement responses in flexible
pavements. Road Materials and Pavement Design,
3(2): 117–149.
Sabita (Southern African Bitumen Association) 2016.
Design and use of asphalt in road pavements. Sabita
Manual 35/TRH8. Cape Town: Sabita.
SANS (South African National Standard) 2011a.
SANS 3001-AS10: Determination of Bulk Density and
Void Content of Compacted Asphalt. Pretoria: SABS
Standards Division.
SANS 2011b. SANS 3001-AS11: Determination of the
Maximum Void-Less Density of Asphalt Mixes and
Volume 62  Number 3  September 2020  Journal of the South African Institution of Civil Engineering
the Quantity Of Binder Absorbed by the Aggregate.
Pretoria: SABS Standards Division.
Su, N & Chen, JS 2002. Engineering properties
of asphalt concrete made with recycled
glass. Resources, Conservation and Recycling,
35: 259–274.
VTRC (Virginia Transportation Research Council)
1998. Effect of glass concentration on stripping of
glasphalt. Final Report. Charlottesville, VA.
Wu, S, Yang, W & Xue, Y 2003. Preparation and
properties of glass-asphalt concrete. Wuhan
University, China.
Yamanaka, M, Gotoh, K, Saruwatari, M &
Mochishita, T‑ 2001. Thermal and mechanical
properties of glass cullet mixed with asphalt as
low-exothermic pavement material. Nagasaki
University, Japan.
Quantifying the economic TECHNICAL PAPER
benefits of gauge changes Journal of the South African
Institution of Civil Engineering

on the South African ISSN 1021-2019

Vol 62 No 3, September 2020, Pages 23–35, Paper 0936

core railway network CHRISTIAN DEMMEREZ DE CHARMOY

completed his MEng (Transportation
Engineering) studies at the University of
Pretoria earlier this year and is currently
C P Demmerez de Charmoy, P J Gräbe working for Hatch in its rail and transit
business unit as a civil perway engineer. He
is also registered as a candidate engineer
and is looking forward to adding value to
In its white paper, the Department of Transport (2017) envisages a future South African core the railway engineering industry.
railway network which requires the conversion and construction of over 8 500 km of standard-
Contact details:
gauge railway track by 2050. The scale of the project would undeniably have a large cost Department of Civil Engineering
attached to it, which needs to be well understood before any of the construction takes place. University of Pretoria
Numerous studies regarding railway gauge have been conducted within South Africa, but no Pretoria
0002
single paper has addressed the issue of which specific railway corridors could economically
T: +27 82 265 5805
benefit from a standard-gauge intervention. E: cdecharmoy@gmail.com
The purpose of this study was to identify which corridors in the South African core network
could potentially benefit from a gauge change intervention. These identified corridors were PROF HANNES GRÄBE (Pr Eng, FSAICE), is a
then economically evaluated to determine which of the corridors would outperform the base civil engineer with experience in the fields
of track technology, geotechnology,
case, which was set as the Market Demand Strategy (MDS) plan. Finally, it was determined if any
advanced laboratory testing, field
of the corridors identified outperformed the base case, to simulate how the operations of the investigations, maintenance models and
corridor would be affected. numerical analysis of track structures. He is
The conclusions of the study indicated that all the corridors in the South African core Associate Professor: Transnet Freight Rail
Chair in Railway Engineering at the
network should follow the plans proposed in the MDS to achieve the maximum return for the
University of Pretoria, where he lectures under- and post-graduate
analysis period, except for the Natal corridor. It was identified that the Natal corridor would courses in railway engineering. He is also responsible for railway research,
benefit most from a standard-gauge single line which would run concurrently with the narrow- as well as for continuing professional development in the form of short
gauge system transporting containers and other general freight along the corridor. courses presented to industry. He holds a PhD degree from the University
of Southampton (UK) and is a Fellow of the South African Institution of
Civil Engineering (SAICE) and is registered with the Engineering Council of
South Africa (ECSA) as a professional engineer.
INTRODUCTION economically beneficial on the identi- Contact details:
South Africa’s rail network has been unable fied corridors, such as adding a third Department of Civil Engineering
to take advantage of rail’s inherent benefits rail, constructing a standard-gauge line University of Pretoria
Pretoria
because of the early South African govern- parallel to the narrow-gauge corridor or
0002
ment’s decision to construct the rail net- completely changing the line from nar- T: +27 420 4723 / 083 286 5305
work in narrow gauge. Rail’s success is built row- to standard-gauge. E: hannes.grabe@up.ac.za
upon its three genetic technologies, namely QQ For the intervention identified, to
bearing, guiding and coupling (Van der understand how this will influence the
Meulen 2010). According to the Department operations on the corridor through the
of Transport (DoT) (2017), South Africa can analysis of the intervention by means of
operate long heavy-haul trains; however, simulation.
South Africa is unable to take advantage
of speed and axle load in the same way
that standard-gauge systems are able to. BACKGROUND
Numerous studies regarding railway gauge Rail gauge has been a contentious issue
have been conducted within South Africa, within South Africa for a long time and
but no single paper has addressed the issue has led to the release of many government
of which specific railway corridors could statements and papers regarding the issue.
benefit from standard-gauge interventions. The NATMAP 2050 report was compiled by
The objectives of the study are sum- The DoT in 2009 and evaluated the issue of
marised as follows: gauge in a global context and locally within
QQ To establish which corridors could South Africa. Since the NATMAP 2050’s
potentially benefit from a standard- release, the DoT (2017) released the National
gauge intervention. Rail Policy (NRP) White Paper which seeks
QQ To identify what type of standard- to reposition rail and would require the con-
gauge intervention would be most version of many corridors to standard-gauge. Keywords: r ailway gauge, economic analysis, agent-based simulation

Demmerez de Charmoy CP, Gräbe PJ. Quantifying the economic benefits of gauge changes on the South African core railway network.
J. S. Afr. Inst. Civ. Eng. 2020:62(3), Art. #0936, 13 pages. http://dx.doi.org/10.17159/2309-8775/2020/v62n3a3 23
South Africa’s core railway
network and key corridors
The South African core network com-
prises 60% of the entire network, while
the remainder of the lines are classified as
“branch lines” (Transnet 2016). The inter-
modal freight lines are rated for 20 tonnes
per axle loading, while the iron ore line
operates on 30 tonne axle loading and the
coal line on 26 tonne axle loading.
Perway condition and railway
line utilisation
Perway is referred to as a section of track
with its numerous components (McNaught
2015). Transnet in more recent years have
placed a larger focus on maintaining assets
rather than constructing new railway
lines. However, the core railway network
is either in an average or poor condition
(Transnet 2016).
Key corridors in South Africa
Transnet (2016) indicates that the South
African core network consists of 18 sec-
tions that can be categorised into the fol-
lowing four systems:
QQ Iron ore and manganese system
QQ Coal system
QQ North-eastern system
QQ Intermodal and general freight system.
The coal export and iron ore export
lines generate 60% of Transnet’s revenue
and together move more than 150 mtpa
(Transnet 2018). The general freight and
intermodal railway corridors struggle to
compete with the road freight transport
sector which holds an 87.5% market share
(Department of Environmental Affairs
2015).
Transnet’s Market Demand Strategy (MDS)
The MDS was compiled by Transnet in
2012 to curb the loss of the freight trans-
portation market share to road. The MDS
seeks to spend more than R330 billion over
the next 30 years and plans to transport
350 million tonnes of freight in the year
2020 (Transnet 2012). Barradas (2014
Engineering News) states that the MDS
intends creating a modal shift from road
to rail.
DoT’s National Rail Policy (NRP)
White Paper
The NRP was compiled by the DoT and
made public in 2017. Figure 1 displays an
essential South African core railway net-
work presented by Van der Meulen (2010)
which is similar to the one presented by
24
Legend
Narrow gauge freight orientated
Standard gauge passenger orientated
Standard gauge freight orientated
Standard gauge very high speed
Hotazel Sishen
Kimberley
Noupoort
Saldanha
Figure 1 A
 n essential South African core railway network (adapted from Van der Meulen 2010)
the DoT in 2017. The line consists of the
following:
QQ 744 km of high-speed standard-gauge
rail (in excess of 250 km/h)
QQ 2 691 km of freight-orientated
standard-gauge
QQ 900 km of passenger-orientated
standard-gauge.
Bridge, tunnel and track structure costs
Railway construction often incorporates
bridges and tunnels to avoid steep gradi-
ents; however, a premium is added onto the
normal track construction costs.
Rostami et al (2013) indicate that the
cost to tunnel is driven by the uncertainty
of the material being tunnelled through,
as well as the bore diameter and type of
tunnel being constructed. Hadi et al (2016)
indicate that the price to construct a bridge
approximately doubles the cost per kilome-
tre of track constructed, depending on the
location’s topography.
The DoT (2009) and Lombard (2017)
indicate that the cost to construct a
standard-gauge railway line from new is
in the range of 7% more expensive than
constructing a narrow-gauge railway line.
Transnet (2016) published the values of
various projects planned in the Long-Term
Planning Framework (LTPF) which reveals
the expected rates per kilometre of narrow-
gauge railway. Lombard (2017) compared
the costs of constructing a coal line as
either a narrow- or standard-gauge line in
Mozambique.
Rolling stock costs
Railway rolling stock prices vary depend-
ing on the intended function of the
rolling stock. The Railway Gazette (2018)
revealed that wagon prices vary depending
Volume 62  Number 3  September 2020  Journal of the South African Institution of Civil Engineering
Beit Bridge
Polokwane
Waterberg
Pretoria
Johannesburg
Richards Bay
Bloemfontein
Durban
Ngqura
on their purpose and are similar in price
regardless of the gauge that they run on.
The Railway Gazette (2018) and Lombard
(2017) both indicate that standard-gauge
locomotives can cost around 50% of the
price of a narrow-gauge diesel or electric
locomotive.
Railway operating costs
Harris (1977) developed a relationship
between operational cost and the density
of a railway line, which displayed that the
greater the density the lower the operating
costs incurred due to economies of scale.
Havenga et al (2012) used the relation-
ship postulated by Harris (1977) to estab-
lish the operating costs of rail operations
in order to identify the cost differences
between road and rail transport. Havenga
et al (2012) were also able to establish the
externality costs associated with road and
rail transport modes to carry freight.
Railway revenue calculation
Revenue is generated through the trans-
portation of freight along a railway line for
clients. Transnet (2018) indicates that the
coal line and the iron ore lines generate
R0.279 per tonne-kilometre and R0.122
per tonne-kilometre respectively. The out-
comes of Transnet’s 2017 and 2018 finan-
cial reports indicate that the average rate
per tonne is between R174.95 and R186.75,
which varies depending on the type of
commodity and the distance required for
it to be transported.
Infrastructure economic
evaluation techniques
Three commonly used project evalu-
ation techniques include the present
worth method, the internal rate of return
 Wang (2007). The main inputs have been
summarised in the figure, with each input
having a further set of inputs and param-
Scenarios eters required, such as various rolling stock
Scenarios
Scenarios characteristics and depot/station spacing.
The operational strategies represent the
real-time control operations aiming at ser-
Main inputs vice restoration. The simulation outputs a
large set of performance measures relating
Network Simulation to the simulation operations.
Dessouky and Leachman (1995)
Signal design QQ Vehicle movements Operational
strategies
indicate that a good method of evaluat-
QQ Interactions
Schedule ing railway alternatives is through the
QQ Depot operations
Rolling stock use of simulation which focuses on the
difference in total delay accumulated in
Demand
an alternative, and a proposed status quo
alternative. Sogin et al (2013) simulated
Output a single-line railway with sidings every
Measures of 24 km and were able to transport up to
performance
40 freight trains per day with zero pas-
senger train trips per day.

Figure 2 A
 rchitecture of a simulation model (redrawn from Koutsopoulos & Wang 2007)
Analysis of a
railway project
Review corridor
Include all
Establish cost
alternatives being
model
evaluated
Figure 3 S ummary of the methodology for analysing a railway project
(IRR) method and the benefit/cost ratio
(Remer & Nieto 1995). The present worth
method is described as a net present value
technique and is limited by the fact that
it assumes that each year the investment
is reinvested at the same rate, which is
not always true when growth is related to
gross domestic product (GDP) (Blank &
Tarquin 1989).
The IRR method is a rate-of-return
technique and is used to calculate the
interest rate for which the present worth of
a project will equal zero. The term internal
indicates that it only considers internal fac-
tors (Park & Sharp-Bette 1990).
The benefit/cost ratio is a ratio tech-
nique and works well as long as the benefits
Economic
evaluation
Simulate most
feasible alternative
Provide
recommendation
can be converted to monetary values. A
project is viewed negatively if the costs out-
weigh the benefits and vice versa (Remer &
Nieto 1995).
Simulation of railway corridors
Simulation is the imitation of the opera-
tion of the real-world process or system
over time and therefore generates artificial
history of a system (Banks et al 2005). The
observation of that artificial history is used
to infer on the operating characteristics of
the real system.
Figure 2 displays the architecture of
a simulation model, which indicates its
inputs, scenarios, operating strategies and
outputs, as postulated by Koutsopoulos and
METHODOLOGY
This section discusses the method of
analysing the freight growth along a
corridor as predicted by Transnet (2016)
before the review and identification of the
various relevant standard-gauge interven-
tion alternatives for each corridor are
conducted.
The methodology also assesses how
the various projects will be evaluated
using different economic evaluation meth-
ods that have been described by Remer &
Nieto (1995).
AnyLogic agent-based simulation mod-
elling was used to model the operations
of the standard-gauge alternatives that
outperformed the base case.
Figure 3 displays the summary of the
methodology for analysing a railway proj-
ect. The procedure displayed was followed
for the analysis of each of the corridors.
Identification of corridors
for intervention
The four principal factors that were used
to determine how a corridor was operating
were annual tonnage, freight growth, infra-
structure condition and capacity. Figure 4
displays the various parameters analysed
for each of the four principles.
Annual tonnage for a corridor
The annual tonnage is the key driver
of revenue generation along a corridor.
Figure 5 displays the present worth of
hypothetical railway corridors to consider
whether a corridor would have a positive

Journal of the South African Institution of Civil Engineering  Volume 62  Number 3  September 2020 25

(1)
Annual
tonnage
(2)
Freight
growth
Figure 4 M
 ethod of reviewing a railway corridor
1 200 000
1 000 000
Present worth (R million)
800 000
600 000
400 000
200 000
0
–200 000
–400 000
–600 000
200
5 mtpa
Figure 5 N
 arrow-gauge to standard-gauge conversion project evaluation for a 30-year analysis
period and 2.5% discount rate
or negative present worth if a narrow- to
standard-gauge conversion were performed
on a double-lined track.
The calculation was influenced by the
amount of tonnage transported per year for
a 30-year analysis period on corridors of
varying length with all possible operational
and capital expenditure costs included. The
present worth reduces as the line length
increases, due to increased capital expen-
diture of the associated infrastructure for
the same amount of tonnage being trans-
ported. This allows corridors in the core
network to be evaluated by comparing their
expected tonnage transported per year and
line length of the corridor with the data in
the figure. The present worth can then be
estimated from the figure. For example,
a corridor that transports 50 mtpa and
is 400 km in length, would be expected
to have a present worth of approximately
R200 000 million if it were converted to
standard gauge. Hudson et al (1997) indi-
cate that the discount rate is the interest
rate less the inflation rate, ultimately giving
the true return of the investment.
26
growth rates was conducted to ensure that
the best alternative is selected per analysed
Electrical
corridor (Remer & Nieto 1995).
Formation
Directional volume split (3)
Infrastructure Track structure
Commodity composition condition
Infrastructure condition
OHTE
The condition of the infrastructure on a
Overall rail corridor is a good indicator of when
maintenance or upgrades are likely to take
Theoretical
place, as well as to what level this would be
Short term (1–5 years) Operational
required. Transnet (2016) provided a sum-
(4)
Medium term (15 years) Practical
Capacity mary of the rail network’s condition, which
Long term (30 years) Used
could be used to identify which lines would
Current and
future traffic require maintenance or an upgrade.
Line capacity
The practical capacity utilisation, which
is 65% of the theoretical capacity, could be
used to assess how congested a corridor
is and if it has capacity available to meet
demand increases on the corridor over
time (Transnet 2016). The practical capa­
city utilisation was used to evaluate how
long a line would be able to operate before
upgrades are required.
Economic evaluation of
railway projects
400 600 800 1 000 1 200 1 400 1 600 The key factor in determining whether an
Route length (km) alternative can be selected is the cost that
20 mtpa 50 mtpa 100 mtpa 150 mtpa is associated with the alternative com-
pared to the base case. Therefore, a base
comparison case was selected (Stanford
University 2005). Transnet’s MDS was
selected as the base case, since the MDS
Tonnage growth is the published plan Transnet intends on
Transnet (2016) indicates that a key driver of following and implementing over the next
the freight growth rate is GDP and predicts 30 years. If any of the alternatives evaluated
the following forecasted overall freight produced a more profitable outcome than
growth rates per annum: a likely scenario the base case, they were deemed to be more
of 3%, a high scenario of 4.1% and a low beneficial than the base case.
scenario of 2.1%. It is likely that the growth The economic evaluation of each of the
rates may vary from year to year and from projects was conducted using the proce-
corridor to corridor; therefore, a sensitivity dure provided in Table 1, which has been
analysis on each corridor’s predicted freight adapted from Remer and Nieto (1995).
Table 1 Economic evaluation steps for railway projects (adapted from Remer & Nieto 1995)
Step Description
1 Define a set of railway investment projects for consideration for the corridor
2 Establish the analysis period for economic study
3 Estimate the cash flow profile for each railway project
4 Specify the time value of money or the minimum attractive rate of return (MARR)
5 Examine the objective and establish criteria to measure effectiveness
6 Apply the relevant project evaluation technique(s)
7 Compare each project proposal for preliminary acceptance or rejection
8 Perform sensitivity analysis
9 Accept or reject a proposal based on established criteria
Volume 62  Number 3  September 2020  Journal of the South African Institution of Civil Engineering
Table 2 A
 ssumptions made in economic calculations (DoT 2009; Lombard 2017; UIC 2017; Joubert Cost model
et al 2001) To determine whether a line could benefit
Item Assumption from a standard-gauge intervention, an
economic method of evaluation was
Axle loads Axle loads would vary depending on the alternative selected for the section.
used over the life of the project. Benefits
Bridges
Bridges would not require any changes except for older, narrower steel include increased profitability, increased
bridges. Allowance must be made for road-over-rail bridges. line capacity to meet long-term traf-
All construction projects were assumed to have 10% contingencies attached fic demands, infrastructure condition
Contingencies
to them. improvements and improved rolling stock
Curvature Would remain as before for line upgrades but may change for new lines. condition. However, the main criteria used
was the comparison of the various projects’
Double-line centres Double lines would remain at 4.00 m centres for line upgrades.
profitability.
For upgrades, banks and formations – would be widened on both sides and Table 2 displays the economic assump-
Earthworks
culverts would be extended.
tions which were required for the various
Electrification Standard-gauge locomotives would require 25 kV AC electrification. alternatives to be compared to the base
Would require widening for line upgrades since standard-gauge track is case. The assumptions were similar to
Horizontal clearance
200 mm wider. the assumptions made by the DoT (2009),
Horizontal curves were not made sharper than 500 m for standard-gauge Lombard (2017) and Joubert et al (2001).
Horizontal curves interventions. High-speed rail requires a curve radius of between 2 500 m and Figure 6 displays the methodology used
5 500 m. to produce the cost model that was used
Would remain as before for line upgrades. For new lines, the ruling grade was to evaluate each of the alternatives for a
selected as 1:100. This allows for 38 wagons to be hauled per locomotive that corridor. The base case and the alterna-
Ruling grades outputs 4 540 kW of power at 454 kN of continuous tractive effort. High-speed tives were all assessed in the same manner
passenger trains were restricted to maximum grades of 1:65 for mixed traffic
instances. to allow for uncomplicated comparison
between all the cases.
Signalling If a third rail were implemented, major upgrades would be required.

