materials

Article
Mastering Yield Stress Evolution and Formwork
Friction for Smart Dynamic Casting
Anna Szabo 1,2 , Lex Reiter 2 , Ena Lloret-Fritschi 1,2 , Fabio Gramazio 1 , Matthias Kohler 1 and
Robert J. Flatt 2, *
1 Institute of Technology in Architecture, ETH Zurich, Stefano-Franscini-Platz 1, 8093 Zürich, Switzerland;
szabo@arch.ethz.ch (A.S.); lloret@arch.ethz.ch (E.L.-F.); gramazio@arch.ethz.ch (F.G.);
kohler@arch.ethz.ch (M.K.)
2 Institute for Building Materials, ETH Zurich, Stefano-Franscini-Platz 3, 8093 Zürich, Switzerland;
reiter@ifb.baug.ethz.ch
* Correspondence: flattr@ethz.ch




Received: 10 April 2020; Accepted: 26 April 2020; Published: 30 April 2020


Abstract: The construction industry is a slow adopter of new technologies and materials. However,
interdisciplinary research efforts in digital fabrication methods with concrete aim to make a real
impact on the way we build by showing faster production, higher quality and enlarged freedom of
design. In this paper, the potential and constraints of a specific digital slip-forming process, smart
dynamic casting (SDC), are investigated with a material-focused approach in the complex task of
producing thin folded structures. Firstly, the workability and the strength evolution of different
material compositions are studied to achieve the constant processing rate for SDC. Secondly, friction
between the formwork walls and the concrete, a key aspect in slip-casting, is studied with a simplified
experimental setup to identify if any of these mixes would provide an advantage for processing.
Finally, a theoretical framework is constructed to link the material properties, the process conditions
and the designed geometry. This framework introduces the ‘SDC number’ as a simplified approach
to formulate the process window, the suitable conditions for slip-forming. The experimental results
prove the assumption of the model that friction is proportional to yield stress for all base compositions
and acceleration methods regardless of the filling history. The results are evaluated in the context
of the narrow process window of thin folded structures as well as the wider process window of
columns. The necessity of consistent strength evolution is underlined for narrow windows. Further,
friction is shown to be the highest initially, thus with both narrow and wide process windows, after a
successful start-up the continuation of slipping is less prone to failure. The proposed theoretical
model could provide material and geometry-specific slipping strategy for start time and slipping rate
during production.

Keywords: smart dynamic casting; set on demand; accelerator; process window; SDC number

1. Introduction
Large-scale digital fabrication processes have been generating an ever-growing research interest
in the last few decades. They showcase potential strategies to enhance efficiency in the construction
industry while offering increased flexibility in design [1]. In the case of digital fabrication with
concrete, the control over early age hydration and rheology [2–7] eliminates the need for formworks or
allows for alternative material-efficient formwork systems instead of bulky conventional formwork
constructions [8].
This paper focuses on smart dynamic casting (SDC), a robotic slip-forming process pioneered
by Dr. Ena Lloret-Fritschi [9]. In SDC, a small formwork with a maximum height of 40 cm is used to

Materials 2020, 13, 2084; doi:10.3390/ma13092084 www.mdpi.com/journal/materials

Materials 2020, 13, 2084 2 of 24

produce one-to-one scale architectural elements such as columns, mullions and folded structures [9].
Materials 2020, 13, x FOR PEER REVIEW 2 of 24
The formwork is moved along a predefined path using a 6-axis robotic arm [9–12] or a linear axis [13–15].
The formwork
15]. The formwork is open at its
is open toptop
at its and bottom,
and it is
bottom, continuously
it is continuouslyfilled
filledfrom
fromthe
thetop
topand
and shapes
shapes the
the
concrete leaving through the bottom along its movement (see
concrete leaving through the bottom along its movement (see Figure 1). Figure 1).

Figure 1. The different fabrication setups of smart dynamic casting (SDC) for producing columns on
the left with a KUKA robotic arm [9], mullions in the middle with a linear axis [16], and thin folded
right with
structures on the right with an
an ABB
ABB robotic
robotic arm
arm [12].
[12].

Shaping
Shaping concrete
concreteininSDC SDC requires
requires thatthat
the the
concrete sets on
concrete setsdemand
on demandand builds up strength
and builds during
up strength
the robotic experiments at an expected rate based on previously conducted
during the robotic experiments at an expected rate based on previously conducted measurements measurements [11,17].
First, a large
[11,17]. First,retarded batch is prepared
a large retarded that is suitable
batch is prepared that isfor pumping
suitable and has a and
for pumping longhasenough
a long open time
enough
for
openthetime
whole forduration
the whole of the processof
duration [17,18]. Then, the
the process concrete
[17,18]. is incrementally
Then, the concrete accelerated
is incrementallywith
chemical
accelerated admixtures
with chemicaland cast into the formwork
admixtures and cast intoin a the
self-compacting
formwork instate without segregation.
a self-compacting By the
state without
time the concrete
segregation. By the exits
timethethemoving
concrete formwork, its strength
exits the moving has evolved
formwork, to support
its strength the weight
has evolved of the
to support
material
the weight of the material above, while maintaining its ability to deform plastically. The capacity to
above, while maintaining its ability to deform plastically. The capacity of the material of
deform at thetobottom
the material deformofatthe theformwork
bottom ofalong different paths
the formwork alongprovides
differentthe ability
paths to produce
provides a set of
the ability to
different
produce ageometries with the
set of different same formwork
geometries with the [10].
same formwork [10].
The formwork
formworkininSDC SDC is related
is relatedto theto geometry
the geometry of theofproduced element.
the produced The process
element. has already
The process has
proved
already itself
provedsuitable for non-standard
itself suitable for non-standard (150 × 150
columnscolumns (150mm rectangular,
× 150 d = 212 mm
mm rectangular, circular,
d = 212 mm
1 = 125 mm,
rcircular, r2 =
r1 = 125 mm,80 mm elliptical
r2 = 80 cross-section)
mm elliptical and mullions
cross-section) with a with
and mullions changing cross-section
a changing (from
cross-section
70 mm × 100 mm to 70 mm × 180 mm) [14]. Then, the SDC
(from 70 mm × 100 mm to 70 mm × 180 mm) [14]. Then, the SDC fabrication of thin folded fabrication of thin folded structures
(25 mm wall wall thickness
thicknessand and500 500mm mmlong longsides)
sides)waswas explored
explored as aaspotentially
a potentially lessless wasteful
wasteful and and
less
less labor-intensive manufacturing method [14,19] for this material-efficient
labor-intensive manufacturing method [14,19] for this material-efficient structural system structural system with
increased stiffness relative to the the thickness
thickness [20]. However, after the failures of the initial empirical empirical
studies itit became
becameobvious
obviousthat thatthethe experimentally
experimentally defined
defined guidelines
guidelines for for the slipping
the slipping criterion
criterion of SDCof
SDC [10] are not applicable for all formwork geometries. The
[10] are not applicable for all formwork geometries. The increased surface area of thinincreased surface area of thin folded
formworks provided stricter slipping slipping criterion,
criterion, meaning
meaning aa narrower
narrower process
process window
window[11]. [11].
This paper takes a material-focused approach to study the process limitations of SDC with a
theoretical framework based on the process window model published in [11]. It discusses different
material compositions that could provide easier processing processing even even for
for thin
thin folded
folded elements.
elements. In the first
part of the paper, the workability and the strength evolution of the different mixes are addressed and
correlated to geometrical limitations for the produced elements. The second part studies the friction
between the formwork walls and the concrete on a simplified experimental setup with the material
compositions defined in part one in order to validate validate assumptions of the process model and extract
parameters
parameters for for further
further calculations.
calculations. The third part discusses the theoretical model for the SDC process
to analyze and consequently handle SDC production and introduces the ‘SDC number’ to formulate
the process window. In doing so, various specific examples are presented, whereby values of the SDC
number specifically referring to our experimental conditions are given (directly by the upper bounds
of the process window). However, it should be understood that these values for the SDC number are
purely illustrative. Overall optimization of SDC will involve numerous factors that can imply a large
Materials 2020, 13, 2084 3 of 24

the process window. In doing so, various specific examples are presented, whereby values of the SDC
number specifically referring to our experimental conditions are given (directly by the upper bounds
of the process window). However, it should be understood that these values for the SDC number are
purely illustrative. Overall optimization of SDC will involve numerous factors that can imply a large
number of optimal combinations. What our model does is to help to identify these, but it does not offer
any “off the shelf universal SDC number” for any arbitrary production using SDC.