Maximum speed would remain at 80 km/h for freight haul. High-speed rail Revenue
Line speed speed was set at 250 km/h and speed restrictions were implemented on tight-
Revenue was used in the analysis for the
curve sections.
true return of the various investments to
Track centres in Yard centres for upgrades would remain the same as currently for narrow- be compared.
yards gauge lines.

Increased vertical and horizontal clearances were required for tunnels if Revenue = r ∙ ttotal ∙ (1 + id)n(1)
Tunnels
upgrades were proposed.

Vertical clearance
Vertical clearance would need to be adjusted if double-stacking were Where:
considered on the line. r is the average rate per tonne,
Vertical curves are designed as a function of line speeds using RV = 0.39 ∙ V2, id is the discount rate, equal to the
where V is in km/h. For line speeds of 250 km/h RV ~ 25 000 m and for line interest rate less the inflation rate and
Vertical curve radius
speeds of 80 km/h RV ~ 2 500 m. Speed restrictions could be implemented for is the true return of an investment
high-speed passenger trains.
(Hudson et al 1997),
t total is the total tonnage for the entire
analysis period, and
n is the analysis period in years.

2015 Capital expenditure Stats SA (2019) indicates that South Africa

(1) (4)
2030 (predicted) Operational costs has achieved an inflation rate of 4.5% per
Tonnage Costs
2045 (predicted) Externality costs annum since 2010, and this was selected as
the inflation rate. The interest rate was set
Track length at 7% (South African Reserve Bank 2019).
(2) Bridge and (5)
Rate per tonne
Infrastructure tunnel length Revenue Costs
Stations and loops Costs can be broken down into operational,
capital expenditure (CAPEX) and rolling
Capacity utilisation stock costs. Maintenance was assumed
Cash flow
Wagon configurations to be equivalent for all alternatives per
(3) (6) Sensitivity analysis
Operations Axle loading Summary on freight growth kilometre, which is the same assumption
Slot spacing Review cash flows that was made by Lombard (2017) and
DoT (2009).
Operational costs were calculated
using the Harris relationship, which is
Figure 6 C
 ost model setup for evaluation of a railway investment calculated by dividing the tonne route

Journal of the South African Institution of Civil Engineering  Volume 62  Number 3  September 2020 27
 kilometres by the track kilometres (Harris
Cost per tonne km (R/tonne – route km) 0.40 1977). An externality cost, which accounts
0.35 for rail’s impact on the environment, was
also included in the operational costs. The
0.30
factor calculated by Havenga (2012) and
0.25 the Department of Environmental Affairs
0.20 (2015) came to R 0.0129/tonne-km. The
operating costs obtained from Transnet’s
0.15
Integrated Report from 2017 for the Natal
0.10 corridor, Transnet’s Integrated Report from
0.05 2018 for the coal line and Lombard’s (2017)
study were used to verify the Harris rela-
0
0 10 20 30 40 50 60 tionship and were plotted on Figure 7.
Density (million tonne route km/track km) Table 3 displays the costs to construct
Harris curve (1977) Lombard (2017) ballasted track, railway bridges and tun-
Transnet (2018) – Coal line Transnet (2017) – NATCOR nels. The cost of constructing ballasted
railway track was obtained from a recent
Figure 7 I ntermodal density change on the Harris curve due to density-driven savings project completed by Transnet (2017) and
(adapted from Harris 1977) a study conducted by Lombard (2017). The
cost of constructing a railway bridge was
Table 3 I nfrastructure costs per kilometre, average locomotive and wagon prices (Transnet 2017; acquired from a study conducted by Hadi
Lombard 2017; Rostami et al 2013; Joubert et al 2001; Hadi et al 2016; Railway Gazette et al (2016) on numerous bridge construc-
2018; Barradas 2014; Transnet 2018; Lombard 2017) tion projects. The rate per metre of tunnel
is related to the tunnel’s bore diameter.
Rate (R million/km)
Infrastructure type These rates were gathered from a study
Narrow-gauge Standard-gauge conducted by Rostami et al (2013). Table 3
Single track ballast 62.0 65.8 also displays the variation in costs between
narrow- and standard-gauge locomotives
Double track ballast 124.0 131.6
which operate on either diesel or electricity.
Bridge single track 73.8 83.9 Lastly, the table summarises the costs of
purchasing the two types of wagons, and it
Bridge double track 147.7 167.9
was noted that standard- and narrow-gauge
Tunnel single track 472.2 512.6 wagons could be bought for similar prices
(Lombard 2017; Transnet 2018).
Tunnel double track 944.3 1 073.7
Lombard (2017) states that the number
Locomotive type Electric (R million/unit) Diesel (R million/unit) of wagons per locomotive is a function of
Narrow-gauge 41.0 32.0 the ruling grade, while Dutton (2014) indi-
cates that, to prevent damage to the wheels
Standard-gauge 25.8 20.2
of a train due to heat, the heat input per
Container CR (R million/unit) Tanker (R million/unit) wheel should not exceed 12 kW in South
African conditions. These checks were
Wagons 1.2 2.5
therefore performed when determining the

MDS plans Base case

Narrow gauge

Dual gauge bogie

Third rail
Alternatives
Bogie changing

Conversion and transhipment Alternative cases

Standard gauge

Ring-fenced line

Figure 8 A
 lternatives for possible use in the cost analysis

28 Volume 62  Number 3  September 2020  Journal of the South African Institution of Civil Engineering
number of wagons per locomotive for the
various railway corridors.
Rank the cases
from most to least
Alternatives for a gauge conversion profitable
Figure 8 displays the various alternatives
which could be implemented along the
different corridors. As each corridor would Main inputs: Economic evaluation methods:
QQ Analysis period
Is there a discrepancy
QQ Present worth
be better suited to specific alternatives (and in the ranking of
QQ Cash flows QQ IRR
each analysis?
not to others), the relevant alternative was QQ Discount rate QQ Benefit/cost ratio

selected for each corridor.

No Yes
Cashflow and economic
evaluation techniques
For each alternative evaluated on a cor- Conduct
Select most
incremental
ridor, the total cost, including operational profitable case
analysis
costs and CAPEX, revenue generation per
year and cash flow, was calculated. The
cash flow was taken as the amount of Figure 9 M
 ethod of economically evaluating a project
revenue less the total costs per year before
taxation, interest, amortisation and depre- Table 4 Properties required to compile a simulation model
ciation, therefore providing the net profit Property Description Units
or cashflow (Hudson et al 1997).
Axle loading Axle loading rating for the line tonnes/axle
The present worth method was used in
conjunction with the benefit/cost ratio and Crossing loop length Length of the crossing loop km
IRR methods. Figure 9 displays the method
Crossing loop spacing Distance between crossing loops km
used to economically evaluate a set of
railway projects. Curve radius Maximum curve radius m

Freight mix Distribution of commodity wagon types –

Simulation model setup
Gradient 1:Ψ where Ψ is the steepest gradient –
Simulation was used to assess the effects of
performing a standard-gauge intervention Line length Length of the line km
on a corridor if the standard-gauge inter- Line speed Maximum speed for the line km/h
vention was the most economical alterna-
Line type Single or double line –
tive for the corridor.
Number of stations/sidings Stations/sidings between the end points of the line –
Identification of line properties Time in depot load/unload Time to load or unload a set of wagons hours
To simulate a railway corridor, various
Traction Diesel or electric –
line property, rolling stock parameters
and operational information are required. Train acceleration/deceleration Rate at which the train accelerates or decelerates m/s2
Table 4 displays the critical properties Trains in system Total number of trains in the system –
required to simulate a railway. These prop-
Wagons per train Number of wagons per train –
erties vary depending on which corridor is
being simulated.
The simulation was conducted using was within 10% of the observed true mean RESULTS AND DISCUSSION
AnyLogic because of the software’s diver- (Balakrishna et al 2007). Since a freight- This section firstly summarises the
sity and ability to perform simulations at a orientated standard-gauge railway does not core network and identifies which
microlevel. The model utilised AnyLogic’s exist within South Africa, the observed val- corridors would potentially benefit from a
rail library, as well as the process ues were calculated using first principles, standard-gauge intervention. A discussion
modelling library. or were in other instances collected from then follows on whether it would be
known standards and documents. economically feasible to construct a
Simulation model parameter calibration, Optimisation of the network was standard-gauge intervention on any of the
validation and optimisation performed with the main aim of reducing South African railway corridors. Lastly,
Data inputs for a simulation model can delay, which is the additional time it takes the outcomes of the simulation model
either be measured directly, or if the inputs a train to travel on a route due to conflicts are discussed.
are unknown, they can be calibrated with other traffic (Dingler et al 2010). Two
(Koutsopoulos & Wang 2007). parameters were varied to optimise the Review of the core
The root mean square percentage proposed standard-gauge intervention, network corridors
error (RMSPE) was used to observe the which included varying the crossing loop Table 5 displays the corridors identified for
differences between the simulated and spacing of single lines and varying the analysis. The nine corridors in the table
observed values, and an acceptable RMSPE number of wagons per train. form part of the following five systems:

Journal of the South African Institution of Civil Engineering  Volume 62  Number 3  September 2020 29
QQ Natal corridor Table 5 S ummary of the future transported volume and the overall condition and utilisation of South
QQ Coal line Africa’s key corridors as analysed for 2030 and 2045, compared to 2015 (Transnet 2015)
QQ Iron ore line
Total volume Corridor
QQ Manganese line Overall
transported (mtpa) growth Utilisation
Corridor condition
QQ North-eastern system. rate per 2045 (%)
2015 2030 2045 2015
annum (%)
Table 5 indicates that as of 2015, only
the coal export line and the iron ore line Natal corridor: Average to
26.28 50.49 73.44 3.96 >130
exceed freight volumes of 30 mtpa. The Reitvallei – Booth poor
Natal corridor is predicted to grow in the
years to follow and to exceed 50 mtpa by Coal line: Average to
89.61 113.62 133.44 1.47 105–130
2030. By observing the overall line condi- Lephalale – Ogies poor

tion, it was also noted that the only line in

Coal line:
good condition is the iron ore line. Most of Ogies – Richards Bay
100.52 107.3 153.31 0.93 Average >130
the network would reach a state of collapse
if no upgrades are performed as indicated Iron ore line:
62.02 63.22 73.83 0.55 Good 105–130
by the utilisation percentage of each cor- Sishen – Saldanha
ridor in the year 2045.
Manganese line:
6.67 16.35 27.85 5.52 Average >130
Hotazel – Ngqura
Results of the economic
evaluation of railway projects
NE System: Average to
The analysis period was set at 30 years, 10.3 14.63 20.84 2.37 95–105
Groenbult – Kaapmuiden poor
the discount rate at 2.5% and the freight
growth rate was varied depending on the NE System: Average to
20.68 26.57 36.15 1.78 80–95
corridor’s predicted growth rate. A sensi- Greenview – Komatipoort poor
tivity analysis was conducted on the freight
growth rates for each of the corridors North-eastern system:
8.1 10.25 14.15 1.73 Average 80–95
Komatipoort – Richards Bay
identified and the construction duration
was also varied for the alternative cases.
North-eastern system:
The analysis results were not sensitive to 1.94 5.1 10.69 6.25 Average 95–105
Musina – Pyramid
the discount rate.
Table 6 displays the results of the
economic analysis conducted on the north- Table 6 E conomic analysis results for the north-eastern, manganese and Gauteng to Cape systems
eastern, manganese and Gauteng to Cape
Corridor Project B/C ratio IRR PW (R million)
systems. The results in all cases indicate
that following the plans described in the MDS 1.186 0.028 19 096
MDS would be the most profitable since Gauteng – Cape Town 3rd Rail 0.226 –0.093 –297 118
their standard-gauge project’s construction
cost is too high for the volumes trans- Conversion 0.242 –0.090 –271 897

ported on these lines. MDS 1.273 0.042 40 775

Table 7 displays the various alternatives
Manganese system 3rd Rail 0.410 –0.067 –180 129
that were evaluated to determine which
case would result in the most economically Conversion 0.336 –0.077 –250 640
beneficial project for the coal line, iron ore
MDS 2.395 3.354 29 669
line and the Natal corridor. Ruling gradi-
North-eastern system
ents, route lengths and axle loads for each 3rd Rail 0.374 –0.059 –70 626
Musina – Pyramid
alternative have been included in the table.
Conversion 0.394 –0.057 –64 773
The Natal corridor cannot have a ruling
gradient of less than 1:80 between Durban MDS 3.856 2.050 93 180
and Pietermaritzburg; therefore, train North-eastern system
3rd Rail 0.950 –0.004 –613
reconfiguration is required once the train Groenbult – Kaapmuiden

has traversed this section of the corridor. Conversion 0.939 –0.005 –2 012
Table 8 displays the outcomes of the MDS 3.351 2.500 159 044
economic analysis for the three railway North-eastern system
corridors. From the analysis the MDS is the 3rd Rail 1.387 0.033 68 266
Greenview – Komatipoort
most profitable project for all the corridors. Conversion 1.412 0.035 70 983
Both the coal and iron ore lines operate
MDS 2.007 2.500 44 731
near capacity and are highly valued by
North-eastern system
Transnet. The Natal corridor transports 3rd Rail 0.578 –0.046 –58 383
Komatipoort – Richards Bay
intermodal freight; however, despite
Conversion 0.614 –0.042 –49 762
being able to double-stack containers on a

30 Volume 62  Number 3  September 2020  Journal of the South African Institution of Civil Engineering
Table 7 The coal line, iron ore line and Natal corridor alternatives for evaluation
Axle load Route Ruling
Corridor Project Case Project description
(tonnes / axle) length (km) grade

MDS A Follow plans described in the MDS. 26 419 1:160

Convert the Ermelo–Richards Bay section to standard-gauge

Coal line:
Tranship B and tranship coal at Ermelo to increase export volumes along 30 419 1:160
Ermelo–
the Ermelo–Richards Bay section.
Richards Bay
Upgrade the current line to allow for a dual-gauge system along
3rd Rail C 30 419 1:160
the Ermelo–Richards Bay section to increase export volumes.

Iron ore line: MDS D Follow plans described in the MDS. 30 877 1:250
Sishen–
Upgrade the current line to allow for a dual-gauge system. The
Saldanha 3rd Rail E 34 877 1:250
line would remain a single line with crossing loops.

MDS F Follow plans described in the MDS. 20 690 1:100 (1:80)

Natal Upgrade the current line to allow a dual-gauge system. Double-

3rd Rail G 26 690 1:100 (1:80)
corridor: stacking of containers allowed.
Durban–
Gauteng Construct a line parallel to the current NG corridor in SG. The
Ring- fenced
H SG line was assumed to be used for only container traffic in the 26 660 1:100 (1:80)
SG
base case. Current NG corridor to follow the plans in the MDS.

Table 8 C
 oal line, iron ore line and Natal corridor economic evaluation for predicted freight standard-gauge line, none of the standard-
growth rates gauge alternatives would be profitable.
PW Figure 10 displays the present worth
Corridor Project (duration) Case B/C ratio IRR values of the cases F, G.2 and H.2 for vary-
(R million)
MDS A 1.619 0.206 349 924
ing freight growth rates. This was only
conducted for a ten-year construction
Transhipment (6 yrs) B.1 1.285 0.047 204 581
period since it is the more likely situation.
Coal line Transhipment (10 yrs) B.2 1.272 0.046 195 250 For Case H, it was noted that an opportu-
3rd Rail (6 yrs) C.1 1.310 0.051 218 377 nity exists along the Natal corridor because
3rd Rail (10 yrs) C.2 1.297 0.049 209 045 the line would be highly underutilised. Rail
MDS D 1.889 0.937 348 887 held a market share of 15% of the freight
transported along the Natal corridor in
3rd Rail (6 yrs) E.1 1.447 0.037 218 329
Iron ore line 2016 and it is predicted that in 2045 the
3rd Rail (10 yrs) E.2 1.219 0.021 121 643
Natal corridor, including both road and
3rd Rail (15 yrs) E.3 0.884 –0.013 –33 067 rail, could transport 223 million tonnes
MDS F 1.403 0.033 109 682 of freight between Gauteng and Durban
3rd Rail (6 yrs) G.1 0.807 –0.016 –125 067 (Naidoo 2015).
Natal corridor 3rd Rail (10 yrs) G.2 0.782 –0.018 –141 686 In Case H, the single-line standard-
gauge railway could be used to transport a
Ring-fenced SG (6 yrs) H.1 0.973 –0.002 –10 224
large portion of the freight along the cor-
Ring-fenced SG (10 yrs) H.2 0.954 –0.004 –21 315
ridor. From the initial operational analysis,
it was noted that the standard-gauge line
could carry an additional 80 mtpa on top of
500 000 the current 18.1 mtpa of containers, which
400 000 the line would carry from the predicted
Present worth (R million)

300 000 rail traffic.

200 000 Figure 11 displays the present worths
100 000 of Cases F and H, where Case H carries a
0 varying amount of additional tonnage on
–100 000 the standard-gauge single line along the
–200 000
Natal corridor. The freight growth rate
was set at the predicted rate of 3.96% per
–300 000
annum as per Transnet’s expected growth
–400 000
0 1 2 3 4 5 rate. Under these conditions, if the stan-
Freight growth rate (%) dard-gauge line were to carry an additional
F G.2 H.2 + 0 H.2 + 50 H.2 + 80 37.5 mtpa, Cases F and H would have the
same present worth values. Only Case H
Figure 10 S ensitivity analysis of the freight growth rate versus project present worth for cases F, would be able to carry additional tonnage
G.2 and H.2 with additional freight transported per annum without jeopardising the future operations

Journal of the South African Institution of Civil Engineering  Volume 62  Number 3  September 2020 31
on the corridor due to its low utilisation
without the prescribed additional tonnage. 500 000
Transporting commodities with rail
400 000

Present worth (R million)

would also induce an environmental benefit
due to the reduced externality costs such as
accidents, noise, congestion and emissions.
A confidence interval has been included
in Figure 11 for a 95% confidence level
with upper and lower limits. The average
externality cost savings of rail over road
transport were obtained from Havenga et al
(2012). This additional benefit of transport-
ing the commodities on road, as opposed to
rail, indicates that Cases F and H have equal
present worth values between an additional
24 and 30 mtpa.
Ruling grade check, simulation
and optimisation of the
standard-gauge line
From the economic analysis conducted, the
only feasible standard-gauge intervention
which may be possible to implement realis-
tically, was the ring-fenced standard-gauge
single line operating concurrently with
the narrow-gauge line between Durban
and Gauteng.
Ruling grade check
Figure 12 displays the 1:100 railway
line and ground elevation longitudinal
profile between Durban and Roodekop,
300 000
200 000
100 000
0
–100 000
0 10 20 30 40 50 60 70 80
Additional tonnage (mtpa)
F H Enviro savings LL Enviro savings UL
Figure 11 P
 resent worth of additional tonnage, from road corridor, transported on Case H
compared to Case F for predicted freight growth rates
Gauteng. The 110 km between Durban 30 km in length, would be required to be
and Pietermaritzburg had to be designed constructed for the 1:100 ruling grade, as
at a ruling grade of 1:80 to avoid a long well as 25 km of bridges.
tunnel being constructed between these Figure 13 displays a railway line
two points. Each locomotive would be designed at a ruling grade of 1:200; how-
able to haul 25 wagons along this section. ever, the extra tunnelling results in an
The tunnel and bridge limits were set at additional R84.6 billion construction costs.
–30 m and 30 m respectively; therefore, any The reduced grade would allow more wag-
cutting deeper than 30 m would require ons to be hauled per locomotive, ultimately
a tunnel to be constructed and any fill increasing the throughput per locomotive
higher than 30 m would require a bridge per annum. High-speed passenger trains
to be constructed. The section between would be able to operate on either of the
Pietermaritzburg and Roodekop was profiles since the maximum grade is less
designed at 1:100. Five tunnels, totalling than 1:65.

650
1 700
600
1 600
550
1 500
1 400 500

Difference between rail and natural ground (m)

1 300 450

1 200 400
1 100 350
1 000
Elevation (m)

300
900 250
800
200
700
150
600
100
500
50
400
0
300
200 –50

100 –100

0 –150
0 50 100 150 200 250 300 350 400 450 500 550 600 650
Chainage (km)
1:100 Elevation Difference Tunnel limit Bridge limit

Figure 12 E levation and 1:100 railway line longitudinal profile between Durban and Roodekop

32 Volume 62  Number 3  September 2020  Journal of the South African Institution of Civil Engineering
 1 700 850
800
1 600
750
1 500 700
650
1 400

Difference between rail and natural ground (m)

600
1 300 550
1 200 500
450
1 100
400
1 000 350
Elevation (m)

650
900
300
800 250
700 200
150
600
100
500 50
0
400
–50
300 –100
200 –150
–250
100
–300
0 –350
0 50 100 150 200 250 300 350 400 450 500 550 600 650
Chainage (km)
1:200 Elevation Difference Tunnel limit Bridge limit

Figure 13 E levation and 1:200 railway line longitudinal profile between Durban and Roodekop

Gauteng
Figure 14 displays the plan layout of the
standard-gauge single line between Durban
and Roodekop, Gauteng at a ruling grade of
1:100. The line closely follows the current
narrow-gauge line. The colour of the points
indicates the grade at which the line would
be constructed.

Simulation and optimisation

of the standard-gauge line
Table 9 displays the scenarios evaluated in
Grade Colour
the simulation. Five crossing loop spacing
0
distances were evaluated to establish how
1:80
1:100 the number of crossing loops affected the
1:150 number of trains that could pass through
1:200 the system per day. It was noted that, as
1:500 the number of crossing loops on the line
1:800 Durban increased, the average number of trains
Figure 14 Plan view of the ring-fenced standard-gauge single line between Durban and Gauteng per day that could pass through the system
increased. The Scott formula was used to
Table 9 Single-line standard-gauge railway average trains per day verify the results of the agent-based simula-
Crossing loop Average trains per Average trains per
tion and these results have been included
Number of loops in Table 9 (Parkinson 1996).
spacing (km) day (simulated) day (Scott formula)
Figure 15 indicates the average delay for
30 22 23.04 23.40
the varying number of crossing loops. It is
35 19 20.90 20.80 noted that, as the number of crossing loops
on the line increased, the delay experienced
40 16 20.57 18.72 per train is reduced. Less time is spent
waiting at crossing loops when there are
60 11 15.10 13.37
more loops, which leads to reduced inter-
action between trains. Despite the costs
80 8 11.73 10.40
associated with constructing additional

Journal of the South African Institution of Civil Engineering  Volume 62  Number 3  September 2020 33

Train delay per 100 km travelled (mm)
Figure 15 D
 elay experienced by trains per 100 km travelled for a varying number of crossing loops
loops, the system can operate more effi-
ciently, and more loops will lead to greater
throughput per annum.
Figure 16 displays the number of tonnes
transported for varying crossing loop spac-
ing and train lengths. The figure indicates
that running longer trains will lead to more
tonnage being transported by the standard-
gauge line. It is therefore most beneficial to
run 210-wagon trains.
As the number of crossing loops
increases, so does the permissible annual
tonnage along the line. In Figure 10 it was
required that the line be able to transport
98.1 mtpa if the line were used to transport
the 80 mtpa of additional freight from the
road. To meet the 98.1 mtpa, a minimum
of 21 crossing loops should be constructed
at spacings of 30 km, and train sets should
transport, on average, at least 203 wagons
per trip.
CONCLUSIONS
For this study the following conclusions
can be made after the analysis and discus-
sion of the results:
Corridors that could benefit from
a standard-gauge intervention
It was noted from the analysis that cor-
ridors would be required to transport a
large amount of tonnage to economically
warrant an intervention. Three corridors
were identified that could economically
benefit from a standard-gauge interven-
tion, namely the coal export line, the iron
ore line and the Natal corridor. The other
corridors within South Africa are not
expected to transport enough tonnage to
warrant a standard-gauge intervention that
is economically viable with current freight
growth projections.
34
network, therefore the break-of-gauge
60 issue is avoided. From the simulation it
was observed that a standard-gauge single
50
line would be able to transport in excess
of 100 mtpa if 21 or more crossing loops
40
are constructed. The delay experienced
30 from train interactions was reduced as the
number of crossing loops was increased. If
20 the line were to be constructed, 21 cross-
ing loops should therefore be constructed
10
to relieve operational stress on the
system when the line begins to reach the
0
0 5 10 15 20 25 expected capacity.
Number of crossing loops
ACKNOWLEDGEMENT
The authors would like to acknowledge
The best suited interventions and thank Transnet Freight Rail for their
for the corridors identified financial support.
From the economic analysis of the coal
line, iron ore line and the Natal corridor
it was identified that none of the corridors REFERENCES
would be able to outperform the base case. Balakrishna, R, Antoniou, C, Ben-Akiva, M,
It was, however, determined that, for the Koutsopoulos, H & Wen, Y.2007. Calibration
Natal corridor, the standard-gauge line of microscopic traffic simulation models.
would be highly underutilised in the base Transportation Research Record, 1999: 198–207.
case analysis and that an opportunity exist- Banks, J, Carson I I, J S, Nelson, B L & Nicol, D M
ed to transport additional freight, which 2005. Discrete-Event System Simulation. 4th ed.
would otherwise be transported by road. If India: Pearson Education.
an additional 37.5 mtpa were transported Barradas, S. 2014. Transnet’s market demand
on the standard-gauge line, Transnet’s strategy, South Africa. Engineering News, 21 March
Market Demand Strategy would not be the 2014.
most economical option. Blank, L T & Tarquin, A J 1989. Engineering Economy.
3rd ed. New York: McGraw-Hill.
The standard-gauge intervention’s Department of Environmental Affairs (DEA) 2015.
effect on operations Freight shift from road to rail. Pretoria: DEA.
A ring-fenced standard-gauge line along Department of Transport (DoT) 2009. National
the Natal corridor would have very lim- Transport Master Plan. NATMAP 2050 – Rail gauge
ited interaction with the narrow-gauge study report. Pretoria: DoT.
120
Permissible tonnes per annum (mt)
100
80
60
40
20
0
0 50 100 150 200 250
Wagons per train
8 Crossing loops – 80 km 10 Crossing loops – 60 km
16 Crossing loops – 40 km 18 Crossing loops – 35 km
21 Crossing loops – 30 km Minimum annual tonnage required
Figure 16 P
 ermissible tonnes through the system as a function of crossing loop spacing and
wagons per train
Volume 62  Number 3  September 2020  Journal of the South African Institution of Civil Engineering
Department of Transport 2017. National Rail Policy.
Draft White Paper. Pretoria: DoT.
Dessouky, M M & Leachman, R C 1995. A simulation
modelling methodology for analysing large complex
rail networks. Simulation, 65: 131–142.
Dingler, M, Koenig, A, Sogin, S & Barkan, C P
2010. Determining the causes of train delay.
Proceedings, 2010 Annual AREMA Conference,
Orlando, FL.
Dutton, C J 2014. Rolling stock: Train consist design
theory. Technical Note. Johannesburg: Transnet
Engineering Services.
Hadi, M, Orabi, W, Xiao, Y & Jia, J 2016. Estimating
total cost of bridge construction using Accelerated
Bridge Construction (ABC) and conventional
methods of construction. Miami, FL: Florida
International University, Department of Civil and
Environmental Engineering.
Harris, R 1977. Economics of traffic density in the rail
freight industry. The Bell Journal of Economics, 8(2):
556–564.
Havenga, J, Simpson, Z & De Bod, A 2012. South
Africa’s domestic intermodal imperative. Research
in Transportation Business and Management,
5: 38–47.
Hudson, W R, Haas, R & Uddin, W 1997. Infrastructure
Management, New York: McGraw-Hill.
Joubert, H S, Pretorius, C, Morgan, J & Mntonintshi, U
2001. Gautrain: Route determination. Proceedings,
20th South African Transport Conference,
16–20 July 2001, Pretoria.
Journal of the South African Institution of Civil Engineering  Volume 62  Number 3  September 2020
Koutsopoulos, H N & Wang, Z 2007. Simulation
of urban rail operations: Application
framework. Transportation Research Record,
2006(1): 84–91.
Lombard, P C 2017. Composing a heavy haul
engineering symphony in differential life cycle
costing. Cape Town: International Heavy Haul
Association.
McNaught, M D 2015. A risk-reliability comparison of
track sections in the passenger railway industry. MSc
Dissertation, Stellenbosch University.
Naidoo, I 2015. SIP2 Logistics and industrial corridor.
Proceedings, Edge Seminar, 19 February 2015,
Durban.
Park, C S & Sharp-Bette, G P 1990. Advanced
Engineering Economics. 1st ed. New York: Wiley.
Parkinson, T 1996. Rail transit capacity. TCRP
Report 13. Washington, DC: Transportation
Research Board, National Research Council.
Railway Gazette 2018. World rolling stock market. April
2018. Available at: https://www.railwaygazette.com/
traction-and-rolling-stock/world-rolling-stock-
market-april-2018/46257.article.
Remer, D S & Nieto, A P 1995. A compendium and
comparison of 25 project evaluation techniques.
Part 1: Net present value and rate of return
methods. International Journal of Production
Economics, 42(1): 79–96.
Rostami, J, Sepehrmanesh, M, Gharahbagh, E A
& Mojtabai, N 2013. Planning level tunnel
cost estimation based on statistical analysis of
historical data. Tunnelling and Underground Space
Technology, 33(2013), 22–33.
Sogin, S L, Lai, Y C, Dick, C T & Barkan, C P L 2013.
Comparison of capacity of single- and double-
track rail lines. Transportation Research Record,
2374(1): 111–118.
South African Reserve Bank (SARB) 2019. Rates.
Pretoria: SARB.
Stanford University 2005. Guidelines for Life Cycle Cost
Analysis (LCCA). Land and buildings. Stanford, CA:
Stanford University.
Stats SA 2019. Falling fuel prices dragging inflation
down. Pretoria: Statistics South Africa.
Transnet 2012. Market demand strategy. Johannesburg:
Transnet Freight Rail.
Transnet 2015. Integrated Report 2015. Johannesburg:
Transnet Freight Rail.
Transnet 2016. Long-term planning framework. In:
Freight demand plan. Johannesburg: Transnet,
pp 13–47.
Transnet 2017. Integrated Report 2017. Johannesburg:
Transnet Freight Rail.
Transnet 2018. Freight rail 2018. Johannesburg:
Transnet Freight Rail.
UIC (International Union of Railways) 2017. High-speed
rail – Fast track to sustainable mobility. Paris: UIC.
Van der Meulen, R D 2010. How might the railway
renaissance evolve in South Africa? Proceedings,
29th Southern African Transport Conference,
Pretoria, pp 630–639.
35
TECHNICAL PAPER Physical model study of
Journal of the South African
Institution of Civil Engineering bedrock scour downstream
ISSN 1021-2019
Vol 62 No 3, September 2020, Pages 36–52, Paper 1000 of dams due to spillway
ADÈLE BOSMAN PrEng is a lecturer in
Hydraulics at the University of Stellenbosch,
plunging jets
South Africa. She is currently studying
towards her PhD (Civil) degree in
developing a numerical model for A Bosman, G R Basson
predicting rock scour. She has more than 11
years’ experience mainly in river hydraulics
and the design of large hydraulic structures.
The erosive power of a free-falling high-velocity water jet, flowing from a dam spillway, could
Contact details: create a scour hole downstream of the dam, endangering the foundation of the dam. Despite
Civil Engineering Department
Water Division
extensive research since the 1950s, there is presently no universally agreed method to predict
Stellenbosch University accurately the equilibrium scour depth caused by plunging jets at dams. These formulae yield
Private Bag X1 a large range of equilibrium scour dimensions. The hydrodynamics of plunging jets and the
Matieland 7602
subsequent scour of a rectangular, horizontal and vertical fissured rock bed were investigated
T: +27 21 808 4356
E: abosman2@sun.ac.za in this study by means of a physical model. Equilibrium scour hole geometries for different
fissured dimensions (simulated with rectangular concrete blocks tightly prepacked in a regular
PROF GERRIT BASSON PrEng MSAICE is a rectangular matrix), for a range of flow rates, plunge pool depths, and dam height scenarios
professor in Hydraulic Engineering in the were experimentally established with 31 model tests. From the results, non-dimensional
Civil Engineering Department at
formulae for the scour hole geometry were developed using multi-linear regression analysis.
Stellenbosch University. He obtained a PhD
from Stellenbosch University in 1996 and The scour depth results from this study were compared to various analytical methods found
has more than 30 years’ experience mainly in literature. The equilibrium scour hole depth established in this study best agrees with that
in the fields of river hydraulics, fluvial predicted by the Critical Pressure method.
morphology and the design of large
hydraulic structures. He has worked on projects in 21 countries and is an
Honorary Vice-President of ICOLD (International Commission on Large
Dams). Introduction depth. New formulae were developed to
Contact details: A free-falling high-velocity water jet, determine the scour hole geometry (depth,
Civil Engineering Department flowing from a dam spillway, could scour length, width and volume) based on stream
Water Division the bedrock downstream of a dam if the power and the rock’s movability number
Stellenbosch University
Private Bag X1
erosive capacity of the jet is large enough. and settling velocity.
Matieland 7602 The foundation of the dam could be endan-
T: +27 21 808 4355 gered by the scour hole. It is therefore
E: grbasson@sun.ac.za
imperative to determine the equilibrium Rock scour
geometry of the scour hole to incorporate Bedrock scour downstream of a dam is a
it at the design stage to ensure the stability complex physical process and the complete
of the dam. understanding of the air-water-rock phase
Well documented cases, such as interaction in the scouring process is
Kariba Dam on the Zambezi River (border required. The free-falling jet could be split
between Zimbabwe and Zambia, Africa) into three different modules, namely the
and Wivenhoe Dam in Australia are prime falling jet module, plunge pool module and
examples where the plunge pool scouring rock mass module as defined by Bollaert
had drastic consequences. The depth of and Lesleighter (2014) in Figure 1.
the scour hole of Kariba Dam is approxi- Various researchers have studied the
mately two-thirds of the dam height of hydrodynamics of a free-falling jet in the
128 m (Noret et al 2012), while that of the air and the plunge pool. It was found that
Wivenhoe Dam is one-third of the dam the fall height (Hw), issuance velocity (Vi),
height of 33.87 m (full supply level to spill- discharge, air entrainment, outlet shape
way bucket lip) (Stratford et al 2013). and initial turbulence (Tu) influence the
The purpose of this study was to behaviour of the jet as it falls through
simulate rock scour due to plunging jets the air (Van Aswegen et al 2001). The jet
in a physical model and to gain a greater trajectory and energy at impingement with
understanding of the scouring process. The the plunge pool surface are dependent
physical model results were then compared on the velocity of the jet (Vj ), whereas the
Keywords: erodibility, impinging jets, physical model, plunge pool, to the predictions of different analytical breakup length (Lb : distance for the jet
scour methods of the equilibrium rock scour core to dissipate in the air) is reliant on

Bosman A, Basson GR. Physical model study of bedrock scour downstream of dams due to spillway plunging jets.
36 J. S. Afr. Inst. Civ. Eng. 2020:62(3), Art. #1000, 17 pages. http://dx.doi.org/10.17159/2309-8775/2020/v62n3a4
the initial jet turbulence intensity (Manso
et al 2008). Reservoir water level
Manso et al (2008) investigated the
Di or Bi
behaviour of the jet as it travels through Vi

the plunge pool. The jet velocity (Vj ),

impact angle (θ) and turbulence at enter-
ing the plunge pool surface, as well as Falling jet
Hw module
the plunge pool depth (Y ) and geometry,
δout
determine the amount of energy that would
be dissipated and how much of the jet core
remains intact at impact with the bedrock. θj Tailwater level
Bj , Vj
According to Ervine and Falvey (1987), the
inner core diffusion angle (αin) of highly yw Original bed level Plunge
turbulent plunging round jets when travel- αout pool
αin Y module
ling through a plunge pool is approximately ys
8°, whereas the outer expansion angle (αout) Rock
mass
can increase up to 15°. module
The plunge pool bed would scour if
the erosive capacity of the jet exceeds the Xult
ability of the rock to resist it. To deter-
mine the conditions of the jet at impact Figure 1 P
 lunging jet modules involved in the rock scouring process (adapted from Bollaert 2002)
with the rock bed, the velocity and pres-
sure decay in the plunge pool need to be Table 1 Empirical formulae to predict plunge pool equilibrium rock scour depth
defined. Ervine and Falvey (1987) defined Formula (empirical) Author(s)
the maximum pressure of a circular jet at
a certain distance below the plunge pool’s H 0.225 q 0.54 Veronese 1 (1937, as cited in
yw + ys = 3.68  w Whittaker & Schleiss 1984)
free surface in terms of the velocity head d90 0.42
and pressure coefficients to account for
the fluctuating nature of the turbulent Veronese 2 (1937, as cited in
yw + ys = 1.32 Hw0.225 q 0.54
jet. Castillo et al (2014) determined the Wittler et al 1995)
mean and fluctuating pressure coefficients
for rectangular jets. Factors affecting the Veronese 3 (1937, as cited in
yw + ys = 1.9 Hw0.225 q 0.54
vulnerability of the erodibility of the rock Noret et al 2012)

are lithology, rock strength, joint spac-

ing, joint orientation and joint condition Yildiz & Uzucek (1994, as cited
yw + ys = 1.9 Hw0.225 q 0.54sinθT
in Alias et al 2008)
(Annandale 2006).
Extensive research has been conducted
on the prediction of rock scour depth due Damle (1966, as cited in Noret
yw + ys = 0.362 Hw0.5 q 0.5
et al 2012)
to plunging jets since the 1950s. However,
there is no universally agreed upon method
to accurately predict the equilibrium scour Chee & Kung (1974, as cited in
yw + ys = 1.663 Hw0.2 q 0.6
Noret et al 2012)
depth. Rock scouring is a complex physi-
cal process that is normally assessed in
Martins (1975, as cited in Noret
practice by empirical methods combining yw + ys = 1.5 Hw0.1q 0.6
et al 2012)
laboratory and field observations with
some physics (Bollaert 2002).
q 0.6(1 + β)0.3 ys0.16 Mason-B (1989, as cited in
Empirical or semi-empirical formulae, yw + ys = 3.39 
g 0.3 dm0.06 Castillo & Carrillo 2014)
developed from physical model tests and
prototype observations generally used to
3.27q 0.6Hw0.05 yw0.15
predict the plunge pool scour depth, are y w + ys =   Mason & Arumugam (1985)
g 0.3 dm0.1
summarised in Table 1. Most of the empiri-
cal formulae for rock scour rely on the fall
H 0.35 q 0.7 Kotoulas (1967, as cited in
height and unit discharge. yw + ys = 0.78  w Whittaker & Schleiss 1984)
d950.4
In Table 1 yw is the plunge pool
⎫y ⎫
yw + ys = 0.6 q 0.5Hw0.25⎪ s ⎪
depth (m), ys is the scour hole depth (m), 0.333