2. Materials and Methods

2.1. Materials
For the experiments described in this paper two different retarded base mix formulations,
a self-compacting concrete (noted ‘mix S’) and an ultra high-performance fibre reinforced concrete
(UHPFRC) (noted ‘mix U’), were tested with three suitable acceleration methods. These are summarized
in Tables 1 and 2, with additional details given in the following paragraphs.

Table 1. Mix designs used in mix S and U prior to acceleration.

Retarded Mix [kg/m3 ] Mix S Mix U

Sand (0–4 mm) 1367.60
Quartz sand (0.125–0.5 mm) 616.40
Glass fibres (Fibre Technologies AR2500H103) 7.35
Cement (CEM I 52.5R or N) 615.06 (R) 547.50 (N)
Silica fume (BASF Masterlife SF100) 32.37 191.60
Limestone 5–15 µm d50% (OMYA Betocarb-HP) 183.10
Limestone 3 µm d50% (OMYA Betoflow-D) 419.10
Water 247.19 192.20
Superplasticizer (BASF MasterGlenium ACE-30) 1.55 6.00
Sucrose (Sigma Aldrich 99.5% purity D(+)) 0.68 1.33
Ca(NO3 )2 (SigmaAldrich) 0.03
w/b (-) 0.38 0.26

Table 2. Composition of the accelerated S_C, S_A and U_A mixes and the w/b ratio of these mixes
after acceleration.

Accelerator Mix S_C Mix S_A Mix U_A

(g/lretarded concrete ) S_C_1 S_C_3 S_C_5 S_A * S_A_1 S_A_3 S_A_5 U_A_1 U_A_3 U_A_5
Sika Rapid C100 15.38 20.00 24.60
Al2 (SO4 )3 ·18H2 O 9.23 9.23 9.23 9.23 13.69 12.59 11.50
Water 31.95 31.95 32.20 32.44 75.48 71.85 68.21
Superplasticizer * 1.85 2.20 2.55
w/b acc. (-) 0.39 0.40 0.40 0.43 0.43 0.43 0.43 0.36 0.36 0.35
* Mass of the same product (BASF MasterGlenium ACE-30) which is a solution containing 70% of water.

2.1.1. Self-Compacting Concrete

The self-compacting concrete, mix S, had a mix design that was adjusted from previous SDC
formulations [11] in order to be more compatible with the demands for the slip-forming of thin folded
structures. It comprised 0–4 mm siliceous sand aggregates, a CEM I 52.5R Portland cement and silica
fume (Table 1).
The admixtures, a retarder (sucrose 99.5% from Sigma-Aldrich, Buchs, Switzerland) and a
superplasticizer (BASF MasterGlenium ACE 30, Basel, Switzeland) were directly added to the mixing
water. Then ‘S’ was mixed with a Collomix Collomatic XM 3-900 type forced action mixer (Gaimersheim,
Germany) for 7 min.
After this, the mixes were activated at specific times after initial mixing by using activators,
for which two types were considered (Table 2):
Materials 2020, 13, 2084 4 of 24

• a set accelerator: SikaRapid C-100 (reportedly based on Ca(NO3 )2 ·4H2 O [21]). These mixes are
noted S_C.
• a combination of aluminium sulfate solution (produced in house by dissolving Al2 (SO4 )3 ·18H2 O
in water to obtain 260 g/L concentration) and a superplasticizer (BASF MasterGlenium ACE 30).
These mixes are noted S_A.

In both cases, the accelerators were added prior to casting. This was done at three different
times after the initial water addition (1, 3 and 5 h), simulating the conditions for the slip-forming
process. For acceleration, the concrete was mixed again with the same mixer for 5 min. The exact
activation time-dependent composition is described with the third digit of the name. For example,
S_C_3 indicates that the retarded S mix was accelerated with accelerator C100, three hours after the
initial mixing with water.
It should also be noted that for the S_C mixes, the amount of accelerator added was increased
over time, while for S_A only the superplasticizer addition increased. Many preliminary trials showed
these choices to give the best performance in each case. Discussing why this is the case is beyond
the scope of this paper. However, to illustrate the benefit of time-dependent dosage, the yield stress
of the accelerated samples with both time-dependent and constant dosages (marked with “◦ ” after
the composition name) are compared in Section 3.1.1 with more information about the choice of
time-dependent dosages.

2.1.2. UHPFRC
The UHPFRC, mix ‘U’, had a mix design based on the formulation developed by Hajiesmaeili and
Denarié [22]. That original formulation was modified to enable its retardation and acceleration on
demand. It contained 0.1–0.4 mm silica sand, a CEM I 52,5N Portland cement, silica fume, two types of
limestone fillers and alkali-resistant Zirconia coated glass fibres (Table 1).
The admixtures (the same retarder and superplasticizer as mix S) were pre-dissolved in the water,
added together to the powder parts and mixed for ten minutes with a forced action mixer. The ‘mix
U’ was activated with the same self-prepared aluminum sulfate solution which was used for S_A,
however, the extra superplasticizer was replaced by water. This modification was necessary to prevent
the rapid skinning on the surface due to the evaporation of water [23]. The addition amounts were
defined in iterative experiments to achieve similar hydration kinetics in the first hour for S_C and S_A
mixes. The resulting accelerated mixes are noted U_A (Table 2).
The activator is added according to the same schedule and mixing protocol as for S_A. For U_A,
the accelerator amount is decreased over time. It is important to remember that the third digit of the
name again refers to the time of addition, but also relates to a modified activator dosage. For example,
sample U_A_3 indicates that mix U was activated by aluminum sulfate 3 h after initial mixing and the
specific dosage used in that case can be looked up in Table 2. As before, the choice of how to vary the
accelerator dosage was the result of a large number of preliminary tests only reported in this work
with the yield stress comparison of the time-dependent and constant acceleration (marked with “◦ ” on
Figure 3). More information about the choice of time-dependent dosages is given in Section 3.1.1.

2.2. Methods

2.2.1. Slump Tests

The workability of all mixes, before and after acceleration, was evaluated using a Hagerman’s
mini-slump cone with dimensions of 60 mm × 70 mm × 100 mm. The S and U retarded mixes were
tested approximately 45 min after water addition. The accelerated mixes were tested 6 min after the
start of activation in order to model the time difference between the start of acceleration and the arrival
in the formwork during a slip-casting experiment. These accelerated samples were activated at 1, 3 and
5 h either with constant or with time-dependent accelerator dosages. The yield stress was calculated
from the slump flow [24].
Materials 2020, 13, 2084 5 of 24

2.2.2. Penetration Tests

Penetrometer
Materials measurements
2020, 13, x FOR PEER REVIEWwere performed to monitor the yield stress evolution of the accelerated
5 of 24
samples, using protocols described in detail in another paper [11]. In short, a cylindrical needle
(d constant
= 19 mm, speed
h = 4ofmm)1 mm/s.
wasThe peak force
repeatedly recordedinduring
immersed these measurements
the sample is related
at different positions to the
with yield
a constant
stress [25].
speed of 1 mm/s. The peak force recorded during these measurements is related to the yield stress [25].

2.2.3.
2.2.3. Calorimetry
Calorimetry
Isothermalcalorimetry
Isothermal calorimetrymeasurements
measurements were
were performed
performedtotodetermine
determinethe
thehydration
hydrationkinetics both
kinetics both
in the retarded and accelerated compositions, according to the protocol described in detail [26].
in the retarded and accelerated compositions, according to the protocol described in detail in in [26].
Additionally, the amounts of heat released in the timespan of the penetrometer measurements were
Additionally, the amounts of heat released in the timespan of the penetrometer measurements were
compared with the measured strength evolutions.
compared with the measured strength evolutions.
2.2.4.
2.2.4. FrictionTests
Friction Tests
A friction test was designed to investigate the stresses occurring during the slipping process at
A friction test was designed to investigate the stresses occurring during the slipping process at
the interface of the concrete and the formwork by pulling out material strips from concrete. To make
the interface of the concrete and the formwork by pulling out material strips from concrete. To make it
it representative of our particular systems, these strips were made of the same material as the interior
representative of our particular systems, these strips were made of the same material as the interior of
of the SDC formwork. Moreover, as explained below, to see the evolution of friction force over time
the SDC formwork. Moreover, as explained below, to see the evolution of friction force over time the
the test is done with 5 strips, each pulled out at a different time.
test is done with were
The strips 5 strips,
0.5 each pulled
mm thick out
and at a of
made different time.
hard, clear PVC sheets with a width of 50 mm and a
height of 350 mm. They were distributed with equal spacing PVC
The strips were 0.5 mm thick and made of hard, clear alongsheets withaxis
the middle a width of 50 mmwith
of a container and a
height of 350
60 × 430 mm.
× 310 cmThey
innerwere distributed
dimensions. For with
betterequal spacing
control of their along the middle
position axis of a container
and displacement, with
the strips
× 430 ×through
60passed 310 cmainner dimensions. For better control of their position and displacement,
loose guiding piece providing a 1mm wide channel on the top and rest tightly in a the strips
passed
slit atthrough a loose
the bottom guiding
(see Figurepiece providing
2). In order toa avoid
1mm wide channel
friction from onthethe top and
bottom slitrest tightly
during thein a
slitmeasurement,
at the bottom each
(see Figure 2). In
strip was orderup
pulled to10mm
avoid friction from the
immediately afterbottom slit during
the container wasthe measurement,
filled (see the
each strip of
middle was pulled
Figure 2). up 10mm immediately after the container was filled (see the middle of Figure 2).

Figure
Figure2. 2.Left:
Left:friction
frictiontest
testbox
box is
is filled
filled with concrete.
concrete. Middle
Middleleft
leftand
andright:
right:section
section
of of
thethe friction
friction tests
tests
box showing the guiding piece on
box showing the guiding piece on top, the top, the slit in the bottom and the final position of the strips
the bottom and the final position of the strips before before
testing. Right:
testing. Right:friction
frictiontest
testperformed
performed on thethe first
first strip.
strip.

The concrete
The concretewas
wasplaced
placedaccording
accordingto
to two
two different protocols:
different protocols:
•  It was first
It was accelerated
first acceleratedasasaawhole
wholeandand then
then cast in the
cast in the box
boxall
allatatonce.
once.
•  It It
was gradually placed using 12 smaller batches, each accelerated
was gradually placed using 12 smaller batches, each accelerated separately separately prior
prior to placing.
to placing.
This procedure
This procedure was
wasdone
donesosoas
asto
toraise
raise the level of
of concrete
concretebyby1.81.8cm
cmevery
every 3 min,
3 min, which
which mimics
mimics
formwork
formwork filling
fillingduring
duringthe
theslip-forming
slip-forming process.
process.
InIn
both
bothcases,
cases,the
theacceleration
acceleration started
started 1h1h after
after water
wateraddition
additionandandthe
the mixing
mixing time
time waswas 2 min.
2 min.
However,
However,for forthe
the“one-go”
“one-go”filling
filling aa forced
forced action mixer
mixer was
wasused,
used,while
whileforforthethegradual
gradual filling
filling an an
overhead
overhead stirrer
stirrer hadhad to be
to be used
used because
because of the
of the smaller
smaller batch
batch sizes.
sizes. WithWith
bothboth filling
filling protocols,
protocols, the the
target
wastarget
that was that the
the strips strips
were 200were
mm 200
deepmm deep in concrete
in concrete after the
after leaving leaving the slit.
bottom bottomAnyslit. Any deviation
deviation from that
from that target height is noted and used for the calculations
target height is noted and used for the calculations accordingly. accordingly.
For each measurement, the box was positioned so that the strip to be pulled out could be
clamped to the cross-bar of a Zwick universal testing machine, while the box was fastened to the
platen of this device. Then, the strip was pulled out at constant 6 mm/min speed for 6 min, imitating
Materials 2020, 13, 2084 6 of 24

Materials
For 2020,
each13, x FOR PEER REVIEW
measurement, 6 of 24
the box was positioned so that the strip to be pulled out could be clamped
to the cross-bar of a Zwick universal testing machine, while the box was fastened to the platen of this
the slipping
device. Then, speed
the stripof was
SDCpulled
used foroutproducing
at constantnon-standard
6 mm/min speed thin for
folded prototypes
6 min, imitating[12]. For each
the slipping
speed of SDC used for producing non-standard thin folded prototypes [12]. For each strip (strips)
strip series, the first test was done 105 min after water addition, while the subsequent tests series,
were
the firstmeasured in 10105
test was done minminintervals afteraddition,
after water that. The forcethe
while needed for the
subsequent displacement
tests (strips) werewas logged
measured
induring
10 minthe pull-out
intervals of the
after that.strips. The needed
The force pull-outforstress was calculated
the displacement wasfrom this during
logged force with the contact
the pull-out of
the strips. The pull-out stress was calculated from this force with the contact surface of the concrete pull-
surface of the concrete and the strips. This test should not be confused with the standardized and
outstrips.
the tests for
Thishardened
test should concrete
not bewhere
confusedreinforcement is pulled out
with the standardized of hardened
pull-out tests forconcrete
hardened samples.
concrete In
our case,
where the materialiswas
reinforcement stillout
pulled a yield stress fluid
of hardened at the samples.
concrete time of testing.
In ourThis
case,isthe
why we define
material wasthestilltest
a
as a “friction test”. However, what is measured is a “pull out stress”, which might
yield stress fluid at the time of testing. This is why we define the test as a “friction test”. However, represent a source
of confusion
what otherwise.
is measured is a “pull out stress”, which might represent a source of confusion otherwise.
In parallel toto
In parallel thethe friction
friction tests,
tests, penetration
penetration teststests
werewere conducted
conducted on companion
on companion samples.
samples. These These
were
prepared from the same retarded base mix and were accelerated with the same compounds at the sameat
were prepared from the same retarded base mix and were accelerated with the same compounds
the same
time as thetime as the
samples samples
tested tested for friction.
for friction.