⎭ dm ⎭
Jaeger (1939, as cited in
Hw is the effective head (m), q is the unit Castillo and Carrillo 2014)
discharge (m3/s/m), d90 and d95 the rock
⎫ 0.97 1.35 ⎫ q ∙ sin θT
y w + ys = ⎪ ⎪
diameter representing the 90% and 95%

⎭√d90 √Hw ⎭1 – 0.175 cot θT

Mirskaulava (1967, as cited in
percentile respectively, θ the impingement – + 0.25ys Mason & Arumugam 1985)
angle with plunge pool surface, dm the

Journal of the South African Institution of Civil Engineering  Volume 62  Number 3  September 2020 37
median diameter, and β the air-water rela-
tionship (refer to Figure 1).
Van Schalkwyk et al (1994; 1995) proved
that the hydraulic erodibility of rock can be
characterised by the Kirsten index (Kirsten
1982) through studying rock scour forma-
tion of 42 spillways in South Africa and
the United States of America. Annandale
(2006) improved on the rock scour concepts
developed by Van Schalkwyk et al (1994;
1995) to develop a classification method
called the Erodibility Index Method (EIM).
The method compares the erosive capacity
of the water jet, to the erosive resistance of
the rock. The erosive capacity of the jet is
defined by stream power per unit area and
varies with depth in the plunge pool calcu-
lated with Equation 1.
γQHw
Ppool = (1)
Ai
where Ai is the jet footprint (impact area
of the jet) at different elevations below
the plunge pool water surface (m 2), γQH
is the stream power that stays the same at
impact with the plunge pool free surface, γ
is the unit weight of water (N/m 3), Q is the
total discharge (m 3/s), Hw is the fall height
(m), and Ppool is the stream power per unit
area in the pool at a certain depth (W/m 2).
The ability of the rock to resist scouring
is defined by the erodibility index K (see
Equation 2), which takes several geological
rock characteristics into account, such as
the mass strength of the rock, block size
and shape, the interparticle friction, and
the joint orientation relative to the flow
direction (Annandale 2006).
K = MS ∙ K b ∙ Kd ∙ JS(2)
Table 2 Erodibility index parameters (adapted from Annandale 2006)
Rock characteristics
Ms = Cr ∙ (0.78) ∙ (UCS)1.05 when UCS ≤ 10 MPa
Ms = Cr ∙ (UCS) when UCS > 10 MPa
Mass strength number (Ms )
Block size number (K b )
Discontinuity bond shear strength (Kd )
38
in which K is the erodibility index, MS is
the mass strength number, K b the block
size number, Kd the discontinuity bond
shear strength number, and JS the relative
ground structure number. Table 2 encapsu-
lates the formulae for the erodibility index
parameters.
The scour threshold for the rock
strength in terms of stream power per unit
area (kW/m2) is based on the erodibility
index K (Annandale 2006):
Prock = 0.48K 0.44 if K ≤ 0.1
Prock = K 0.75 if
K > 0.1
According to the EIM, the maximum scour
depth is reached when the available stream
power is equal to the resistive capacity of
the rock (Ppool ≤ Prock ).
The EIM does, however, have
some limitations, as the method only
determines the depth at the centreline of
the jet and cannot determine the extent
of the scour hole. The EIM provides a
generalised evaluation for rock scour,
as it does not explicitly consider the
mechanisms of block removal, brittle
fracture, or fatigue failure. Although
the geometric influence of rock joint
orientation is considered, it is handled in
a two-dimensional context, and therefore
does not capture the essence of block
removal in three-dimensional fractured
rock masses (Kieffer & Goodman 2012).
Pells (2016) argued that, to determine
the extent of the scour of unlined dam
spillways in rock, a gradation indicating
erosion regions should rather be used
than a threshold line. The erosion regions
presented by Pells (2016) compared well
Formulae
gρr
Cr =
γr
RQD
Kb =
Jn
RQD = (115 – 3.3Jc)
⎫ ⎫
Jc = ⎪ ⎪+3
3 Jx, Jy, Jz : Average spacing of joint sets
⎭(Jx ∙ Jy ∙ Jz) ⎭
0.33
Jr
Kd =
Ja
Volume 62  Number 3  September 2020  Journal of the South African Institution of Civil Engineering
with the independent data set presented
by Van Schalkwyk et al (1995).
Bollaert (2002) also investigated
rock scour due to falling jets. For this
investigation, near-prototype laboratory
experiments were carried out to reproduce
the dynamic pressure fluctuations at
the plunge pool bottom and inside the
rock joints. Bollaert (2002) developed the
Comprehensive Scour Model (CSM) to
predict the equilibrium scour depth for
both open-ended and closed joints. The
(3) CSM consists of three methods describing
the failure of rock, namely Comprehensive
(4) Fracture Mechanics (CFM), Dynamic
Impulsion method (DI), and Quasi-Steady
Impulsion method (QSI). The CFM
method assesses the instantaneous or time-
dependent joint propagation due to joint
water pressures (fatigue failure and brittle
fracturing) of closed-ended joints.
The DI method analyses sudden rock
block ejection due to uplift pressures on
individual rock blocks from the completely
jointed rock bed. The net impulse or maxi-
mum dynamic impulsion underneath a
rock block can be obtained by time integra-
tion of the net forces on the block (Bollaert
et al 2015):
∆t,pulse
I∆t,pulse = ∫0 (Fu – F0 – Gb – Fsh ) ∙ dt
= m ∙ V∆t,pulse(5)
in which Fu and F0 are the forces under
and over the block (N) respectively, Gb is
the immersed weight of the block (N), Fsh
represents the shear and interlocking forces
on the block (N), m is the mass of the block
(kg), V∆t,pulse is the average velocity experi-
enced by the rock block during time period
∆t (m/s), ∆t,pulse is the time interval of
Parameters
UCS : Unconfined compressive strength
Cr : Coefficient of relative density
ρr : Rock density
γ r : Unit weight of rock (27 ∙ 103 N/m3)
RQD values range between 5 and 100
Jn : Joint set number ranges between 1 and 5
Jc : Joint count number
Jr : Joint roughness number
Ja : Joint alteration number

Qtotal
Bg
Jet impingement
region
yup
Qup δ with the critical net uplift pressure (Pu)
Figure 2 Jet deflection on rock bed (adapted from Bollaert 2012)
certain pressure pulse(s) and I∆t,pulse is the
impulse on the rock block (Ns).
The first step in the DI method is to
define the maximum net impulsion (Imax)
as the product of a net force and a time
period. The corresponding pressure is
⎫V 2⎫
made non-dimensional by the jet’s kinetic
energy ⎪ ⎪. This results in a net uplift
⎭ 2g ⎭
pressure coefficient, Cup . The time period
is non-dimensionalised by the travel
⎫ L⎫
period that is characteristic for pressure
waves inside rock joints, i.e. Tup  =  ⎪2 ⎪,
⎭ c⎭
with L the joint length (L = xb + 2zb ), with
xb the width of the rock block and zb the
height of the rock block, and the mean
pressure wave celerity c taken as 100 m/s.
Hence, the non-dimensional impulsion
coefficient CI is defined by the product
V 2L
Cup ∙ Tup = j [m ∙ s]. The maximum
‌g ∙ c
net impulsion (Imax) is obtained by multi­
V 2L
plication of CI by j . Prototype-scaled
g∙c
analysis of uplift pressures resulted in the
following expression for CI (Bollaert et al
2015):
⎫Y⎫ ⎫Y⎫
CI = 0.0035 ∙ ⎪ ⎪ – 0.119⎪ ⎪ + 1.22(6)
2
⎭ Dj ⎭ ⎭ Dj ⎭
with Y the total plunge pool depth (yw + ys )
as defined in Figure 1, and Dj the jet
diameter at impact with the plunge pool
free surface.
The immersed weight of the rock block
is defined by Equation 7.
Journal of the South African Institution of Civil Engineering  Volume 62  Number 3  September 2020
Wall jet region
Qdown
ydown
Rock bed
Gb = xb2 ∙ zb (γr – γw)(7)
Failure of a block is expressed by the
displacement it undergoes due to the net
impulsion. This is obtained by transfor-
mation of V∆t,pulse in Equation 5 into a
net uplift displacement, hup defined by
Equation 8. As the block is ejected out of
the rock bed, the kinetic energy (velocity)
applied to the block is transformed into
potential energy.
V∆t,pulse2
hup = (8)
2g
in which hup is the height the rock block is
lifted (m).
If the height that the block is lifted (hup )
during the time period (∆t) is high enough,
the particle or rock block would be ejected
out of the matrix and become mobilised.
Tightly jointed rock, for example, would
require a vertical displacement close to its
height (hup /zb ≥ 1) while rock which is not
as tightly jointed, would require a lower
value of hup /zb .
The critical pressure for scour initiation
of a rock bed subjected to an impinging
jet could be used in a similar manner as
the DI method to determine when a rock
block would become dislodged from the
Table 3 Discharge distribution at impingement for different jet angles (Bollaert 2012)
Jet angle (δ) with rock bed 10°
Qup 1.5%
Qdown 98.5%
bed. This scour prediction method is
called the Critical Pressure Method in
this current study. A stable equilibrium
scour hole is achieved when the condition
Pactual < Pcritical is satisfied. The height a
block should be lifted, generally hup /zb ≥ 1,
for the particle to be displaced out of the
bed matrix and become mobilised is quan-
tified as follows (Annandale 2006):
⎫ L⎫
hup = ⎪2 ⎪ ∙
2
1
⎭ c ⎭ 2g ∙ xb ∙ zb2 ∙ ρr2
∙
4
[Pu ∙ xb2 – Gb – Fsh]2(9)
underneath a rock block with open-ended
joints being calculated with Equation 10
(Bollaert 2002):
Vj2
Pu = γCI (10)
2g
The QSI method (Bollaert 2012 calculates
the peeling of protruding rock blocks along
thin layers in the wall jet region parallel
to the plunge pool floor as illustrated in
Figure 2. The QSI method is dependent on
the quasi-steady high-velocity wall jets par-
allel to the pool floor and the subsequent
quasi-steady uplift forces on rock blocks,
which are caused by the protrusions of the
rock blocks. The rock blocks are said to
be plucked or peeled out of the bed matrix
due to the wall jet, when the quasi-steady
uplift force (Equation 11) can overcome
the immersed weight (Equations 7) of the
protruding rock block.
Vxi, max2
FQSL = Cuplift ∙ ρw ∙ Aexp ∙ (11)
2
In Equation 11 Cuplift is the net uplift
pressure coefficient, ρw is the density
of water, Aexp is the exposure area of a
rock block, and Vxi, max (Equation 14) is
the wall jet velocity x-distance from the
impingement point.
The deflection of the jet at the pool bot-
tom occurs in both the up and downstream
directions. The importance of each of these
deflections directly depends on the jet
impingement angle (δ) of the jet upon impact
with the rock bed, as set out in Table 3.
20° 30° 40° 90°
6.0% 7.0% 12.0% 50.0%
94.0% 93.0% 88.0% 50.0%
39
 The wall jet thicknesses yup and ydown Table 4 Physical model test conditions
are determined with Equations 12 and 13 Prototype value for limiting
respectively, with initial discharge Qtotal Item Model value
model scale of 1:20 scale
and jet thickness Bg impinging on a flatbed 3m 60 m (Hmin)
(Bollaert 2012). Drop height, H
4m 80 m (Hmed)
(relative to bedrock surface)
Qup yup 1 5m 100 m (Hmax)
= = ∙ (1 – cos δ)(12)
Qtotal Bg 2 0.224 m3/s 80 m3/s/m (Qmax)
Discharge, Q
0.13 m3/s 45 m3/s/m (Qmed)
(bmodel = 0.25 m)
Qup ydown 1
= = ∙ (1 + cos δ)(13) 0.1 m3/s 35 m3/s/m (Qmin)
Qtotal Bg 2
0.5 m 10 m (TWLmin)
Plunge pool depth, TWL
1m 20 m (TWLmax)
Once the jet deflects, the wall jets may be
characterised by their initial flow velocity Rock size 0.1 m × 0.1 m × 0.05 m 2.0 m × 2.0 m × 1.0 m (rock size 1)
VZbottom (Equation 15) and their initial x×y×z 0.1 m × 0.1 m × 0.075 m 2.0 m × 2.0 m x 1.5 m (rock size 2)
thickness yup or ydown at the point of deflec- vertical (90⁰)
tion. Initiating from this impingement loca-
Rock joint angle 45⁰ against the flow direction, illustrated in Figure 5
tion, the wall jets develop radially outwards
45⁰ in the flow direction (135⁰), illustrated in Figure 5
following self-preserving velocity profiles as
given by Equation 14 (Beltaos & Rajaratnam
1973 as cited in Bollaert et al 2015). Vxi, max
expresses the decay of the maximum 1 000 300 2 700 5 469

cross-sectional jet velocity with the relative

distance from the start of the wall jet (lateral
distance X divided by the initial thickness of
1 Issuance canal width
the deflected jet yup or ydown): = 250 mm

Vxi, max 3.5

= (14)
V Zbottom X
6 000

4 2
5 000

ydown TWL
4 000

2 300
The initial flow velocity at impingement
VZbottom (Equation 15) depends on the dif-
fusion angle (α in Figure 1) of the imping- Circular basin
3
ing jet and on its development length
through the plunge pool (yw + ys). VZbottom 5 800

continuously changes during scour forma-

tion (Bollaert et al 2015). The velocity Figure 3 M
 odel setup: (1) issuance canal, (2) plunge pool, (3) rectangular hand-packed
decay through the plunge pool would concrete blocks to simulate a uniform rock bed mass with regular open-ended joints,
increase for a greater jet diffusion angle (α) (4) scaffolding (adapted from a General Arrangement Drawing by PERI Formwork
Scaffolding Engineering (Pty) Ltd)
VZbottom Zcore
= (for rectangular jets)(15)
Vj yw + ys

with Zcore the distance required for the jet

core to diffuse in the plunge pool depth
(m), which is generally taken as four times
the jet thickness (m) at impact (Bg).

Experimental Work

Model scale
Bollaert and Schleiss (2005) determined
that the pressures inside the rock joints are
accurately reproduced for a physical model
scale of approximately 1:10. Boushaba et al
(2013) found that the scale effects on the
plunging jet velocity, initial turbulence,
plunge pool aeration and the dynamic Figure 4 Photograph of experimental setup for lowest spillway height

40 Volume 62  Number 3  September 2020  Journal of the South African Institution of Civil Engineering
 45°
Dipping angle of “horizontal”
joints against flow direction
Figure 5 Dipped joint structure orientation
pressures at the bedrock surface are negli-
gible for a physical model scale not exceed-
ing 1:20. Model scales greater than 1:100
could cause incorrect aeration results.
Therefore, a model scale not exceeding
1:20 is recommended for the experimental
results from this study. Additionally, with
the available laboratory height, a 1:20
model scale is able to replicate realistic
prototype spillway heights (i.e. 60 m up
to 100 m).
Model layout
The experimental investigation was
conducted in the hydraulics laboratory
of the Civil Engineering Department at
Stellenbosch University, South Africa.
The three-dimensional physical model
(based on Froude scale laws) consisted of
a rectangular horizontal issuance canal,
replicating an uncontrolled spillway. Flow
straighteners were installed in the issuance
(a)
Figure 6 Bed deformation process (viewed from upstream)
Journal of the South African Institution of Civil Engineering  Volume 62  Number 3  September 2020
Horizontal and vertical joints
canal to ensure that uniform, fully devel-
oped flow would be reached. The issuance
canal could be adjusted to three different
fixed heights above the movable rock bed:
3 m, 4 m and 5 m (prototype heights for
1:20 model scale: 60 m, 80 m and 100 m
respectively). The tailwater depth at the
plunge pool was also adjustable, namely
0.5 m and 1 m (prototype: 10 m and 20 m
respectively). Plunging jets were evalu-
ated for unit discharges of 0.10 m3/s/m to
0.224 m3/s/m (prototype: 35 m3/s/m to
80 m3/s/m). The broken-up rock bed mass
was modelled by using tightly hand-packed
concrete paver blocks (cobblestones)
emulating a uniform three-dimensional
open-ended horizontal and vertical rock
joint network. The two rock sizes tested
were rectangular concrete blocks with x,
y, z dimensions of 0.1 m × 0.1 m × 0.05 m
and 0.1 m × 0.1 m × 0.075 m (prototype:
2 m × 2 m × 1 m and 2 m × 2m × 1.5 m
(b)
45°
Dipping angle of “horizontal”
joints in flow direction
respectively). A similar prototype rock
density of 2 650 kg/‌m3 should be used in
the model since gravity cannot be scaled.
The paver block densities used were
2 355.4 kg/‌m3 and 2 388.1 kg/m3 respec-
tively. The test conditions are summarised
in Table 4.
A total of 31 experimental tests were
carried out. Ten per cent of the experi-
ments were repeated to ensure repeatability
of the results, and showed a maximum
deviation of 15%. Figure 3 shows the
laboratory model setup (dimensions shown
in Figure 3 are model dimensions in mm).
Figure 4 shows a photograph of the experi-
mental setup. The three joint structures
investigated are illustrated in Figure 5.
During each of the experimental tests,
the scouring of rock by the free-falling
rectangular jet that exited the spillway was
monitored until scouring in the plunge
pool stopped and equilibrium conditions
(c)
41
were reached (after 6 to 8 hours of testing),
called Case A. The bed profile of the equi- Bed level
(m)
librium scour hole was then surveyed. The –0.5
1.0 –0.6
same test was then repeated, but instead –0.7
of leaving the scoured blocks from the 0.5 –0.8
–0.9
pool, the deposited rocks downstream of –1.0

Width
Flow direction –1.1
the scour hole were continuously removed 0
–1.2
until equilibrium conditions were reached –1.3
–0.5 –1.4
again (called Case B) after approximately –1.5
–1.6
16 to 20 hours, where after the so-formed –1.0 –1.7
equilibrium bed profile was surveyed again. –1.8
–1.9
Figure 6 (p 41) shows the bed deformation –3.0 –2.5 –2.0 –1.5 –1.0 –0.5 0 0.5 1.0 1.5 2.0 2.5 –2.0
(a) Length –2.1
process for the horizontal and vertical –2.2
aligned joint structures. The topography of
the bed profile was surveyed using a three-
Bed level
dimensional laser scanner, Z+F Imager (m)
5006h, at a high resolution of 10 000 1.0 –0.5
–0.6
pixels/360°. –0.7
0.5 –0.8
–0.9
Results of experimental work –1.0
Flow direction
Width