3.3.Part
Part1:1:Mastering
Masteringthe
theYield
YieldStress
StressEvolution
Evolution

3.1.Results
3.1. Results

3.1.1.
3.1.1.Spread
SpreadFlow
Flow
Figure
Figure3 represents
3 represents the the
yieldyield
stressstress
obtained by the spread
obtained by the flow measurements
spread that were conducted
flow measurements that were
on both retarded
conducted on both andretarded
accelerated
andsamples, investigating
accelerated the rheological
samples, investigating theconsequences of accelerationof
rheological consequences
for fluidity at for
acceleration thefluidity
time of at
casting (values
the time given in
of casting Table given
(values A1 of intheTable
Appendix
A1 of A).
theThe non-accelerated
appendix). The non-
mixes, S and U mix, are shown for time 0 with open symbols. The accelerated samples
accelerated mixes, S and U mix, are shown for time 0 with open symbols. The accelerated samples are reported as
aare
function of the
reported as atime at which
function thetime
of the accelerator
at whichwastheadded (1, 3, 5was
accelerator h). added (1, 3, 5 h).

Figure 3. Mini-slump test results expressed as yield stress. On the left, the accelerator dosages were
Figure 3. Mini-slump test results expressed as yield stress. On the left, the accelerator dosages were
constant over time. On the right, time-dependent accelerator amounts were used.
constant over time. On the right, time-dependent accelerator amounts were used.

The non-accelerated mix U has significantly larger yield stress than the S mix, despite the S mix
The non-accelerated mix U has significantly larger yield stress than the S mix, despite the S mix
still being fluid enough for pumping [11]. The effect of adding an accelerator is shown for both constant
still being fluid enough for pumping [11]. The effect of adding an accelerator is shown for both
dosages (Figure 3, left) and time-dependent ones (Figure 3, right).
constant dosages (Figure 3, left) and time-dependent ones (Figure 3, right).
For the S mixes, it can be seen that these additions caused an initial drop in the yield stress which
For the S mixes, it can be seen that these additions caused an initial drop in the yield stress which
was increasingly less important as the addition time is increased from 1 to 5 h, both for the C and
was increasingly less important as the addition time is increased from 1 to 5 h, both for the C and the
the A type accelerators. However, Figure 3 (right) shows that if the accelerator dosage is increased
A type accelerators. However, Figure 3 (right) shows that if the accelerator dosage is increased as
as defined in Table 2, the yield stress can be maintained. This is important for maintaining constant
defined in Table 2, the yield stress can be maintained. This is important for maintaining constant
process conditions in SDC.
process conditions in SDC.
When changing the accelerator of the S mix to the aluminium sulfate solution (S_A*) there was
When changing the accelerator of the S mix to the aluminium sulfate solution (S_A*) there was
flow loss (see Table A1 in Appendix A), which led us to consider additional and simultaneous additions
flow loss (see Table A1 in Appendix), which led us to consider additional and simultaneous additions
of superplasticizer to avoid this problem. However, a dosage giving initial yield stress similar to S_C_1
of superplasticizer to avoid this problem. However, a dosage giving initial yield stress similar to
S_C_1 leads to a great increase in yield stress at later acceleration times than what was obtained
previously (S_C_1°). As before, this change could be mitigated by changing the superplasticizer
Materials 2020, 13, 2084 7 of 24
Materials 2020, 13, x FOR PEER REVIEW 7 of 24

leads to a of
content great increase inin
the accelerator yield stress
relation at later
to the time acceleration times than
of activation (Table what
2), once was
again obtaineda previously
providing solution
(S_C_1 ◦ ). As before, this change could be mitigated by changing the superplasticizer content of the
to maintain yield stress constant throughout the process (Figure 3, right).
accelerator in relation
For the to the
U_A mixes, thetime ofstress
yield activation (Table
increased 2), once
upon again providing
acceleration, a solution
contrary to the S_A to maintain
mixes. In
thisstress
yield case, however,
constant the yield stresses
throughout remained
the process low over
(Figure time, even with a constant accelerator dosage
3, right).
(U_A_1°).
For the DueU_Atomixes,
this, there was no
the yield need
stress to add additional
increased amounts ofcontrary
upon acceleration, superplasticizer. Additionally,
to the S_A mixes. In this
to mitigate the small yield stress increase taking place at late acceleration times,
case, however, the yield stresses remained low over time, even with a constant accelerator dosage small reductions of
the aluminium
(U_A_1 ◦ ). Due tosulfate solution
this, there was proved
no needtotobeadd most effective (Figure
additional amounts 3, of
right).
superplasticizer. Additionally,
to mitigate the small yield stress increase taking place at late acceleration times, small reductions of the
3.1.2. Penetration Tests
aluminium sulfate solution proved to be most effective (Figure 3, right).
Figure 4 shows that the evolution of penetration resistance in S_C mixes is very similar,
3.1.2. Penetration
irrespective Tests
of the time at which the accelerator is added. This, however, relates to the situation in
which the4accelerator
Figure shows thatdosage is adapted
the evolution to maintainresistance
of penetration constant initial
in S_Cyield
mixesstress after
is very its addition,
similar, as
irrespective
explained above (Figure 4) [11]. The initial penetration resistance is expected to be
of the time at which the accelerator is added. This, however, relates to the situation in which the similar, since this
also represents a measurement of yield stress [27]. However, the fact that after this the increase in
accelerator dosage is adapted to maintain constant initial yield stress after its addition, as explained
yield stress should be similar is not self-evident. For SDC, it is, however, a very useful result because
above (Figure 4) [11]. The initial penetration resistance is expected to be similar, since this also
it maintained similar material evolution kinetics throughout the process. In particular, for the
represents a measurement of yield stress [27]. However, the fact that after this the increase in yield
empirically defined slipping range of 4–6 N of penetration resistance it defined a fixed time of
stress should be similar is not self-evident. For SDC, it is, however, a very useful result because it
occurrence (50 min) as well as a fixed window of operation.
maintained similar material evolution kinetics throughout the process. In particular, for the empirically
With respect to the theoretical model presented later in this paper, we can point out that when
defined
the dataslipping range
are plotted onofa 4–6 N ofscale,
log-log penetration resistance
their appearance asita defined a fixed
straight line time of
indicates occurrence
that (50 min)
the yield stress
as is
well as a fixed
increasing as awindow
power law of operation.
of the time elapsed after addition of the accelerator.

Figure
Figure4. 4.Strength
Strengthevolution
evolutionof of the
the S_C
S_C Mix with
with C100
C100accelerator
acceleratorwith
withincreased
increased dosage
dosage over
over time.
time.
(Left) Lin-lin
(Left) Lin-linscale
scalereproduced
reproduced fromfrom[11].
[11].(Right)
(Right) Log-log
Log-log scale.
scale. The The straight
straight line obtained
line obtained on the on
log-the
log-log scale
log scale highlights
highlights thethe fact
fact that
that thethe yield
yield stress
stress evolution
evolution cancan well
well be be fitted
fitted byby a power-law
a power-law (see
(see
Equation
Equation (3)).
(3)).