Although various flow rates, drop heights,
and tailwater levels were tested, only
results of selected tests are presented in
this paper since all the tests followed simi-
lar trends.
Figure 7 shows the contour maps of
the surveyed scour holes of the model
for Cases A and B respectively for the
maximum discharge, highest drop height,
and deepest plunge pool. It can be seen
from the contour maps that the scour hole
depth, length, width and volume increase
when the deposited rocks downstream of
the hole are removed, and the horizontal
distance from the spillway to the onset of
scour decreases.
The longitudinal and lateral bed profiles
for the model for the various flow rates,
listed in Table 4 (35 m3/s/m, 45 m3/s/m
and 80 m3/s/m), for the highest fall height
(Hmax = 100 m) and the deepest plunge
pool depth (TWLmax = 20 m) are shown
in Figure 8 for Case A (scour and deposi-
tion) and Case B (scour with deposited
rock removed). Figure 8 illustrates that the
scour depth, width and length increase as
the discharge increases. The scour results
for Case B presented in Figure 8 also indi-
cate that greater scour could occur if the
deposited rocks downstream of the scour
hole are removed by floodwaters. Test 2 in
Figure 8 is a repeated test of Test 1.
Figure 9 indicates that the scour depth,
width and length for the three different
drop heights (60 m, 80 m and 100 m) were
of similar magnitude. A possible reason
for the low sensitivity towards the drop
height could be that the jet was already
fully developed (the jet core is fully dis-
sipated) at the impingement with the
0 –1.1
–1.2
–1.3
–0.5 –1.4
–1.5
–1.0 –1.6
–1.7
–1.8
–1.9
–3.0 –2.5 –2.0 –1.5 –1.0 –0.5 0 0.5 1.0 1.5 2.0 2.5 –2.0
(b) Length –2.1
–2.2
 bserved final bed level for Q = 80 m3/s/m, H = 100 m, TWL = 20 m; (a) Case A (scour and
Figure 7 O
deposition) and (b) Case B (scour with deposited rock removed)
plunge pool water surface for all three to scour failure when the joint structure
drop heights. is dipped 45° in the direction of the flow.
Figure 10 shows that the plunge pool The experimental results (Table 6) indicate
acts as a water cushion, since the scouring that the rock with a joint structure dipped
depth, width and length decrease with an 45° against the flow direction has a greater
increase in the plunge pool depth, in accor- scour resistance to the erosive capacity of
dance with Van Aswegen et al (2001) and the water.
Bollaert et al (2012).
The bed profiles depicted in Figure 11
indicate that the scour depth, width and Development of FORMULAE
length decrease as the rock block size based on experimental
increases. Larger rock blocks provide great- data OF THIS STUDY
er erosive resistance than smaller blocks, A multi-linear regression analysis was
due to its submerged weight, in accordance done on the physical model results to find
with Annandale (2006). non-dimensional formulae that define
The geologic structure of the rock can the scour hole geometry (depth, length,
significantly influence the erodibility of width and volume). Three regression
the rock. The effect of the rock joint angle models were analysed, namely Linear,
relative to the flow direction was therefore Logarithmic Transformed and Linear
investigated by means of Figure 12. The Logarithmic, represented by Equations 16
three joint angles under investigation to 18 respectively.
were 45° dipped against the direction of
the flow, vertical (90° to the horizontal Q 0 = k + a ∙ x1 + b ∙ x2 + c ∙ x3 + … (16)
axis) and 45° in the direction of the flow
(thus 135°). ln(Q 0 ) = ln(k) + a ∙ ln(x1) + b ∙ ln(x2) +
Similar to the well-known rock scour c ∙ ln(x3) + … (17)
case of Ricobayo Dam in Spain (Annandale
2006), the scour results shown in Figure 12 Q 0 = k + a ∙ ln(x1) + b ∙ ln(x2) +
indicate that the rock is more conducive c ∙ ln(x3) + … (18)

42 Volume 62  Number 3  September 2020  Journal of the South African Institution of Civil Engineering
 15 2
0
10
–2
Elevation (m)

Elevation (m)
5 –4
0 –6
–8
–5 –10
–10 –12
–15 –14
0 20 40 60 80 100 120 –30 –20 –10 0 10 20 30
Length (m) Width (m)
Qmin-Test 1 (35 m3/s/m) Qmed (45 m3/s/m) Qmin-Test 1 (35 m3/s/m) Qmed (45 m3/s/m)
Qmin-Test 2 (35 m3/s/m) Qmax (80 m3/s/m) Qmin-Test 2 (35 m3/s/m) Qmax (80 m3/s/m)
Longitudinal cross-section: Case A – distance from spillway Lateral cross-section: Case A – distance from issuance centreline

15 2
0
10
–2
Elevation (m)

Elevation (m)
5
–4
0 –6
–8
–5
–10
–10
–12
–15 –14
0 20 40 60 80 100 120 –30 –20 –10 0 10 20 30
Length (m) Width (m)
Qmin-Test 1 (35 m3/s/m) Qmed (45 m3/s/m) Qmin-Test 1 (35 m3/s/m) Qmed (45 m3/s/m)
Qmin-Test 2 (35 m3/s/m) Qmax (80 m3/s/m) Qmin-Test 2 (35 m3/s/m) Qmax (80 m3/s/m)
Longitudinal cross-section: Case B – distance from spillway Lateral cross-section: Case B – distance from issuance centreline

Figure 8 Observed prototype bed profile indicating the effect of discharge (H = 100 m, TWL = 20 m)

15 4
2
10
0
Elevation (m)

Elevation (m)

5 –2
–4
0
–6
–5 –8
–10
–10
–12
–15 –14
0 20 40 60 80 100 120 –30 –20 –10 0 10 20 30
Length (m) Width (m)
Hmin (60 m) Hmed (80 m) Hmax (100 m) Hmin (60 m) Hmed (80 m) Hmax (100 m)
Longitudinal cross-section: Case A – distance from spillway Lateral cross-section: Case A – distance from issuance centreline

15 4
2
10
0
Elevation (m)

Elevation (m)

5 –2
–4
0
–6
–5 –8
–10
–10
–12
–15 –14
0 20 40 60 80 100 120 –30 –20 –10 0 10 20 30
Length (m) Width (m)
Hmin (60 m) Hmed (80 m) Hmax (100 m) Hmin (60 m) Hmed (80 m) Hmax (100 m)
Longitudinal cross-section: Case B – distance from spillway Lateral cross-section: Case B – distance from issuance centreline

Figure 9 Observed bed profile indicating the effect of drop height (Q = 80 m3/s/m, TWL = 20 m)

Journal of the South African Institution of Civil Engineering  Volume 62  Number 3  September 2020 43
 25 25
20 20
15 15
Elevation (m)

Elevation (m)
10
10
5
5
0
0
–5
–10 –5
–15 –10
–20 –15
0 20 40 60 80 100 120 –30 –20 –10 0 10 20 30
Length (m) Width (m)
Bed level for TWLmax TWLmin (10 m) Bed level for TWLmax TWLmin (10 m)
Bed level for TWLmin TWLmax (20 m) Bed level for TWLmin TWLmax (20 m)
Longitudinal cross-section: Case A – distance from spillway Lateral cross-section: Case A – distance from issuance centreline

25 25
20 20
15 15
Elevation (m)

Elevation (m)
10 10
5 5
0 0
–5 –5
–10 –10
–15 –15
–20 –20
0 20 40 60 80 100 120 –30 –20 –10 0 10 20 30
Length (m) Width (m)
Bed level for TWLmax TWLmin (10 m) Bed level for TWLmax TWLmin (10 m)
Bed level for TWLmin TWLmax (20 m) Bed level for TWLmin TWLmax (20 m)
Longitudinal cross-section: Case B – dstance from spillway Lateral cross-section: Case B – distance from issuance centreline

Figure 10 Observed bed profile indicating effect of the plunge pool depth (Q = 80 m3/s/m, H = 100 m)

15 2
10 0
5 –2
Elevation (m)

Elevation (m)

–4
0 –6
–5 –8
–10
–10
–12
–15 –14
0 20 40 60 80 100 120 –30 –20 –10 0 10 20 30
Length (m) Width (m)
Rock size 1 Rock size 2 – Test 1 Rock size 2 – Test 2 Rock size 1 Rock size 2 – Test 1 Rock size 2 – Test 2
Longitudinal cross-section: Case A – distance from spillway Lateral cross-section: Case A – distance from issuance centreline

15 2
0
10
–2
Elevation (m)

Elevation (m)

5
–4
0 –6
–8
–5
–10
–10
–12
–15 –14
0 20 40 60 80 100 120 –30 –20 –10 0 10 20 30
Length (m) Width (m)
Rock size 1 Rock size 2 – Test 1 Rock size 2 – Test 2 Rock size 1 Rock size 2 – Test 1 Rock size 2 – Test 2
Longitudinal cross-section: Case B – dstance from spillway Lateral cross-section: Case B – distance from issuance centreline

Figure 11 Observed bed profile indicating effect of rock block size (Q = 80 m3/s/m, H = 100 m, TWL = 20 m)

44 Volume 62  Number 3  September 2020  Journal of the South African Institution of Civil Engineering
 10 4
8
6 2
4
Elevation (m)

Elevation (m)
0
2
0 –2
–2
–4 –4
–6 –6
–8
–10 –8
0 20 40 60 80 100 120 –30 –20 –10 0 10 20 30
Length (m) Width (m)
90° joint angle 45° against flow direction 90° joint angle 45° against flow direction
45° in flow direction 45° in flow direction
Longitudinal cross-section: Case A – distance from spillway Lateral cross-section: Case A – distance from issuance centreline

10 4
8
6 2
4
Elevation (m)

Elevation (m)
0
2
0 –2
–2
–4 –4
–6 –6
–8
–10 –8
0 20 40 60 80 100 120 –30 –20 –10 0 10 20 30
Length (m) Width (m)
90° joint angle 45° against flow direction 90° joint angle 45° against flow direction
45° in flow direction 45° in flow direction
Longitudinal cross-section: Case B – dstance from spillway Lateral cross-section: Case B – distance from issuance centreline

Figure 12 Observed equilibrium bed profile indicating effect of the rock joint angle (Q = 80 m3/s/m, H = 100 m, TWL = 20 m)

where Q 0 is the dependent variable, the
independent variable, x1,2,3 … and k, a, b,
and c are constants.
Several parameters were incorporated
in the regression analysis, such as rock
diameter, discharge, drop height, plunge
pool depth, stream power, and uplift pres-
sure. The main parameters influencing the
flow and rock scour were selected based
on the extensive literature study conducted
and experimental tests.
Instead of incorporating bed shear
stress explicitly, the movability number
(Equation 19) based on the settling velo­city
(Vss) and the particle Reynolds number
with v the kinematic viscosity of water =
1.13 × 10 –6 m²/s at 15˚C, and V*, the shear
velocity, is defined as:
V* = √g ∙ ydown ∙ Sf(21)
with ydown denoting the jet thickness
downstream of the impingement region as
illustrated in Figure 2. The energy slope
Sf required for determining the shear
velocity was determined from Chèzy
(Equation 22) by assuming the hydraulic
Awall jet
radius R =  (area of jet deflected
Wj rock
downstream of impingement/jet width at
⎫12 × R⎫
Qdown = Awall jet × 18log⎪ ⎪√R ∙ Sf(22)
⎭ d50 ⎭
with Qdown the respective discharge down-
stream of the impingement region, based
on the total flow and jet impinging angle, δ
(Table 3), R the hydraulic radius, and Sf the
energy slope. The wall jet flow area is defined
as Awall jet = ydown × Wj rock, with Wj rock being
the width of the jet at impingement with
the rock bed, taking the jet outer spreading
angle as 15° (Ervine & Falvey 1987) for highly
turbulent jets. The median particle size (d50)
is defined by Equation 23 (Pabst & Gregorova
2007) with Vblock the rock block volume (m3):

⎫6 ⎫
(Equation 20) were used in analysing the impingement with the bedrock as illus-
d50 = ⎪ Vblock⎪
1/3

⎭π ⎭
rock scour. The movability number and trated in Figure 2). The surface roughness
(23)
particle Reynolds number Rep (Armitage was approximated as ks = d50 median
& Rooseboom 2010) have the benefit that particle size (Equation 23). The settling
they are based on the shear velocity V* velocity was elected as the representative The parameters describing the scour hole
(Equation 21) the particles are experiencing of the transportability of large aggregate, geometry were calculated using ordinary
due to the plunging jet, and the particle size. since it accounts for size, shape and least squares regressions, with due consid-
density of the rock blocks. A settling tank eration given to the parsimony principle.
V* was used in the laboratory to determine Due to the availability of many possible
MN = (19)
Vss that the blocks have a median prototype regressors, parsimony is emphasised to
settling velocity (VSS ) of 4.19 m/s and ensure avoiding over-fitting. The parameters
V*d50 4.99 m/s for the small and large rock were made dimensionless by applying the
Rep = (20)
v blocks respectively. Buckingham theorem (Albrecht et al 2013).

Journal of the South African Institution of Civil Engineering  Volume 62  Number 3  September 2020 45
  egression formulae for equilibrium scour hole geometry for horizontal and vertical open joints*
Table 5 R
Equation R2

⎫ Ds ⎫ ⎫d ⎫ ⎫ V ⎫ ⎫H ⎫ ⎫V ⎫ ⎫F ⎫ ⎫y ⎫
ln⎪ CaseA ⎪ = –
 468.532ln⎪ 50 ⎪ – 322.102ln⎪ i ⎪ + 4.502ln⎪ w⎪ – 483.385ln⎪ SS ⎪ – 0.804ln⎪ lift ⎪ – 0.817ln⎪ w ⎪
⎭ Hw ⎭ ⎭ Hw ⎭ ⎭√gDi⎭ ⎭ Lb ⎭ ⎭ Vi ⎭ ⎭ Gb ⎭ ⎭ Hw ⎭
⎫V D ⎫ ⎫ q2 ⎫ ⎫ ⎫ ⎫ P ⎫ ⎫ V* ⎫
– 473.991ln⎪ i i ⎪ + 158.264ln⎪ ⎪ – 0.419ln⎪ ⎪ + 0.006ln⎪ ⎪ – 473.536ln⎪ ⎪
PSP
⎭ v ⎭ ⎭ gHw ⎭ ⎭Puplift ∙ VSS ⎭ ⎭Pcrit⎭ ⎭VSS ⎭
Depth
0.998
Case A 3

⎫V*d50⎫
+ 474.587ln⎪ ⎪
⎭ v ⎭

⎫d ⎫ ⎫ V ⎫ ⎫V ⎫ ⎫F ⎫ ⎫ ⎫ ⎫ yw ⎫
 0.196ln⎪ 50 ⎪ – 0.578ln⎪ i ⎪ – 0.942ln⎪ SS ⎪ – 0.216ln⎪ lift ⎪ – 0.129ln⎪ ⎪ – 0.079ln⎪ ⎪
DsCaseB PSP
⎭ Hw ⎭ ⎭√gDi⎭ ⎭ Vi ⎭ ⎭ Gb ⎭ ⎭ Puplift ∙ Vj ⎭ ⎭ Hw ⎭
=–
Hw
⎫V D ⎫ ⎫ q2 ⎫ ⎫ P ⎫ ⎫ V* ⎫ ⎫V*d50⎫
Depth

– 0.279ln⎪ i i ⎪ + 0.129ln⎪ ⎪ + 0.006ln⎪ ⎪ – 0.145ln⎪ ⎪ + 0.315ln⎪ ⎪

0.992
Case B

⎭ v ⎭ ⎭ gHw ⎭
3 ⎭Pcrit⎭ ⎭VSS ⎭ ⎭ v ⎭

⎫d ⎫ ⎫ V ⎫ ⎫H ⎫ ⎫V ⎫ ⎫F ⎫ ⎫y ⎫
= – 416.262ln⎪ 50 ⎪ – 281.275ln⎪ i ⎪ + 2.804ln⎪ w⎪ – 421.760ln⎪ SS ⎪ – 0.666ln⎪ lift ⎪ – 0.78ln⎪ w ⎪
LsCaseA
Hw ⎭ Hw ⎭ ⎭√gDi⎭ ⎭ Lb ⎭ ⎭ Vi ⎭ ⎭ Gb ⎭ ⎭ Hw ⎭
⎫V D ⎫ ⎫ q2 ⎫ ⎫ ⎫ ⎫ P ⎫ ⎫ V* ⎫
– 418.577ln⎪ i i ⎪ + 140.444ln⎪ ⎪ – 0.157ln⎪ ⎪ – 0.076ln⎪ ⎪ – 417.815ln⎪ ⎪
PSP
⎭ v ⎭ ⎭ gHw3 ⎭ ⎭Puplift ∙ VSS ⎭ ⎭Pcrit⎭ ⎭VSS ⎭
Length
0.987
Case A

⎫V*d50⎫
+ 418.985ln⎪ ⎪
⎭ v ⎭

⎫ d ⎫3.048 ⎫ Vi ⎫–0.1878 ⎫Hw⎫1.427 ⎫ VSS ⎫–3.052 ⎫ Flift ⎫0.485 ⎫ yw ⎫–0.081 ⎫ Vi Di ⎫0.639 ⎫ q 2 ⎫0.002
= ⎪ 50 ⎪ ∙  ⎪ ⎪ ∙  ⎪ ⎪ ∙  ⎪ ⎪ ∙  ⎪ ⎪ ∙  ⎪ ⎪ ∙  ⎪ ⎪ ∙  ⎪ ⎪
LsCaseB
Hw ⎭ Hw ⎭ ⎭√gDi⎭ ⎭ Lb ⎭ ⎭ Vi ⎭ ⎭ Gb ⎭ ⎭ Hw ⎭ ⎭ v ⎭ ⎭ gHw3 ⎭
⎫ ⎫0.019 ⎫ P ⎫–0.001 ⎫ V* ⎫0.872 ⎫V*d50⎫–0.377
Length

∙  ⎪ ⎪ ∙  ⎪ ⎪ ∙  ⎪ ⎪ ∙  ⎪ ⎪
0.984
Case B PSP
⎭Puplift ∙ VSS⎭ ⎭Pcrit⎭ ⎭VSS ⎭ ⎭ v ⎭

⎫ Ws ⎫ ⎫d ⎫ ⎫ V ⎫ ⎫H ⎫ ⎫V ⎫ ⎫F ⎫ ⎫y ⎫
ln⎪ CaseA ⎪ = –
 35.453⎪ 50 ⎪ – 28.412ln⎪ i ⎪ + 2.079ln⎪ w⎪ – 43.005ln⎪ SS ⎪ – 0.879ln⎪ lift ⎪ + 0.102ln⎪ w ⎪
⎭ Hw ⎭ ⎭ Hw ⎭ ⎭√gDi⎭ ⎭ Lb ⎭ ⎭ Vi ⎭ ⎭ Gb ⎭ ⎭ Hw ⎭
⎫V D ⎫ ⎫ q2 ⎫ ⎫ ⎫ ⎫ P ⎫ ⎫ V* ⎫
– 37.850ln⎪ i i ⎪ + 12.705ln⎪ ⎪ – 0.229ln⎪ ⎪ + 0.079ln⎪ ⎪ – 37.585ln⎪ ⎪
PSP
⎭ v ⎭ ⎭ gHw ⎭ ⎭Puplift ∙ VSS ⎭ ⎭Pcrit⎭ ⎭VSS ⎭
Width
0.998
Case A 3

⎫V*d50⎫
+ 38.229ln⎪ ⎪
⎭ v ⎭

⎫ d ⎫–2.228 ⎫ Vi ⎫4.390 ⎫Hw⎫–1.844 ⎫ VSS ⎫6.531 ⎫ Flift ⎫0.853 ⎫ yw ⎫0.048 ⎫ Vi Di ⎫0.466 ⎫ q 2 ⎫0.614
= ⎪ 50 ⎪ ∙  ⎪ ⎪ ∙  ⎪ ⎪ ∙  ⎪ ⎪ ∙  ⎪ ⎪ ∙  ⎪ ⎪ ∙  ⎪ ⎪ ∙  ⎪ ⎪
WsCaseB
Hw ⎭ Hw ⎭ ⎭√gDi⎭ ⎭ Lb ⎭ ⎭ Vi ⎭ ⎭ Gb ⎭ ⎭ Hw ⎭ ⎭ v ⎭ ⎭ gHw3 ⎭
⎫ ⎫0.379 ⎫ P ⎫0.021 ⎫ V* ⎫0.393 ⎫V*d50⎫–0.565
Width