With respect toyield

A consistent the theoretical modelover
stress evolution presented
the 5 hlater in thiswas
timespan paper,
alsowe can point
achieved byout
the that when the
aluminate-
based accelerator (S_A series), provided by the concentrated aluminium sulfate solution
data are plotted on a log-log scale, their appearance as a straight line indicates that the yield stress is combined
with a superplasticizer
increasing as a power law (see
of Figure
the time 5). elapsed
In this case, it required
after addition the superplasticizer
of the accelerator. dosage to be varied.
AsAbefore, the fact
consistent yieldthat similar
stress initial over
evolution yieldthe
stresses were obtained
5 h timespan was alsoisachieved
not surprising, since dosages
by the aluminate-based
were adjusted to obtain similar flow spreads. The fact that the yield stress
accelerator (S_A series), provided by the concentrated aluminium sulfate solution combined growth over time shouldwith
be the same is not trivial, but it is, as previously mentioned, very useful.
a superplasticizer (see Figure 5). In this case, it required the superplasticizer dosage to be varied.
As can
As before, best that
the fact be seen in the
similar log-log
initial yieldplot (Figure
stresses 5, right),
were thisisacceleration
obtained method
not surprising, sincealso leads were
dosages to
power-law increases in yield stress over time. In this case, however, it provides lower exponents
adjusted to obtain similar flow spreads. The fact that the yield stress growth over time should be the
(lower slopes on the log-log plot). Apart from this, we observed that the late measurements showed
same is not trivial, but it is, as previously mentioned, very useful.
larger scatter than for the S_C mixes. The experimental slipping range was reached again around 50
min.
Materials 2020, 13, 2084 8 of 24
Materials
Materials2020,
2020,13,
13,xxFOR
FORPEER
PEERREVIEW
REVIEW 88of
of24
24

Figure
Figure 5. 5.5.
Figure Strength
Strength evolution
Strengthevolution
evolutionof of the
ofthe S_A
theS_A
S_AMixMix with
with aluminium
Mix with aluminium sulfate
aluminium sulfatesolution
sulfate solutionas
solution asasaccelerator
accelerator
accelerator andand
and
simultaneous
simultaneous
simultaneous superplasticizer
superplasticizer
superplasticizer addition
addition of an
additionofofan increased
anincreased amount
increasedamount over
amount over time. (Left)
over time. (Left) lin-lin
(Left)lin-lin scale.
lin-linscale. (Right)
scale.(Right)
(Right)
log-log
log-log
log-log scale.
scale.
scale. The
The
The straight line
straightline
straight obtained
lineobtained on
obtainedon the
onthe log-log
log-log scale
the log-log scale highlights
highlightsthe
highlights thefact
the factthat
fact thatthe
that the
theyield
yield
yieldstress
stress
stress
evolution
evolution
evolution can can
can well
bebe
well
well be fittedbyby
fitted
fitted bya apower-law
apower-law (see
power-law(see Equation
(seeEquation (3)).
Equation(3)).
(3)).

AsThecanU_A
The best mix
U_A mix also
be seen shows
also in similar
similar penetration
the log-log
shows plot (Figureresistance
penetration 5, right), evolutions
resistance after
after acceleration
this acceleration
evolutions method also
acceleration addition,
leads to
addition,
irrespective
irrespective
power-law of the
of the in
increases addition
addition time (Figure
time (Figure
yield stress 6). Similarly,
6). Similarly,
over time. In this case,as in
as in the two previous
the twoitprevious
however, cases,
providescases,
lower this
this was
wasobtained
exponents obtained
(lower
by
by adjusting
adjusting the
the accelerator
accelerator dosage
dosage over
over time
time to
to get
get the
the same
same initial
initial
slopes on the log-log plot). Apart from this, we observed that the late measurements showed largeryield
yield stress.
stress. However,
However, in
in contrast
contrast
to
to the
the two previous
twofor
previous cases,
cases, the U_A
theThe mixes
mixes required
U_Aexperimentalrequired the
the dosage
dosage to be
be decreased
towas decreased rather than
than increased.
ratheraround increased.
scatter than the S_C mixes. slipping range reached again 50 min.
The
The reasons
reasons for these
for these differences
differences are due
are due to the underlying
to the underlying mechanism
mechanism of each acceleration
acceleration method
The U_A mix also shows similar penetration resistance evolutionsofafter
each acceleration method
addition,
and
and are
are beyond
beyond the
the scope
scope of
of this
this paper,
paper, which
which focusses
focusses instead
instead on
onshowing
showing how
how consistent
consistent behaviour
behaviour
irrespective of the addition time (Figure 6). Similarly, as in the two previous cases, this was obtained
can
can bebe obtained
obtained in
in order
order to enable
todosage aamore
enableover more robust
robust SDC
SDC process.
by adjusting the accelerator time to get theprocess.
same initial yield stress. However, in contrast
Importantly,
Importantly, for for the
the process
process modelling, we
we once
once again found
found that
that aa power-law
power-law can can describe
describe the
to the two previous cases, the U_Amodelling,
mixes required theagain
dosage to be decreased rather than the
increased.
strength
strength evolution
evolution well well after
after accelerator
accelerator addition
addition (Figure
(Figure 6, 6, right).
right). The
The exponents
exponents of of the
the strength
strength
The reasons for not
evolution these differences are due to the underlying mechanism of each acceleration method
evolution do do not depend
depend on on the
the time
time of
of accelerator
accelerator addition
addition and and are
are in
in aa similar
similar range
range to to the
the S_C
S_C
and are
samples. beyond the scope of this paper, which focusses instead on showing how consistent behaviour
samples. However,
However, the the target
target value
value ofof 4–6N
4–6N of of penetration
penetration resistance
resistance was
was reached
reached slightly
slightly earlier
earlier
canthan
be obtained in order to enable a more robust SDC process.
than before.
before.

Figure
Figure
Figure 6. 6.
6.Strength
Strength
Strength evolution
evolutionofof
evolution the
ofthe U_A
theU_A
U_AMixMix with
Mixwith aluminium
aluminiumsulfate
withaluminium sulfatesolution
sulfate solutionas
solution asan
as anaccelerator
an accelerator
acceleratorwith
with
withaa a
decreased
decreased amount
amount over
over time.
time. (Left)
(Left) Lin-lin
Lin-lin scale.
scale. (Right)
(Right) Log-log
Log-log scale.
scale. The
The straight
straight
decreased amount over time. (Left) Lin-lin scale. (Right) Log-log scale. The straight line obtained on line
line obtained
obtained on
on
thethe
the log-log
log-log
log-log scale
scale
scale highlights
highlights
highlights thethe
the
factfact
fact
thatthat
thethe
that the yield
yield
yield stress
stress
stress evolution
evolution
evolution can can
can well
bebe
well
well be fitted
byby
fitted
fitted by aa power-law
power-law
a power-law (see
(see Equation
(see Equation (3)).
Equation (3)). (3)).

3.1.3.
3.1.3. Calorimetry
Calorimetry
Importantly, for the process modelling, we once again found that a power-law can describe the
strengthAsevolution
As shown
shown in inwell
Figureafter
Figure 7, accelerator
7, the
the non-accelerated
non-acceleratedaddition (Figure
SS mix
mix had aa6,
had right). The
particularly
particularly exponents
long
long inductionof
induction the strength
period,
period, with
with
evolution
an onsetdo
an onset of not
of the depend
the main
main on theat
hydration
hydration time
at of accelerator
around
around 7–10
7–10 h.
h. Thisaddition
This time
time frameandprovides
frame are in aworkability
provides similar range
workability to thefor
retention
retention S_C
for
samples.
aawhole However, the target value of 4–6N of penetration resistance was reached
whole work-day that is needed for our typical SDC production conditions. The figure also showsaa
work-day that is needed for our typical SDC production conditions. The figure slightly
also earlier
shows
strong
than reduction
reduction in
before.
strong in the
the duration
duration of of the
the induction
induction period
period when
when accelerator
accelerator CC was
was added.
added. However,
However,
while
while the
the initial
initial hours
hours are
are very
very similar,
similar, the
the hydration
hydration kinetics
kinetics differ
differ after
after about
about 44 h.
h. The
The shape
shape of
of the
the
three
three curves
curves also
also suggests
suggests that
that there
there is
is an
an inversion
inversion of
ofthe
the aluminate
aluminate and
and silicate
silicate hydration,
hydration, with
with the
the
Materials 2020, 13, 2084 9 of 24

3.1.3. Calorimetry
As shown in Figure 7, the non-accelerated S mix had a particularly long induction period, with an
onset of the main hydration at around 7–10 h. This time frame provides workability retention for a
whole work-day that is needed for our typical SDC production conditions. The figure also shows a
strong
Materialsreduction in the
2020, 13, x FOR PEERduration
REVIEW of the induction period when accelerator C was added. However, 9 of 24
while the initial hours are very similar, the hydration kinetics differ after about 4 h. The shape of the
most curves
Materials
three intense
2020, 13, peak
x FOR
also probably
PEER
suggests that reflecting
REVIEW there is ansulfate depletion
inversion and the main
of the aluminate silicatehydration,
and silicate peak following 9 of
with at
24
the
around 15–20 h depending on the time of accelerator addition [5]. This implies that
most intense peak probably reflecting sulfate depletion and the main silicate peak following at around the main strength
most
gain isintense
15–20 h still
dependingpeakdelayed,
rather probably
on the time reflecting
which sulfate
is something
of accelerator depletion
to consider
addition and
when
[5]. This the mainthatsilicate
producing
implies peak
tallmain
the following
structures.
strength gain at
is
around 15–20 h depending on the time of accelerator addition [5]. This
still rather delayed, which is something to consider when producing tall structures. implies that the main strength
gain is still rather delayed, which is something to consider when producing tall structures.

Figure 7. Hydration kinetics of the S mix accelerated with Sika Rapid C100.

Figure 7. Hydration kinetics of the S mix accelerated with Sika Rapid C100.
This pattern is not
Figure 7. seen in thekinetics
Hydration S_A samples (Figure
of the S mix 8). The
accelerated shape
with Sika of their
Rapid hydration curves is
C100.
veryThis
similar and does not suggest an inversion of the aluminate and silicate
pattern is not seen in the S_A samples (Figure 8). The shape of their hydration peaks. The silicate peak
curves is
Thisto
appears pattern
occur is not seen
around 10 in quickly
h, the S_Afollowed
samples by (Figure
the 8). Thedepletion
sulfate shape ofpeak.
their Apart
hydrationfrom curves
this, is
we
very similar and does not suggest an inversion of the aluminate and silicate peaks. The silicate peak
very similar
also observed and does
a around not
much lower suggest an inversion of the aluminate and silicate peaks. The silicate peak
appears to occur 10 h, heat flow
quickly and thus
followed byathemuch lower
sulfate reaction
depletion rate Apart
peak. in the from
induction period.
this, we also
appears
In fact, to occur
the values around
are 10 h,toquickly
similar those offollowed
the by the sulfate
non-accelerated S depletion peak. Apart from this, we
formulation.
observed a much lower heat flow and thus a much lower reaction rate in the induction period. In fact,
also observed
Importantlya much lower
for SDC heat flow
process andthe
control, thus a much
onset of lower reaction
hydration occursrate in the
earlier withinduction period.
this acceleration
the values are similar to those of the non-accelerated S formulation.
In fact, the
method values
than areS_C
for the similar to those
mixes and isof theindependent
also non-accelerated S formulation.
of addition time.
Importantly for SDC process control, the onset of hydration occurs earlier with this acceleration
method than for the S_C mixes and is also independent of addition time.

Figure 8.8. Hydration

Figure Hydration kinetics
kinetics of
of the
the SS mix
mix accelerated
accelerated with
with the
the combination
combination of
of aluminium
aluminium sulfate
sulfate
solution and a superplasticizer.
solution and a superplasticizer.
Figure 8. Hydration kinetics of the S mix accelerated with the combination of aluminium sulfate
Importantly for SDC process control, the onset of hydration occurs earlier with this acceleration
Lookingand
solution next at the U_A samples, it is evident that the induction periods are shorter and the
a superplasticizer.
method than for the S_C mixes and is also independent of addition time.
heat flows are intermediate between S_C and S_A (Figure 9). The hydration curves do not overlap as
well Looking next onset
as S_A, with at thetimes
U_A occurring
samples, it is evident
later that
for earlier the induction
additions. periods
Finally, arenoshorter
there is and the
peak inversion
heat flowsaluminates
between are intermediate between S_C and S_A (Figure 9). The hydration curves do not overlap as
and silicates.
well as S_A, with onset times occurring later for earlier additions. Finally, there is no peak inversion
between aluminates and silicates.
Materials 2020, 13, 2084 10 of 24

Looking next at the U_A samples, it is evident that the induction periods are shorter and the
heat flows are intermediate between S_C and S_A (Figure 9). The hydration curves do not overlap as
well as S_A, with onset times occurring later for earlier additions. Finally, there is no peak inversion
between aluminates
Materials 2020, and silicates.
13, x FOR PEER REVIEW 10 of 24

Figure 9. Heat flow over time from time of acceleration for the U mix accelerated with diluted
Figure 9. Heat flow over time from time of acceleration for the U mix accelerated with diluted
aluminium sulfate solution.
aluminium sulfate solution.
3.2. Discussion
3.2. Discussion
3.2.1. Initial Yield Stress after Accelerator Addition
3.2.1.To
Initial
begin Yield
this Stress after Accelerator
discussion, let us recallAddition
that we are considering a retarded concrete batch being
pumped and activated
To begin before being
this discussion, let us delivered
recall thatintowe an areSDC mould that
considering is slipped
a retarded vertically.
concrete batchFor this
being
process
pumpedtoand runactivated
smoothlybefore
throughout
being the production
delivered into an of SDC
an element,
mould we thatwould like the
is slipped residence
vertically. Fortime
this
of the concrete
process in the formwork
to run smoothly throughout to bethe
constant.
production Thisof means that thewe
an element, time needed
would like to
thereach the target
residence time
“exit-yield stress”
of the concrete in should be constant.
the formwork to be constant. This means that the time needed to reach the target
The results
“exit-yield stress”show that this
should is only possible if the accelerator dosage is varied over time. The concrete
be constant.
is already ageingshow
The results beforethat
it gets
thispumped and as aifresult
is only possible even smalldosage
the accelerator amounts is of hydration
varied products
over time. The
formed in this period can influence rheological behaviour substantially
concrete is already ageing before it gets pumped and as a result even small amounts of hydration because they create a high
specific
products surface
formed that has aperiod
in this profound consequence
can influence on yieldbehaviour
rheological stress [28,29].substantially because they create
a highFor the S_A
specific series,that
surface thehasaddition of an consequence
a profound accelerator initially
on yield reduces the yield stress with respect
stress [28,29].
to theForbase
the concrete.
S_A series,For the both
additionaccelerators, this may
of an accelerator be partially
initially reducesdue the to thestress
yield waterwiththeyrespect
contain.to
Further,
the base the presence
concrete. Forofboth
calcium ions can this
accelerators, additionally promotedue
may be partially superplasticizer
to the water they adsorption.
contain.For the A
Further,
accelerator,
the presence theofadditional
calcium ions amount canofadditionally
superplasticizer promoteplayssuperplasticizer
an important role. Thus, in both
adsorption. For thecases,A
we can expect
accelerator, thethat the accelerator
additional amount addition initially increases
of superplasticizer playstheanamount
important of adsorbed
role. Thus, superplasticizers.
in both cases,
In
weany case,
can the fact
expect thatthe
that the superplasticizer
accelerator addition addition decreases
initially the yieldthe
increases stress of the base
amount concrete
of adsorbed
also shows that increasing its dosage can counteract the yield stress increase
superplasticizers. In any case, the fact that the superplasticizer addition decreases the yield stress of obtained for constant
dosages
the baseinconcrete
the accelerated compositions.
also shows that increasing its dosage can counteract the yield stress increase
For the
obtained forUconstant
series, the accelerator
dosages in theaddition
accelerated actually increases the yield stress. This shows that lower
compositions.
dosagesForshould be used
the U series, theataccelerator
later addition times,actually
addition or else the yield stress
increases is higher
the yield stress.than
Thisfor early that
shows hydration
lower
times.
dosages The reasonbe
should theused
accelerator
at lateraddition
addition increases
times, or yield stress
else the in the U
yield mix while
stress decreasing
is higher than for it in the
early
Shydration
mix is thattimes.
the A accelerator does not include any superplasticizer in this
The reason the accelerator addition increases yield stress in the U mix while case. Therefore, the addition
of the aluminium
decreasing it in thesulfate
S mix isis that
creating
the Aadditional
acceleratorsurfaces
does not(most
include likely
anyettringite) that consume
superplasticizer the
in this case.
superplasticizer.
Therefore, the addition Thus, the increase
of the in thesulfate
aluminium effective volume fraction
is creating additional causes yield(most
surfaces stresslikely
to increase.
ettringite)
that consume the superplasticizer. Thus, the increase in the effective volume fraction causes yield
stress to increase.