∙  ⎪ ⎪ ∙  ⎪ ⎪ ∙  ⎪ ⎪ ∙  ⎪ ⎪
0.996
Case B PSP
⎭Puplift ∙ VSS⎭ ⎭Pcrit⎭ ⎭VSS ⎭ ⎭ v ⎭

⎫Vol ⎫ ⎫d ⎫ ⎫ V ⎫ ⎫H ⎫ ⎫V ⎫ ⎫F ⎫ ⎫y ⎫
ln⎪ CaseA⎪ = –
 131.752ln⎪ 50 ⎪ – 106.469ln⎪ i ⎪ + 6.433ln⎪ w⎪ – 159.640ln⎪ SS ⎪ + 1.567ln⎪ lift ⎪ – 0.724ln⎪ w ⎪
⎭ Hw ⎭
3 ⎭ Hw ⎭ ⎭√gDi⎭ ⎭ Lb ⎭ ⎭ Vi ⎭ ⎭ Gb ⎭ ⎭ Hw ⎭
⎫V D ⎫ ⎫ q2 ⎫ ⎫ ⎫ ⎫ P ⎫ ⎫ V* ⎫
– 145.502ln⎪ i i ⎪ + 47.595ln⎪ ⎪ – 0.012ln⎪ ⎪ – 0134ln⎪ ⎪ – 146.724ln⎪ ⎪
Scour PSP
⎭ v ⎭ ⎭ gHw ⎭ ⎭Puplift ∙ VSS ⎭ ⎭Pcrit⎭ ⎭VSS ⎭
volume 0.999
3

⎫V*d50⎫
Case A

+ 146.313ln⎪ ⎪
⎭ v ⎭

⎫ d ⎫3.061 ⎫ Vi ⎫–5.062 ⎫Hw⎫2.519 ⎫ VSS ⎫–8.050 ⎫ Flift ⎫1.330 ⎫ yw ⎫–0.470 ⎫ Vi Di ⎫–2.522 ⎫ q 2 ⎫1.238
= ⎪ 50 ⎪ ∙  ⎪ ⎪ ∙  ⎪ ⎪ ∙  ⎪ ⎪ ∙  ⎪ ⎪ ∙  ⎪ ⎪ ∙  ⎪ ⎪ ∙  ⎪ ⎪
VolCaseB
Scour Hw3 ⎭ Hw ⎭ ⎭√gDi⎭ ⎭ Lb ⎭ ⎭ Vi ⎭ ⎭ Gb ⎭ ⎭ Hw ⎭ ⎭ v ⎭ ⎭ gHw3 ⎭
⎫ ⎫0.276 ⎫ P ⎫0.010 ⎫ V* ⎫–2.566 ⎫V*d50⎫2.966
volume
∙  ⎪ ⎪ ∙  ⎪ ⎪ ∙  ⎪ ⎪ ∙  ⎪ ⎪
0.998
PSP
⎭Puplift ∙ VSS⎭ ⎭Pcrit⎭ ⎭VSS ⎭ ⎭ v ⎭
Case B

* Ranges: Discharge: 35 m3/s/m – 80 m3/s/m; Drop height: 60 m – 100 m; Tailwater depth: 10 m – 20 m; horizontal and vertical joint network

The Logarithmic model provided the
best fit for the scour hole dimensions
(scour depth, length, width and volume),
except the scour hole depth for Case B was
predicted best by the Linear Log model.
Table 5 summarises the non-dimensional
formulae for vertical rock joints obtained
from the regression analysis. To apply
these regression formulae, the equilibrium
scour depth should first be determined
by a trial-and-error procedure using the
scour formula in Table 5, initially for a
jet impinging onto the original flat bed.
Iteration ceases when a reasonable per-
centage difference is found. Thereafter the

46 Volume 62  Number 3  September 2020  Journal of the South African Institution of Civil Engineering
scour parameters (i.e. movability number Table 6 Stability factors due to joint angle
and dynamic pressure head) at the equi- Dipping angle of “horizontal” Dipping angle of “horizontal”
librium scour depth can be used in order Scour hole geometry
joints against flow direction joints in flow direction
to obtain the other scour hole geometries
Case A 0.93 1.29
(length, width and volume) by using the Depth
formulae presented in Table 5. The regres- Case B 0.89 1.06
sion formulae should be applied inside the Case A 0.98 0.77
prototype ranges in which the formulae Length Q < 45 m3/s/m: Q > 45 m3/s/m:
were developed (discharge: 35 m 3/s/m – Case B 0.82
0.42 0.89
80 m 3/s/m; drop height: 60 m – 100 m;
tailwater depth: 10 m – 20 m; horizontal Case A 1.01 1.28
Width
and vertical joint network). Case B 0.9 1.13
The stabilising (or destabilising) fac-
Q < 45 m3/s/m: Q > 45 m3/s/m:
tors for joint structures orientated 45° in Case A 0.83
1.22 0.48
and against the direction of the flow are Volume
Q < 45 m3/s/m: Q > 45 m3/s/m:
summarised in Table 6. The joint struc- Case B 0.82
1.09 0.53
ture angle factors should be applied to the
predicted scour geometry results obtained Ranges: Discharge: 35 m3/s/m – 80 m3/s/m; Drop height: 60 m – 100 m;
Tailwater depth: 10 m – 20 m; 45° and 135° joints
for the horizontal and vertical joint
formulae in Table 5. The variations in the
stabilising factors are small. However, results. The regression analysis results indi- the scour depth 14.3%, for the scour length
additional experimental testing is required cate that a good correlation was achieved 22.5%, for the scour width 8.6%, and for the
to ascertain whether linear interpolation between the predicted and observed values. scour volume 14.4%.
is possible between the stabilising factors The regression formulae were confirmed
for different degrees of joint structures inside the ranges in which the developed
between 0° and 45° in and against the formulae were based by using experimental Comparison of physical
flow direction. tests that did not form part of the regres- model results with scour
The regression analysis results for sion analysis but with parameter values prediction methods
the scour hole parameters (depth, length, inside the ranges of the tests. The devel- fROm literature
width and volume) for the Case B sce- oped formulae have a good agreement with Eleven empirical formulae (Table 1), and the
nario (deposited rocks removed) shown in the observed confirmation values with the EIM and CSM traditionally employed to
Figure 13 with the 95% confidence band following percentage errors (i.e. percentage predict rock scour due to plunging jets were
indicates the correlation between the for- differences between test values and those evaluated against the results from the physi-
mulae predicted and the confirmation test predicted by the developed formulae) – for cal model study for a full-scale prototype.

(a) Scour depth (b) Scour hole length

0.25 R2 = 0.992 0.5 R2 = 0.984
predicted

0.20
predicted

0
0.15
–0.5
Hw
Ls
Hw
Ds

0.10
ln

–1.0
0.05

0.05 0.10 0.15 0.20 0.25 –1.0 –0.5 0 0.5

Ds Ls
ln
Hw observed Hw observed

(c) Scour hole width (d) Scour hole volume

–2
R2 = 0.996 R2 = 0.998
–0.4
predicted
predicted

–3
–0.8 –4
Hw3
Vol
Ws
Hw

–1.2 –5
ln

ln

–6
–1.6
–1.2 –0.8 –0.4 –6 –5 –4 –3 –2
Ws Vol
ln ln
Hw observed Hw3 observed

Figure 13 Regression analysis outcome for Case B scenario with 95% confidence band

Journal of the South African Institution of Civil Engineering  Volume 62  Number 3  September 2020 47

Uniform open-ended
joint rock bed
Figure 14 Prototype equilibrium scour depth from physical model and literature scour prediction
methods for Q = 45 m3/s/m, H = 100 m, TWL = 20 m
Only the DI and QSI methods of the CSM,
and not the CFM method, were compared
with the physical model test results, since
the CFM method analyses fatigue failure
and brittle fracturing of closed-ended joints
and the model’s rock mass comprised a fully
formed joint network.
The prototype scour hole depth formed
in a movable bed based on the physical
model study is compared in Figure 14 to
the scour prediction methods found in
literature for the 45 m3/s/m flood, a drop
height of 80 m and a plunge pool depth of
20 m.
The scour depth prediction methods
from literature were found to yield a wide
range of varying results for the same
input conditions as shown in Figure 14.
Figure 15 is a correlation plot of scour
depths observed in the experimental tests
versus the scour depths predicted by
15 selected formulae from literature by
applying prototype Case B physical model
test conditions as input in the respective
formulae and methods. Figure 16 presents
the statistical spread of the data in Figure
15. The percentage difference between the
observed prototype depth from the experi-
mental tests and the depths obtained from
the respective literature methods is given
dspredicted – dsobserved
.
by dsobserved × 100
48
Damle (1966): 2.1 m
Martins (1975): 2.9 m
Physical model of this study (Case A): 4.2 m
Physical model of this study (Case B): 7.8 m
Vernonese 2 (1984): 7.9 m; Critical Pressure: 8.0 m
Mason (1985): 9.6 m
EIM (2006): 10.5 m
Yildiz (1994): 16.9 m
QSI (2012): 17.5 m
Chee (1974): 19.5 m
Katoulas (1967); Veronese 3 (1984): 20.1 m
DI (2002): 30.0 m
Veronese 1 (1984): 38.5 m
The percentage difference results
shown in Figure 16 indicate that the
methods are in weak agreement with
one another and generally overestimate
the observed scour depths with a mean
difference of 99% (over-predict) for the
Case B scenario. The analytical method
that agrees best with the experimental
results is the Critical Pressure method
(Annandale 2006), since the median per-
centage difference is close to zero (-25.2%).
However, the safest scour prediction
method (accurate and conservative) would
be the EIM (Annandale 2006), followed
by Mason and Arumugam’s (1985) empiri-
cal formula, since the first quartile (25th
percentile) nears the zero percentage dif-
ference (-12% and 15% respectively) with a
narrow statistical spread.
The Critical Pressure method
(Annandale 2006) proves to agree the clos-
est with the laboratory data, presumably
because it relies on the critical pressure for
scour initiation of a rock bed subjected to
an impinging jet. The critical pressure is
dependent on the submerged weight of the
rock block, block dimensions, uplift pres-
sures inside the joints, diffusion of the jet
through the air and plunge pool, as well as
the velocity of the jet and drop height.
The EIM presumably performed well
because it relies on rock material properties
Volume 62  Number 3  September 2020  Journal of the South African Institution of Civil Engineering
and the stream power of the jet as it travels
through the air and plunge pool. Figure 17
indicates that the EIM erosion threshold is
compatible with the laboratory experimen-
tal data for the Case B scenario when the
scoured blocks were continually removed,
since the data plots above the threshold
line in the “scour” region (green markers).
Mason and Arumugam (1985) proved
to be the only identified empirical formula
that incorporates the plunge pool depth
with gravity acceleration. The implication
of this is that models taking the plunge
pool depth, and consequently the dis-
sipation of the jet’s energy through the
plunge pool, into consideration could
offer better scour depth predictions. The
experimental test results reiterate the
experimental finding that the scour depth
greatly depends on the plunge pool depth.
The Jaeger equation (1939, as cited in
Castillo & Carrillo 2014) also incorporated
the plunge pool depth and had a median
percentage difference close to zero (25%);
however, it had a wide percentage differ-
ence statistical spread.
Yildiz and Uzucek (1994, as cited in
Alias et al 2008) and Mirskaulava (1967, as
cited in Mason & Arumugam 1985) rely on
the impingement angle of the jet with the
plunge pool surface. Both equations have
a narrow statistical spread; however, the
Yildiz and Uzucek equation overestimates
the scour depth, whereas the Mirskaulava
equation underestimates the scour depth.
Also, the median percentage difference
calculated using the Mirskaulava equation
(-78.6%) is closer to zero compared to the
Yildiz and Uzucek equation (126.1%). A
possible reason is that the Mirskaulava
equation also incorporates the plunge pool
depth and 90% percentile rock dimension,
whereas the Yildiz and Uzucek equation
does not.
The remainder of the empirical formu-
lae are based on the drop height and unit
discharge with varying coefficients. A wide
range in scour depths, both under- and
overpredicted, were calculated, indicating
that the scour depth does not rely solely on
the discharge and drop height, and that the
scour depth is sensitive to the parameter
coefficients. Furthermore, these equations
do not incorporate the rock block size, thus
the same scour depth would be predicted
if the discharge and drop height are kept
constant for various particle size scenarios.
These empirical formulae are therefore
considered less applicable for full-scale
prototype cases.
 80 90
Over-predicted Over-predicted
70 80
Predicted scour depth (m)

Predicted scour depth (m)

60 70
60
50
50
40
40
30
30
20 20
10 10
Under-predicted Under-predicted
0 0
0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80 90
Observed scour depth (m) Observed scour depth (m)
Veronese 1 Veronese 2 Veronese 3 Yildiz y=x Katoulas Jaeger Mirskaulava y=x

60 100
Over-predicted Over-predicted
90
Predicted scour depth (m)

Predicted scour depth (m)

50 80
40 70
60
30 50
40
20 30
10 20
10
Under-predicted Under-predicted
0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60 70 80 90 100
Observed scour depth (m) Observed scour depth (m)
Damle Chee Mason Martins y=x EIM Critical pressure DI QSI y=x

Figure 15 Comparison of equilibrium scour depths from the experimental work and the different prediction methods (Case B scenario)

The DI and QSI methods (Bollaert into account, whereas the QSI method is physical model scales, and that both meth-
2012) demonstrated the most significant the only identified scour prediction model ods were developed for vertical falling jets.
spread of differences between all the meth- that takes wall jets into consideration. Both The developed regression formula
ods. The DI method takes the uplift pulsat- methods over­estimated the scour depth, to predict the scour depth agreed well
ing forces integrated over a pulse period presumably due to the challenges posed by with the experimental results. There is a

Under-predicted Over-predicted
Method (%) (%)
–200 –100 0 100 200 300 400 500 600 700 800 900
1  Veronese 1 (1937)
2  Veronese 2 (1937)
3  Veronese 3 (1937)
4  Yildiz and Uzucek (1994)
5  Damle (1966)
6  Chee and Kung (1974)
7  Martins (1975)
8  Mason and Arumugam (1985)
9  Katoulas (1967)
10  Jaeger (1939)
11  Mirskaulava (1967)
12  EIM (Annandale 2006)
13  Critical pressure (Annandale 2006)
14  DI (Bollaert 2002)
15  QSI (Bollaert 2012)
16  New Formula (Case B)

Figure 16 Scour depth distribution as a percentage difference for the different scour prediction methods from the experimental work (Case B)

Journal of the South African Institution of Civil Engineering  Volume 62  Number 3  September 2020 49

10 000.0
1 000.0
Steam power (kW/m2)
100.0
10.0
1.0
0.1
0.01 0.1
Erosion
Annandale criteria
Figure 17 EIM erosion potential threshold (Annandale et al 1996) overlaid with experimental data
compact statistical range of differences
with a median difference percentage close
to zero (0.8%), with the first quartile just
below zero (-2%: overprediction) compared
to the literature scour prediction methods
as seen in Figure 16.
The sum of squared residuals (SSR),
as defined by Equation 24, was compared
against the various scour prediction meth-
ods in Figure 18 for the Case B scenario
(scour with deposited rocks removed).
New formula
Veronese 1 (1937)
Veronese 2 (1937)
Veronese 3 (1937)
Yildiz and Uzucek (1994)
Scour prediction methods
Damle (1966)
Chee and Kung (1974)
Martins (1975)
Mason and Arumugam (1985)
Katoulas (1967)
Jaeger (1939)
Mirskaulava (1967)
EIM (Annandale 2006)
Critical pressure (Annandale 2006)
DI (Bollaert 2002)
QSI (Bollaert 2002)
Figure 18 Comparison of the sum of squared residuals for the various scour prediction methods (Case B)
50
ranked seventh and eighth respectively in
the least total SSR.
Scour occurs
Conclusions
This paper presents the results of equi-
librium plunge pool scour hole geometry
obtained from a physical model study. A
model scale not exceeding 1:20 is recom-
mended for the experimental results from
this study in order to limit scale effects.
The physical model was used to simulate
and test the following conditions:
QQ Rectangular plunging jets issuing
horizontally.
No scour QQ Jet issuance levels of 3 m, 4 m, and 5 m
1 10 100 1 000 10 000 (prototype: 60 m, 80 m and 100 m)
Erodibility index above the rock bed level.
No erosion Current study QQ Plunge pool tailwater depths of 0.5 m
0% probability failure 100% probability failure and 1.0 m (prototype: 10 m and 20 m).
QQ Unit discharges of 0.1 m3/s/m to
0.224 m3/s/m (prototype: 35 m3/s/m to
80 m3/s/m).
SSR = ∑(dsobserved – dspredicted )2(24) QQ Two rock sizes of 0.1 m × 0.1 m ×
0.075 m and 0.1 m × 0.1 m × 0.05 m
The developed regression formula to (prototype: 2.0 m × 2.0 m × 1.5 m and
determine the scour hole depth ranked 2.0 m × 2.0 m × 1.0 m).
the highest with the least total SSR QQ Rock joint angles of 45°, 90° and 135°
26;13 (underpredicted; overpredicted), with the horizontal axis.
followed by Martins (1975) and Damle Results from the physical model were used
(1966). However, Martins’ and Damle’s to develop non-dimensional multi-linear
empirical formulae have a high number of regression formulae to predict the scour
underpredictions. The EIM and Critical hole geometry (depth, length, width and
Pressure methods (Annandale 2006) volume). The proposed regression formulae
26;13
0;38111
69;884
0;8919
0;5828
657;61
0;9129
366;151
26;1161
0;10248
120;16476
945;24
184;1580
461;5055
99;28317
340;20417
0 5 000 10 000 15 000 20 000 25 000 30 000 35 000 40 000
Sum of squared residuals
Under-predicted Over-predicted
Volume 62  Number 3  September 2020  Journal of the South African Institution of Civil Engineering
were confirmed to be inside the ranges
Notation
in which the formulae were developed by
using experimental test data that did not The following symbols are used in this paper:
form part of the regression analysis. These
Bi Jet issuance thickness
proposed regression formulae agree reason-
ably well with the experimental data on q
which they are based. Bg = Core thickness at impingement (Castillo et al 2014)
√2gHw
The scour depth results from the physi-
cal model were compared to various ana- q Impingement jet thickness for horizontal
Bj = + 4 × (1.14Tu)√Bi (√2Hw – 2√Bi)
lytical methods found in literature. It was √2gHw discharge channel (Castillo et al 2014)
found that, regarding the more frequently
Thickness of jet impinging on the rock bed, with
used formulae from literature, the Critical
Bj rock jet’s outer spreading angle as 15°
Pressure method agrees the closest with (Ervine & Falvey 1987)

⎫Y⎫ ⎫Y⎫
the experimental data and regression for-
CI = 0.035 ⎪ ⎪ – 0.119⎪ ⎪ + 1.22
2

⎭ Bg ⎭ ⎭ Bg ⎭
mulae developed in this study, whereas the Dynamic impulsion coefficient (Bollaert et al 2015)
EIM method (Annandale 2006) followed by
Mason and Arumugam’s (1985) empirical
Vj2
formula were more conservative than the Fuplift = CI γw (xb ∙ yb ) Lift force underneath block (Bollaert 2002)
formulae developed in this study. 2g

g Gravitational acceleration constant

Acknowledgements Gb = Vb(ρr – ρw )g Submerged weight of rock block (Bollaert 2002)

The assistance toward this research of
Hw Fall height
PERI Formwork Scaffolding, Continental
Cobbles, Kaytech Engineered Fabrics, Lb = 6q0.32 Jet breakup length (Horeni 1956)
and Horts Geo-Solutions, as well as
Mr NF Katzke and Mr DE Bosman, is P 1% Exceedance dynamic pressure head

acknowledged with gratitude. The opinions 1

expressed, and conclusions arrived at, are Pcrit = Puplift ∙ Critical pressure head
γ
those of the authors.
γQw Hw Stream power per unit area at rock bed
PSP =
Bj rock ∙ Wj rock (Annandale 2006)
REFERENCES
Albrecht, M, Nachtsheim, C, Albrecht, T & Cook, R CI γVj2
Puplift = Uplift pressure (Annandale 2006)
2013. Experimental design for engineering 2g
dimensional analysis. Technometrics, 55(3): 257–270.
q Unit discharge
DOI: 10.1080/00401706.2012.746207.
Alias, N, Mohamed, T, Ghazali, A & Mohd, M 2008. Vb Rock volume
Impact of takeoff angle of bucket type energy
dissipater on scour hole. American Journal of Vi Issuance velocity
Applied Sciences, 5(2): 117–121.
Velocity at impingement with plunge pool
Annandale, G, Smith, S, Nairns, R & Jones, J 1996.
Vj = √Vi2 + 2gHw
(Annandale 2006)
Scour Power. Civil Engineering, 66(7): 58–60.
VSS Rock block settling velocity
Annandale, G W 2006. Scour Technology: Mechanics
and Engineering Practice. New York: McGraw-Hill. V* = √9.81 ∙ γw ∙ Sf Shear velocity
Armitage, N & Rooseboom, A 2010. The link between
Movability Number and Incipient Motion in river Wi Flow width at issuance
sediments. Water SA, 36(1): 89–96. Thickness of jet impinging on the rock bed, with
Bollaert, E 2002. Transient water pressures in joints Wj rock jet’s outer spreading angle as 15°
and formation of rock scour due to high-velocity jet (Ervine & Falvey 1987)
impact. PhD Thesis, Lausanne, Switzerland: Ecole xb, yb, zb, d50 Rock block size
Polytechnique Fédérale de Lausanne.
Bollaert, E 2012. Wall jet rock scour in plunge pools: A ys Scour depth
quasi-3D prediction model. Hydropower and Dams,
yw Initial plunge pool depth
XX: 1–9.
Bollaert, E & Lesleighter, E 2014. Spillway rock scour ρr Rock density
experience and analysis – The Australian scene over
the past four decades. Proceedings, 5th International ρw Water density
Symposium on Hydraulic Structures, Brisbane,
v Kinematic viscosity of water at 15°C = 1.13×10–6 m2/s
Australia.