3.2.2. Penetration Resistance/Strength Evolution

After ensuring that we could have fixed yield stress after acceleration, and a consistent evolution
during slipping, we are interested in knowing the rate of evolution of this yield stress. There are two
reasons for this. First, as the yield stress increases with time after acceleration, there is a distribution
Materials 2020, 13, 2084 11 of 24

3.2.2. Penetration Resistance/Strength Evolution

After ensuring that we could have fixed yield stress after acceleration, and a consistent evolution
during slipping, we are interested in knowing the rate of evolution of this yield stress. There are two
reasons for this. First, as the yield stress increases with time after acceleration, there is a distribution
of yield stresses in the formwork. This means that the total force needed to pull up the formwork
must be determined from the integration of yield stress from the bottom to the top of the formwork.
The integration can be done over the known residence time of concrete in the formwork: the oldest
concrete being at the bottom and the youngest at the top (more information is given in Section 4.1).
The second reason for our interest in the rate of yield stress growth is that it can tell us if the concrete
continues to gain enough strength to carry the additional weight of the object being produced.
We gain insight into these questions by examining penetration resistance measurements, which
provide crucially important information for SDC as well as for other digital fabrication processes [30].
Our results mainly show that provided the post-acceleration yield stress is kept constant (as described
in the previous section) the yield stress evolution is very similar regardless of the time of acceleration.
We also observed that the rate of yield stress change after acceleration can be quite different between
the different cases. Importantly, we see that the mixes S_C and U_A provide the least scatter and
therefore offer very appealing solutions for a smooth production control by SDC.
In the case of the C accelerator, it should be noted that the manufacturer defines maximum dosages
that ought not to be exceeded. This puts a limit on the rate of strength gain and therefore also on that
of SDC production. As an example, for 5 h production time, with a 30 cm high formwork and a mix
that requires 60 min to reach the process window, the slipping speed can be calculated as 5 mm/min.
This allows for a maximum 150 cm slipping path during the experiment, instead of the desired 300 cm
that would allow for more meaningful construction scale components.
As the A accelerator is a solution containing a significant amount of water, the theoretical upper
limit of the dosage leads to the highest possible w/b ratio allowed by the building code. By increasing
the concentration of the accelerator to a stable suspension, full-scale components could be produced
with several meters per hour of vertical building rate in reasonable times [26].

3.2.3. Cement Hydration

The calorimetry results of the accelerated samples in the first hour all show an increased heat release
that we can relate to rapid precipitation of hydration products following the addition of accelerator
(more so with aluminium sulfate). This does not directly translate to yield stress increase because,
as noted above, in the S concrete compositions additional water and superplasticizer adsorption
is expected.
As already noted in Section 3.1.3, all three S_C samples (Figure 7) show signs of sulfate
depletion [31]. This means that the silicate hydration remains quite delayed and that the sample
will have an intermediate strength for an extended period of time. This may limit the maximum
stable height of SDC objects. Excessively delayed silicate hydration probably also makes the finished
elements produced with the S_C mix more susceptible to drying shrinkage. During production and
curing afterwards without formwork may lead to loss of water before most of the C3S has hydrated.
In contrast, the aluminium sulfate type of acceleration, either with the ‘S mix’ (Figure 8) or with the
‘U mix’ (Figure A1), shows no flash set or sulfate depletion. One reason is that this type of accelerator
contains aluminium and sulfates in the ratio needed to produce ettringite. The time when the main
silicate peak (S_A) or the merged main hydration peak (U_A) occurs largely depends on the accelerator
dosage and it is discussed in another paper [26].
Materials 2020, 13, 2084 12 of 24

4. Part 2: Friction Tests

4.1. Results
In what follows, the results of the friction tests are reported as pull-out stress versus displacement.
The pull-out stress is calculated using the contact surface of the concrete with the immersed strip at
each point of the measurement.
Results for S_C, S_A and U_A with an accelerator addition 1 h after initial mixing are reported
in Figure 10. The figure on the left shows the situation in which the box is filled in one go by the
concrete, while for the figure on the right the concrete was added gradually in small layers prepared
individually
Materials 2020,one
13, x after the REVIEW
FOR PEER other to mimic the continuous SDC process. 12 of 24

Figure 10. Pull-out stress of the different compositions recorded in friction tests. (Left) Concrete is cast
Figure 10. Pull-out stress of the different compositions recorded in friction tests. (Left) Concrete is cast
in one go. (Right) Gradual filling mimicking SDC process.
in one go. (Right) Gradual filling mimicking SDC process.