Journal of the South African Institution of Civil Engineering  Volume 62  Number 3  September 2020 51
Bollaert, E & Schleiss, A2005. Physically based
model for evaluation of rock scour due to
high-velocity jet impact. Journal of Hydraulic
Engineering, 131(3): 153–165. DOI: 10.1061/
(ASCE)0733-9429(2005)131:3(153).
Bollaert, E, Munodawafa, M & Mazvidza, D 2012.
Kariba Dam plunge pool scour: Quasi-3D numerical
predictions. Proceedings, ICSE 6-265, Paris,
pp 627–634.
Bollaert, E, Stratford, C & Lesleighter, E 2015.
Numerical Modelling of Rock Scour: Case Study
of Wivenhoe Dam (Australia). London: Taylor &
Francis.
Boushaba, F, Manso, P, Schleiss, A, Yachouti, A &
Daoudi, S 2013. Numerical and experimental high-
Reynolds jet diffusion and impact pressures in flat
and laterally confined aerated pools. International
Journal of Hydraulic Engineering, 2(6): 133–141.
DOI: 10.5923/j.ijhe.20130206.02.
Castillo, L & Carrillo, J 2014. Scour analysis
downstream of Paute-Cardenillo Dam. Proceedings,
3rd IAHR Europe Congress, Porto, Portugal,
pp 1–10.
Castillo, L, Carillo, J & Blàzquez, A 2014. Plunge pool
dynamic pressures: A temporal analysis in the
nappe flow case. Journal of Hydraulic Research,
53(1): 101–118. DOI: 10.1080/00221686.2014.968226.
52
Ervine, D & Falvey, H 1987. Behaviour of turbulent
water jets in the atmosphere and in plunge pools.
Proceedings of the Institution of Civil Engineers,
Part 2, 83(1): 295–314.
Horeni, P 1956. Disintegration of a free jet of water
in air. Byzkumny ustav vodohospodarsky prace a
studie, Sesit 93.
Kieffer, D & Goodman, R 2012. Assessing scour
potential of unlined rock spillways with the Block
Scour Spectrum. Geomechanics and Tunnelling,
5: 527–536.
Kirsten, H 1982. Classification system for excavation
in natural materials. The Civil Engineer in South
Africa, 24: 293–308.
Manso, P, Bollaert, E & Schleiss, A 2008. Evaluation of
high-velocity plunging jet-issuing characteristics as
a basis for plunge pool analysis. Journal of Hydraulic
Research, 46(2): 147–157.
Mason, P & Arumugam, K 1985. Free jet scour
below dams and flip buckets. Journal of Hydraulic
Engineering, 111(2): 220–235.
Noret, C, Girird, J, Munodawafa, M & Mazvidza, D
2012. Kariba Dam on Zambezi River: Stabilizing the
natural plunge pool. Proceedings, ICSE 6-265, Paris.
Pabst, W & Gregorova, E 2007. Characterization of
particles and particle systems. Proceedings, ICT,
Prague, p 122.
Volume 62  Number 3  September 2020  Journal of the South African Institution of Civil Engineering
Pells, S 2016. Erosion of rock in spillways. PhD Thesis.
Kensington, Australia: University of New South
Wales.
Stratford, C, Bollaert, E & Lesleighter, E 2013. Plunge
pool rock scour analysis techniques: Wivenhoe
Dam spillway, Australia. Proceedings, Hydro 2013
Conference, Innsbruck, Austria.
Van Aswegen, W J, Dunkley, E, Blake, K R K 2001.
Plunge pool scour reproduction in physical hydraulic
models. Stellenbosch: Water Research Commission.
Van Schalkwyk, A, Dooge, N & Pitsiou, S 1995. Rock
mass characteristics for evaluation of erodability.
Proceedings, 11th European Conference on Soil
Mechanics and Foundation Engineering, Volume 3,
Copenhagen, Denmark.
Van Schalkwyk, A, Jordaan, J & Dooge, N 1994. Die
erodeerbaarheid van verskillende rotsformasies
onder variërende vloeitoestande. Report WRC
302/1/95. Pretoria: Water Research Commission.
Whittaker, J & Schleiss, A 1984. Scour related to
energy dissipators for high head structures. Zürich:
Eidgenössischen Technischen Hochschule.
Wittler, R, Mefford, B, Abt, S, Ruff, J & Annandale,
G 1995. Spillway and dam foundation erosion:
Predicting progressive erosion extents. Proceedings,
International Conference on Water Resources
Engineering, San Antionio, TX, pp 1–5.
The effect of lane TECHNICAL PAPER
blockage on signalised Journal of the South African
Institution of Civil Engineering

intersection performance – ISSN 1021-2019

Vol 62 No 3, September 2020, Pages 53–60, Paper 0771

analysing and modelling PROF ERDEM DOĞAN is Assistant Professor

in the Department of Civil Engineering at
Kirikkale University, Turkey, where he has
been a faculty member since 2014. His
E Doğan, E Korkmaz, A P Akgüngör research interests are in the fields of traffic
flow, artificial intelligence applications in
traffic control, and prediction models.

Unexpected stops or entry/exit manoeuvres of vehicles on the road may cause the related lane
to become blocked. When this blocking happens in a signalised intersection zone, it also affects Contact details:
Department of Civil Engineering
intersection performance. Determining the extent of this effect will assist traffic engineers Kirikkale University
with intersection design and performance analysis. In this study, the effects of Lane Blockage Yahsihan / Kirikkale 71451
(LB) on intersection performance under various traffic conditions were analysed according Turkey
T: +90 318 357 42 42 Ext 1214
to two performance criteria. ANN (Artificial Neural Network) models were also developed to
E: edogan@kku.edu.tr
enable the prediction of intersection performance. As a result of the analysis, it was clearly
determined that the effect of LB on intersection performance was limited at v/c <0.5. However, PROF ERSIN KORKMAZ obtained degrees in
it was determined that the intersection performance may decrease between 10% and 110% Electrical and Electronics Engineering in
under the condition of 0.5 < v/c, depending on the LB frequency and duration. Additionally, 2011 and in Civil Engineering in 2012, both
from Erciyes University, Turkey. He has been
the developed ANN models have R > 0.95 and will therefore be useful in LB-related intersection
working as Assistant Professor in the Civil
performance analysis. Engineering Department at Kırıkkale
University since 2019. He completed his MSc
degree in 2016 and received his PhD in 2019,
and has published many articles and papers in national and international
INTRODUCTION Research indicated that bus-dwelling has
journals. He is interested in delay analysis, signalling, optimisation
A temporary stop occurring in the signal- a negative effect on delay when it happens techniques and the analysis of traffic accidents.
ised intersection approach lane usually in the range of 200 m from the stop line.
Contact details:
causes Lane Blockage (LB) on the right/ It was also found that the delay becomes Department of Civil Engineering
left (depending on driving-hand side) lane. worse the closer the bus-dwelling happens Kirikkale University
The reasons for this stop are random (e.g. to the stop line. Qi et al (2009) attempted Yahsihan / Kirikkale 71451
Turkey
accident, malfunction, etc) or non-random to develop probability models between T: +90 318 357 42 42 Ext 1080
(presence of a bus stop). In either case, LB delays and unwanted events occurring on E: ersinkorkmaz@kku.edu.tr
can significantly affect the performance the road. Ghasemlou et al (2016) examined
of the intersection. Precautions should be the impact of bus stops on road capacity by PROF ALI PAYIDAR AKGÜNGÖR is Professor
taken to prevent or alleviate intersection using the Cell Transmission Model. Their in the Civil Engineering Department at
Kirikkale University, Turkey, where he is the
performance reduction caused by LB. research showed that the negative effect of head of the Transportation Division and
Therefore, determining the traffic situation LB due to bus stops is significantly increased teaches transportation courses. He received
levels where the negative effects of LBs are closer to the stop line. Other related studies his BS and MSc degrees in Civil Engineering
from Istanbul Technical University, Turkey,
high is important for regulating counter- were done by Rodriguez-Seda and Benekohal
in 1989 and 1993 respectively, and earned
measures. In this study, the effects of LB (2006), Ghasemlou et al (2012), and Amita his PhD in Transportation Engineering from the University of Pittsburgh,
Frequency (F-LB) and LB Duration (D-LB) et al (2015). Ghasemlou et al (2014) also USA, in 1998. His research interests are traffic operation, traffic signal
at an isolated intersection were analysed tried to determine the effect of heavy vehicle control, signalised intersections, mathematical models of delay
estimation, queuing analysis and meta-heuristic optimisation techniques.
at various volume/capacity (v/c) ratios. In traffic on a signalised intersection. Although
addition, two ANN models were developed some attempts were made to resolve this Contact details:
Department of Civil Engineering
to predict Total Delay (T-Delay) and Move/ issue, there are still situations that have
Kirikkale University
Total time ratio (M/T) using the analysis not been fully clarified. For instance, as far Yahsihan / Kirikkale 71451
results. The (M/T) values are the ratio as we know, most studies have examined Turkey
of the moving time of the vehicles in the non-random bus stops and the effects of LB T: +90 318 357 42 42 Ext 1255
E: akgungor@kku.edu.tr
intersection during the simulation period on the system as a result of these stops. In
to the total time in the system. our study, random occurrence of LB was
Although studies about LB are limited also considered. In addition, control delay
in literature, from previous studies it was calculations as per the Highway Capacity
seen that researchers generally studied Manual (HCM) (TRB 2010) are suggested
the effects of bus movements on the for non-LB events. It is also stated that, in
intersection. Wong et al (1998) studied the the case of an LB event, this effect should Keywords: lane blockage, signalised intersection, ANNs, modelling,
effects of bus-dwelling in an intersection. be taken into account for delay analysis. delay, intersection performance

Doğan E, Korkmaz E, Akgüngör AP. The effect of lane blockage on signalised intersection performance – analysing and modelling.
J. S. Afr. Inst. Civ. Eng. 2020:62(3), Art. #0771, 8 pages. http://dx.doi.org/10.17159/2309-8775/2020/v62n3a5 53
With this study, if there is an LB event in
the intersection region, it will assist in the
performance analysis of the intersection and
in determining the level of performance.
This study therefore tried to address the gap
in this important issue.
In the next section (the methodological
section) the geometric and signal system
of the experimental intersection are illus-
trated, the LB event is described, and the
simulation parameters explained. In the
section following that the results of the LB
event analyses are given. Additionally, the
models are developed via ANNs and tested
according to the performance criteria. In
the last section, the analysis results are
discussed, and conclusions are reached.
INTERSECTION GEOMETRY, LANE
BLOCKAGE AND SIMULATION
Because safety is a problem it is very dif-
ficult to perform tests on the road itself
when determining how an LB event affects
intersection performance. Therefore,
and for the sake of safety and simplicity,
these studies were done using simulation
programs (today there are a number of
sophisticated traffic simulation programs
available for traffic studies). With the help
of these programs, variables that affect
traffic flow, including driver behaviour,
can be analysed by keeping other variables
constant. In this study, the simulations of
the LB event were performed with the TSIS
v6.3 program. In the CORSIM program,
situations occurring in less than 60 seconds
and blocking the lane are described as
short-term events. Examples of these events
are parallel parking, parking manoeuvres
and stops, as well as drop-off and pick-up of
passengers in the lane without a bus pocket.
The frequency and duration of these events
can be adjusted by entering them into the
simulation program so that their impact
on intersection performance can be ana-
lysed. TSIS v6.3 is a widely used program
consisting of CORSIM which simulates
microscopic traffic, TRAFED used for crea-
tion of CORSIM traffic network, and other
auxiliary modules (McTrans 2012).
The intersection shown in Figure 1 was
created with the TSIS program. A four-leg
intersection was used in the simulation,
which is widely used in field applications.
This intersection has two lanes of which the
width is about 3 m (10 feet) for all approach-
es. The intersection phase plan consists of
four phases, as shown in Figure 1. The green-
light periods of this intersection for all phase
54
Phase diagram
30 sec 30 sec 30 sec
φ1 φ2 φ3
L < 30 m
The vehicle blocking the right lane
The blocked vehicle, need to change lane
The moving vehicle
The standing vehicle
Figure 1 F our-leg intersection, phase diagram and lane blockage state
plans are 30 seconds. The vehicle volume in
each approach leg is distributed equally for
the right, left and straight movements. In
addition, it is assumed that there is no heavy
vehicle traffic in the intersection, but that
only passenger vehicles use this intersec-
tion. The aim is to decrease the number of
variables and to observe the LB effect more
clearly within these assumptions.
It is known that the negative effect of
the LB event on intersection performance
is elevated if the event is close to the stop
line (Ghasemlou et al 2016). Therefore,
situations in which an LB is 30 m away
or closer to the stop line were examined
in the simulations. Since the program
can only analyse LB events for the right
lane, the cases in which the LB happened
in the right lane were investigated. Due
to this restriction in the program, the
study was only performed for right-hand
drivers. The total duration (TimeSim) of
each simulation was determined as 3 600
seconds. Considering that the green phase
Table 1 Assumptions for intersection geometry, signal control and simulation
Assumptions
Intersection geometry
Number of lanes and width
Phase plan
Green, amber and all-red phases
Vehicle types
Vehicle movements
Mean start-up lost time
Mean discharge headway
Volume 62  Number 3  September 2020  Journal of the South African Institution of Civil Engineering
30 sec
φ4
is 30 seconds, the amber 3 seconds and the
all-red 1 second, approximately 26.5 cycles
would occur within 3 600 seconds.
Thus, the queuing effect, due to vehicles
being unable to pass the stop lane during
the green phase, formed part of the analysis
results. In addition, the simulations were
repeated 100 times with different arrival
headways, by keeping other parameters
constant. Then the average values of simu-
lation outputs were calculated and used for
comparison purposes. The assumptions for
intersection geometry, signal control and
simulation are shown in Table 1.
Four-leg intersections with two lanes
on each leg are the most common types of
intersections. This intersection type has
been dealt with in many simulation-based
studies (Soh et al 2010; Yang et al 2012).
The values of the mean start-up lost time
and mean discharge headway were selected
according to the Highway Capacity
Manual 2010 (TRB 2010). In studies where
capacity delay is investigated, the effect of
Type and value
Four-leg intersection
Two lanes and 3 m (10 feet)
Four-phase signal plan
30, 3 and 1 second(s)
Passenger car
Right, left and straight-through movements
Two-second phase
1.8 seconds
 T-Delay values grouped by v/c
60
v/c
50
0.1
0.3
40 0.5
T-Delay for LB (veh-h)
0.7
0.9
30
1.1
20
10
0 0.15
0 5
T-Delay for No-LB (veh-h)
Figure 2 P
 erformance criteria scatter graphs for No-LB and LB states grouped according to v/c ratios
heavy vehicles is also taken into account
(Ghasemlou et al 2014; Qi et al 2009). Since
only the effect of LB was investigated in
this study, the passenger car was used as
the vehicle type.
In the simulation, three variables (D-LB,
F-LB and v/c) were analysed by constantly
keeping two of them in simulation. D-LB
and F-LB variables were considered as {10,
20, …, 60 sec} and {10, 20, …, 60 LB/h}
respectively; the v/c ratio was also utilised
as {0.1, 0.3, …,1.1}. Therefore, the number
of simulations reached were 21 600 (6 × 6 ×
6 × 100). Another variable used in the anal-
ysis and in the development of the models
was the Event Total Duration (ETD). The
ETD was calculated by multiplying the fre-
quency and the Duration of the LB Event,
as illustrated Equation 1.
ETD = F-LB × D-LB
The fact that the ETD value is 3 600
at the intersection arm, in other words
(F-LB = 60 LB/h and D-LB = 60 sec), indi-
cates the situation in which the relevant
lane is constantly blocked.
EFFECT OF LB EVENT ON
INTERSECTION PERFORMANCE
Failure of traffic flow for any reason in the
lane of a signalised intersection will affect
the intersection performance. It can be
predicted that the effect of the LB event to
intersection performance will vary with
Journal of the South African Institution of Civil Engineering  Volume 62  Number 3  September 2020
0.50
0.45
0.40
0.35
M/T for LB
0.30
0.25
0.20
10 15 20 25 30
D-LB, F-LB and (v/c). Therefore, changes in
the performance of the intersection have
been examined according to different values
of D-LB, F-LB and v/c. As performance cri-
teria, the total delay time (T-Delay) and the
Move/Total time (M/T) values were con-
sidered. The T-Delay is a value expressed as
vehicle-hours and represents the total delay
values of all vehicles in the intersection dur-
ing the simulation period (Roess 2011).
Scatter plots for the presence and
absence of the LB event (LB vs No-LB)
are given in Figure 2. These graphs are
separated according to the two performance
criteria and grouped according to the v/c
ratios. It can be seen from Figure 2 that, as
the v/c ratio increases, the T-Delay values
increase and the M/T ratio decreases. In
the case of No-LB, the values of T-Delay
and M/T increase linearly on the dashed
(1) line shown in Figure 2, depending on the
increase in v/c. However, with the change
of F-LB and D-LB values, the spots begin to
scatter. The figure shows that the T-Delay
values started becoming affected after the
v/c ratio reached 0.7. Because with v/c = 0.7
(turquoise dots), the data started to scatter
along the vertical axis. This situation shows
itself more clearly when v/c reaches the 0.9
and 1.1 values. This tendency is similar in
the scattering of M/T values. However, M/T
values started to increase their scattering on
the graph after the v/c ratio had reached 0.5.
It is understood that the increase in
delay value is linear in under-saturated
traffic flow, in other words in the region
M/T values grouped by v/c
v/c
0.1
0.3
0.5
0.7
0.9
1.1
0.30 0.35 0.40 0.45 0.50
M/T for no LB
where the v/c ratio is less than 1. However,
it can be seen that the delay values expo-
nentially increase when the v/c value is
1.1. For instance, in the case of No-LB,
the T-Delay value, which is 3.27 veh-hr for
v/c  = 0.1, increases to 25.5 veh-hr when
v/c = 1.1. When the LB situation occurs, it
can be seen from Figure 2 that the T-Delay
value reaches 52 veh-hr for v/c = 1.1. For
M/T values, it is observed that M/T is
about 0.49 when the v/c is 0.1. With the v/c
reaching 1.1, the M/T value drops to 0.30
without the effect of LB events. When con-
sidering the LB effect, it is understood that
this has decreased to approximately 0.16.
The effect of LB events occurring at
distances close to the stop line on the delay
is shown in Figure 3 as a percentage change.
Depending on the v/c ratios, the percentage
of change between the number of delays
occurring in the absence and presence of LB
events is calculated with Equation 2.
|NoED – ED|
PCD = × 100 (2)
NoED
Where:
PCD = percentage change in the amount
of delay (%)
NoED = the level of delay when there is
no LB (veh-hr)
ED = the level of delay when there is
an LB (veh-hr).
When Figure 2 is analysed, it is observed
that there is an increase of 2.11% and 5.18%
55
in PCD values respectively in cases where
the v/c is 0.1 and 0.3, when the LB event 120
times reach 3 600 seconds. When the v/c v/c
reaches 0.7, this value is approximately 17%. 0.1
It can be seen that the PCD values increase 100 0.3
linearly with the increase of the v/c from 0.1 0.5
0.7
to 0.7. However, when the v/c value reaches 0.9
0.9, it is understood that the effect of the 80 1.1

LB event time has increased exponentially

(blue point scattering). In this case, there is

PCD (%)
60
a delay increment of about 60% when the LB
event duration is at its highest point (3 600
seconds). The state of v/c = 1.1, where the
40
traffic flow is higher than the capacity, is
given as red dots in Figure 3. It can be seen
that the PCD values, which had increased
20
rapidly since the beginning, increase by
approximately 103% when the LB event
duration reaches 3 600 seconds.
0
The scatter graph of the M/T values is 0 500 1 000 1 500 2 000 2 500 3 000 3 500 4 000
shown in Figure 4. The scatter pattern is ETD (sec)
similar to that in Figure 3. However, it is
observed that there is a difference in the Figure 3 P
 ercentage change in total delay for ETD
rate of change of M/T after short-term
events. The percentage change of M/T
(PCMT) is calculated by using Equation. 3. 50
v/c
|NoEMT – EMT| 45
PCMT = × 100 (3) 0.1
NoEMT 40 0.3
0.5
0.7
Where: 35 0.9
PCMT = percentage change in the 1.1
30
amount of M/T (%)
PCMT (%)

NoEMT = the level of M/T when there is 25

no LB (veh-hr)
EMT = the level of M/T when there is 20
an LB (veh-hr).
15

When Figure 4 is examined, it can be seen 10

that the maximum increase of PCMT is
about 48%. It is understood that this point 5
is where v/c is equal to 1.1 and the ETD
0
value is 3 600 seconds. Similar to the PCD 0 500 1 000 1 500 2 000 2 500 3 000 3 500 4 000
graphic in Figure 3, it is seen that the ETD (sec)
PCMT value decreases rapidly when the v/c
ratio and ETD values decrease. Figure 4 Percentage change of M/T for ETD
When Figures 3 and 4 are examined, it
is seen that there are some multiple points the obtained results from the hypothesis that high Frequency (F) or Event Duration
corresponding to the same v/c ratios and test are given in Table 2. (ED) does not have a statistical effect on
the same ETD values. For example, for In the case of the same ETD value, the performance. In this case, it is expressed
ETD = 3 000 seconds, it can be observed results of the two-sample t-test showed that ETD is the main effective.
that two red points (v/c = 1.1) or two blue
points (v/c = 0.9) occur at the same vertical Table 2 Hypothesis test results of frequency and event duration
axis. The reason for this is that they have Difference
the same ETD value after the multiplica- Performance Comparison t-stat P-value statistically
tion of different F-LB and D-LB values. significant
Hypothesis testing was performed to deter-
PCD [High-F & Low-ED] vs [High-ED & Low-F] –0.5394 0.5903 No
mine whether the different F-LB and D-LB
values giving the same ETD duration in the PCMT [High-F & Low-ED] vs [High-ED & Low-F] –0.4781 0.6331 No
delay and M/T graphs were effective, and

56 Volume 62  Number 3  September 2020  Journal of the South African Institution of Civil Engineering
 : Linear transfer function
: Hyperbolic tangent
ETD
: Artificial neorun group
v/c
Figure 5 T he network architecture which was used to estimate delay and M/T value input variables
FORECASTING DELAY
AND M/T WITH ANNS
As seen in Figures 3 and 4, different ETD
and v/c values are effective in changing the
T-Delay and M/T. It is also observed that
performance values change specifically at
different v/c ratios. Therefore, it is anticipated
that the parametric models can be applied
successfully. However, to be successful the
models have to be designed according to
each v/c value. For these reasons, it would
be appropriate to develop models by using
Artificial Neural Networks (ANNs) in order
to estimate T-Delay and M/T values. An
ANN is an artificial intelligence technique
inspired by biological neurons. By connecting
these neurons to each other with various
coefficients, artificial neural networks occur.
These networks are trained for different
events and are now widely used to develop
successful models. Various ANN-based stud-
ies exist, such as the development of delay
models for signalised intersections (Murat
& Baskan 2006; Murat 2006), signal control
(Lin & Lee 1991; Srinivasan et al 2006) and
queue length estimation (Chang & Su 1995).
The ANN network used for prediction of
the T-Delay and M/T were trained with the
Levenberg-Marquardt algorithm (Levenberg
1944; Marquardt 1963), and the network
architecture is shown in Figure 5. The hyper-
bolic tangent function in the hidden layer of
the network and the linear transfer function
in the output layer were used.
It is understood from Figures 2 and 3
that the ETD and v/c values directly affect
T-Delay and M/T. For this reason, ETD and
v/c variables were used as input variables in
ANN models. In this study, two different
ANN models estimating T-Delay and M/T
values are given in Figure 5.
For the training of the ANNs, 70% of
total data (216 × 0.70 = 152 data items), 15%
Journal of the South African Institution of Civil Engineering  Volume 62  Number 3  September 2020
sigmoid transfer function
T-Delay
(32 data items) for validation and 15% (32
data items) for testing were used. The data
used for testing and training was not con-
stant and consecutive but were randomly
taken from the data set. Additionally,
the performance of these models was
compared with the Mean Absolute Error
(MAE), Mean Absolute Percentage Error
(MAPE), Mean Squared Error (MSE), and
Root Mean Squared Error (RMSE) criteria
given in Equations 4–7.
1 N
MAE =
N
∑ |Fi – Oi|(4)
i=1
| |
1 N Fi – Oi
MAPE =
N
∑ Oi
∙ 100
i=1
N (Fi – Oi)2
MSE = ∑ N
(6)
i=1
N (Fi – Oi)2 ½
RMSE = ∑ N
(7)
i=1
Where:
Fi = the prediction (forecast) value
Oi = the observed value
N = the number of observations.
In ANNs training, ten-fold cross-validation
was used, i.e. ANNs were trained in ten
different training and validation sets divided
from the data set (McLachlan et al 2005).
The average of the error values obtained for
these ten sets was considered as the error
value of the ANNs. Thus, the performance
of ANNs with different network architec-
tures has been revealed more objectively.
To investigate the effect of the different
hidden layers on the performance of ANNs
for T-Delay estimation, the estimation
errors of the ANNs models with the num-
ber of hidden layers between 1 and 40 are
: Linear transfer function
: Hyperbolic tangent
ETD sigmoid transfer function
: Artificial neorun group
M/T
v/c
shown in Figure 6. The highest error value
is observed in the T-Delay model with only
one hidden layer neuron. In general, the
error values of the networks with different
neuron numbers than 1 are close to each
other and the lowest error values for the
T-Delay occurred in the 2 × 21 × 1 (number
of inputs – hidden neurons – output) net-
work architecture. The MAPE value at this
point is approximately 1.5%.
When comparing the estimates of the
training, validation and test sets of the
T-Delay model with the actual values, the
correlation (R) values approaching 1 were
found and are shown in Figure 7. It can be
seen that the prediction and actual values,
(5)
especially below 20 veh-hr delay values, are
located at the R = 1 line. At higher delay
values, it is understood that there is some
scattering. There are cases in the dataset
where v/c is less than 1 and larger. It is
understood from Figures 4 and 5 that the
intersection performance is affected dif-
ferently in these two situations. The model
consists of data covering both situations.
Therefore, situations where the v/c is less
than 1 (much data) are better predicted.
However, although the forecast was suc-
cessful in cases where v/c was higher
than 1, it was not as good as the forecast in
situations where v/c was less than 1. This
situation explains the scattering in Figure 7.
The effect of the different number of
hidden layers on the performance of ANNs
for M/T estimation is shown in Figure 8.
The highest error value was observed in
the M/T model with one hidden layer of
neurons, similar to the T-Delay model. With
the increase in the number of neurons, these
error values suddenly decrease. The lowest
error value is determined by the 2 × 19 × 1
network architecture. The MAPE value in
this network architecture is below 1%.
57
 MAE MAPE
1.6 20
1.4
1.2 15

1.0
Error

Error
10
0.8
0.6 5
0.4
0.2 0
0 10 20 30 40 0 10 20 30 40
Number of hidden neurons Number of hidden neurons

RMSE MSE
2.0 3.5

1.8 3.0

1.6 2.5

1.4 2.0
Error

Error
1.2 1.5

1.0 1.0

0.8 0.5
0 10 20 30 40 0 10 20 30 40
Number of hidden neurons Number of hidden neurons

Figure 6 P
 erformance of different numbers of hidden neurons belonging to the T-Delay estimation

Training: R = 0.99727 Validation: R = 0.98744

50 50
40 40

30 30
Prediction

Prediction

20 20

10 10

10 20 30 40 50 10 20 30 40 50
Actual Actual

Test: R = 0.99769 All: R = 0.99582

50 50

40 40
Prediction

Prediction

30 30

20 20

10 10

10 20 30 40 50 10 20 30 40 50
Actual Actual

Figure 7 C
 orrelation values and scatter plots of the T-Delay model

58 Volume 62  Number 3  September 2020  Journal of the South African Institution of Civil Engineering
 MAE MAPE
10 3.0

2.5
8
Error (× 10 –3)

2.0

Error
6
1.5

4
1.0

2 0.5
0 10 20 30 40 0 10 20 30 40
Number of hidden neurons Number of hidden neurons

RMSE MSE
0.015 16
14
12

Error (× 10 –5)
10
Error

0.01
8
6
4
0.005 2
0 10 20 30 40 0 10 20 30 40
Number of hidden neurons Number of hidden neurons

Figure 8 P
 erformance of different numbers of hidden neurons belonging to the M/T estimation

Training: R = 0.99831 Validation: R = 0.99802

0.45 0.45

0.40 0.40
Prediction

Prediction

0.35 0.35

0.30 0.30

0.25 0.25

0.20 0.20

0.20 0.25 0.30 0.35 0.40 0.45 0.20 0.25 0.30 0.35 0.40 0.45
Actual Actual

Test: R = 0.9996 All: R = 0.99837

0.45 0.45

0.40
0.40
Prediction

Prediction

0.35
0.35
0.30
0.30
0.25

0.25 0.20

0.25 0.30 0.35 0.40 0.45 0.20 0.25 0.30 0.35 0.40 0.45
Actual Actual

Figure 9 Correlation values and scatter plots of the M/T model

Journal of the South African Institution of Civil Engineering  Volume 62  Number 3  September 2020 59
 The correlation values and scatter
graphs of the training, validation and test
sets of the developed 2 × 19 × 1 ANNs
model for predicting the M/T ratio are
shown in Figure 9 (p 59). It can be seen
from Figure 9 that the correlation values
are very close to 1 for training, validation
and test sets. It is understood from the fig-
ure that the actual and predicted values for
above 0.3 of the M/T ratios are less scat-
tered than those below 0.3 for M/T ratios.
DISCUSSION AND CONCLUSION
In this study, the analysis of LB that
occurred in the signalised intersection
region was investigated. In the analysis,
varied v/c rates, LB frequencies and dura-
tions were considered. The effects of these
variables on intersection performance were
determined by creating many simulations.
T-Delay and MT ratios were observed as
performance criteria. In addition, LB and
intersection performance were modelled
using the results of the analysis. Other
results regarding analysis and models are
listed as follows:
1. Increasing ETD and v/c values decrease
the intersection performance.
2. If the ETDs of two LB events with
different F-LB and D-LB are equal, the
effects of these two different events on
intersection performance are considered
statistically the same.
3. The effect of the LB event on delay
significantly increases from v/c = 0.7.
This increase is also observed at lower
v/c rates for M/T.
4. For v/c <1, the performance is seriously
reduced only when the ETD gets closer
to the simulation time. On the other
hand, T-Delay and M/T ratios become
very sensitive to the increase in ETD
as it approaches the saturation level of
the intersection. Therefore, planning
should be organised for rapid response
to LB events at periods of vc >1 at
intersections.
5. At v/c ≤0.5, the increases in T-Delay
and M/T as a result of the ETD increase
are approximately linear. However, at
v/c >0.5 these increases are exponen-
tial. Further studies to investigate the
starting point of this change would
contribute to a better understanding of
the LB event.
6. The developed ANN-Latency and
ANN-MT models have very high cor-
relations. In addition, model errors
60
of training and test observations are
close to each other and minimal. This
indicates that models can generalise the
event, that is they do not memorise it.
As a result, the developed models can
be used in real-world applications.
LB occurs for a variety of reasons and it
is very difficult to prevent all of them.
However, their negative effects can be
alleviated by various measures. It may be
advisable to develop controls to avoid LB
events, especially in densely signalised
intersections. It is recommended that bus
stops be removed from within intersec-
tion zones, or that bus schedules are
rearranged according to the v/c ratios.
Additionally, designing increased shoulder
and lane widths could help reduce the
negative impact of an LB event. It is recom-
mended that response teams be located
at critical intersections during periods
of high v/c, thereby shortening the event
response time.
DISCLOSURE STATEMENT
No potential conflict of interest was
reported by the authors.
REFERENCES
Amita, J, Singh, J S & Kumar, G P 2015. Prediction
of bus travel time using artificial neural network.
International Journal for Traffic and Transport
Engineering, 5(4): 410–424. doi: 10.7708/
ijtte.2015.5(4).06
Chang, G L & Su, C C 1995. Predicting intersection
queue with neural network models. Transportation
Research Part C: Emerging Technologies, 3(3): 175–
191. doi: 10.1016/0968-090X(95)00005-4
Ghasemlou, K, Aydın, M M, Tanyel, S, Topal, A
& Çalışkanelli, S P 2012. Effect of dwell time
on performance of signalized intersections.
Proceedings, 10th International Congress on
Advances in Civil Engineering, Ankara, Turkey,
17–19 October 2012, pp 1–10.
Ghasemlou, K, Aydin, M M & Yildirim, M S 2016. An
investigation on lane blockage effects at signalized
intersections. International Journal for Traffic &
Transport Engineering, 6(3): 289–302. doi: 10.7708/
ijtte.2016.6(3).05
Ghasemlou, K, Topal, A, Aydın, M M, Tanyel, S
& Kazemi Afshar, A A 2014. Investigation
of the effect of heavy vehicles on capacity of
signalized intersections based on Bayes’ theorem.
Proceedings, 2nd Transportation & Development
Congress: Planes, Trains, and Automobiles,
Orlando, FL, 8–11 June 2014, pp 376–387. doi:
10.1061/9780784413586.036
Volume 62  Number 3  September 2020  Journal of the South African Institution of Civil Engineering
Levenberg, K 1944. A method for the solution of certain
non-linear problems in least squares. Quarterly
Journal of Applied Mathematics, 2(2): 164–168.
Lin, C T & Lee, C S G 1991. Neural-network-based
fuzzy logic control and decision system. IEEE
Transactions on Computers, 40(12): 1320–1336. doi:
10.1109/12.106218
Marquardt, D W 1963. An algorithm for least-squares
estimation of nonlinear parameters. Journal of the
Society for Industrial and Applied Mathematics,
11(2): 431–441.
McLachlan, G, Do, K A & Ambroise, C 2005. Analyzing
microarray gene expression data. New York: Wiley.
McTrans 2012. TSIS-CORSIM. Available at: http://
mctrans.ce.ufl.edu/mct/index.php/tsis-corsim.
Murat, Y S 2006. Comparison of fuzzy logic and
artificial neural networks approaches in vehicle
delay modeling. Transportation Research Part
C: Emerging Technologies, 14(5): 316–334. doi:
10.1016/j.trc.2006.08.003
Murat, YS & Baskan, O 2006. Modeling vehicle delays
at signalized junctions: Artificial neural networks
approach. Journal of Scientific and Industrial
Research, 65(7): 558–564.
Qi, Y, Teng, H & Martinelli, D R 2009. An investigation
of incident frequency, duration and lanes blockage
for determining traffic delay. Journal of Advanced
Transportation, 43(3): 275–299. doi: 10.1002/
atr.5670430303
Rodriguez-Seda, J & Benekohal, R 2006. Methodology
for delay-based passenger car equivalencies for
urban transit buses. Transportation Research
Record: Journal of the Transportation Research
Board, 1988: 127–137. doi: 10.3141/1988-18
Roess, R P, Prassas, E S & McShane, W R 2011. Traffic
engineering. New York: Pearson/Prentice Hall.
Soh, A C, Rhung, L G & Sarkan, H M 2010. MATLAB
simulation of fuzzy traffic controller for multilane
isolated intersection. International Journal on
Computer Science and Engineering, 2(4): 924–933.
Srinivasan, D, Choy, M C & Cheu, R L 2006. Neural
networks for real-time traffic signal control. IEEE
Transactions on Intelligent Transportation Systems,
7(3): 261–272. doi: 10.1109/TITS.2006.874716
TRB (Transportation Research Board) 2010. Highway
Capacity Manual (HCM 2010). Washington, DC:
Transportation Research Board, National Research
Council.
Wong, S C, Yang, H, Yeung, W A, Cheuk, S L &
Lo, M K 1998. Delay at signal-controlled
intersection with bus stop upstream. Journal of
Transportation Engineering, 124(3): 229–234. doi:
10.1061/(ASCE)0733-947X(1998)124:3(229)
Yang, W, Zhang, L, He, Z, & Zhuang, L 2012.
Optimized two-stage fuzzy control for urban
traffic signals at isolated intersection and Paramics
simulation. Proceedings, 15th International IEEE
Conference on Intelligent Transportation Systems,
16–19 September 2012, Anchorage, AK, pp 391–396.
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.