Figure 10 shows that the values for direct casting are higher than for progressive addition. This is
Figure 10 shows that the values for direct casting are higher than for progressive addition. This
because
is becausein the latter
in the case,
latter less
case, time
less timehashaselapsed
elapsed since
since acceleration
acceleration ininthe
theupper
upper layers
layers thanthanforfor
thethe
concrete in the base than when the concrete is filled all at once. Apart from these
concrete in the base than when the concrete is filled all at once. Apart from these differences, we note differences, we note
that
thatininallallcases
casesthethestresses
stressesare are initially
initially higher
higher andand then
thendecrease
decreasetowards
towardsa aplateau.
plateau. Thus,
Thus, eacheach
of of
these
these tests are characterised by a peak stress and a plateau stress taken as the average of stress for for
tests are characterised by a peak stress and a plateau stress taken as the average of stress
displacements
displacementsbetween between15 15and
and 20
20 mm,
mm, when these these stresses
stressesarearefound
foundtotobeberoughly
roughly constant.
constant.
Interestingly,
Interestingly,the theranking
rankingof of plateau stresses isisnot
plateau stresses notthethesame
samedepending
depending onon
thethe filling
filling modemode (see(see
Figure
Figure 10).
10).ForForthe
thedirect
directfilling,
filling, the highest
highest peak
peakand andplateau
plateaustresses
stresses are
are found
found forfor
S_C,S_C, followed
followed by by
U_Aand
U_A andthenthenS_A.
S_A.For Forthethe progressive
progressive filling,
filling,the
thehighest
higheststresses
stressesare inin
are U_A.
U_A.However,
However, its its
peakpeakis is
lower
lower thanthat
than thatofofthe
theS_A.
S_A.
ToTo determine
determine thethe relation
relation ofof these
these friction
friction measurementstotoyield
measurements yield stress,
stress, wewe first
first report
report them
them with
with respect to the penetration resistance determined for the material with
respect to the penetration resistance determined for the material with the same activation time as the same activation time
asone
the the one
presentpresent at the
at the bottom
bottom of of
thethe box.This
box. Thisisisdone
done for
for the
the peak
peakandandthe
theplateau
plateau pull-out
pull-out stresses
stresses
respectively in the left and right parts of Figure 11. Independently of the concrete
respectively in the left and right parts of Figure 11. Independently of the concrete type or the filling type or the filling
mode,both
mode, bothdatadataseries
seriescancan bebe well
well fitted
fitted with
with aa line
lineforced
forcedthrough
throughthe theorigin. This
origin. This is important
is important for for
our paper since it establishes the possibility that the friction force, represented here by the pull-out
our paper since it establishes the possibility that the friction force, represented here by the pull-out
force, is proportional to the yield stress found at the bottom of the SDC formwork.
force, is proportional to the yield stress found at the bottom of the SDC formwork.
To more directly relate the friction measurements to yield stress, and to account for the yield
stress variation throughout the gradually placed samples, we computed average yield stress from
penetrometer measurements done on samples representing the concrete at the top and bottom of the
box τ0,h7 and τ0,h1 in Figure 12.
Plotting the peak and plateau pull-out stress versus the average yield stress, we again see a linear
relation (see Figure 13). However, the rate should depend not only on the material composition but
also on the filling method. For the gradual filling, the average penetration resistance was obtained by
measuring the penetration resistance of the material equivalent to both the first and the last batch filled

Figure 11. Relation of pull-out stresses to penetration resistance of the material at the bottom of the
friction box. (Left) Peak pull-out stress, (right) plateau pull-out stress. Both graphs show the three
 Figure 10). For the direct filling, the highest peak and plateau stresses are found for S_C, followed by
U_A and then S_A. For the progressive filling, the highest stresses are in U_A. However, its peak is
lower than that of the S_A.
To determine the relation of these friction measurements to yield stress, we first report them
with respect to the penetration resistance determined for the material with the same activation time
Materials one13,
as the2020, 2084 at the bottom of the box. This is done for the peak and the plateau pull-out stresses
present 13 of 24

respectively in the left and right parts of Figure 11. Independently of the concrete type or the filling
mode, both data series can be well fitted with a line forced through the origin. This is important for
into the box. For the same composition, the friction is 2–3 times lower when the container is filled
our paper since it establishes the possibility that the friction force, represented here by the pull-out
gradually compared to filling it at once.
force, is proportional to the yield stress found at the bottom of the SDC formwork.

Materials 2020, 13, x FOR PEER REVIEW 13 of 24

Figure
Figure Relation
11.11. Relationofofpull-out
pull-outstresses
stressesto topenetration resistanceof
penetration resistance ofthe
thematerial
materialatatthe
thebottom
bottom of of
thethe
friction box.
friction box.(Left)
(Left)Peak
Peakpull-out
pull-outstress,
stress, (right)
(right) plateau pull-out stress.
plateau pull-out stress.Both
Bothgraphs
graphsshow show thethe three
three
Materials
concrete2020,compositions
concrete 13, x FOR PEERand
compositions REVIEW
andthe
thetwo
twofilling
fillingmodes.
modes. 13 of 24

To more directly relate the friction measurements to yield stress, and to account for the yield
stress variation throughout the gradually placed samples, we computed average yield stress from
penetrometer measurements done on samples representing the concrete at the top and bottom of the
box 𝜏0,ℎ7 𝑎𝑛𝑑 𝜏0,ℎ1 in Figure 12.

Figure 12. Yield stress of the concrete filled into the container at once (left) and gradually (right).

Plotting the peak and plateau pull-out stress versus the average yield stress, we again see a linear
relation (see Figure 13). However, the rate should depend not only on the material composition but
also on the filling method. For the gradual filling, the average penetration resistance was obtained by
measuring the penetration resistance of the material equivalent to both the first and the last batch
filledFigure
into the box.
Yield
12.12.
Figure
Forstress
theofof
Yieldstress
same
the composition,
theconcrete
concretefilled
thethe
filled into
friction
into the
is 2–3
container
container
times
atonce
at once lower
(left)
(left)
when the(right).
andgradually
and graduallycontainer
(right). is
filled gradually compared to filling it at once.
Plotting the peak and plateau pull-out stress versus the average yield stress, we again see a linear
relation (see Figure 13). However, the rate should depend not only on the material composition but
also on the filling method. For the gradual filling, the average penetration resistance was obtained by
measuring the penetration resistance of the material equivalent to both the first and the last batch
filled into the box. For the same composition, the friction is 2–3 times lower when the container is
filled gradually compared to filling it at once.

Figure 13.13.Relation
Figure Relationofofpull-out
pull-out stresses
stresses to average yieldstress
average yield stressmeasured
measuredbybythethepenetrometer.
penetrometer. (Left)
(Left)
peak pull-out
peak pull-outstress,
stress,(right)
(right)plateau
plateau pull-out stress. Both
pull-out stress. Bothgraphs
graphsshow
showthe
the three
three concrete
concrete compositions
compositions
and
and the
the twotwofilling
fillingmodes.
modes.

4.2. Discussion
During SDC, the upper limit of the process window is related to the friction that results from the
sum Figure
of the 13.
spatially
Relationvariable shear
of pull-out stress
stresses along yield
to average the surface of the formwork.
stress measured The friction
by the penetrometer. (Left) tests
peak pull-out
presented in this stress,
paper (right)
were plateau
designedpull-out stress. Both
to determine graphslaws
friction showthat
the three
mayconcrete
be usedcompositions
to describe this
andas
process the twoas
well filling modes.
the parameters that they contain.
In our experiment, the surface in contact with the formwork was expected to be a cement paste,
4.2. Discussion
which allows the formation of a lubrication layer along the wall. This explains why the pull-out
Materials 2020, 13, 2084 14 of 24

4.2. Discussion
During SDC, the upper limit of the process window is related to the friction that results from
the sum of the spatially variable shear stress along the surface of the formwork. The friction tests
presented in this paper were designed to determine friction laws that may be used to describe this
process as well as the parameters that they contain.
In our experiment, the surface in contact with the formwork was expected to be a cement paste,
which allows the formation of a lubrication layer along the wall. This explains why the pull-out
stresses that we measure are substantially lower than the average yield stress of the concrete. Values
are in fact roughly an order of magnitude lower, which corresponds to the range of yield stresses of the
cement paste.
Our results show that the friction force depends more or less linearly on the yield stress and
that the frictional coefficients depend on the mix design, accelerator and filling mode. Qualitatively,
it is understood that the gradual filling mode reduces the overall friction force due to the gradient
of material age in the box with lower average age than with simple filling. However, the fact that
the filling mode inverses the ranking of friction forces is less trivial. This result implies that the rate
of structural build-up plays an important role in this process, and that SDC may only be reliably
simulated when this is accounted for. Specifically, it is important to obtain trustworthy measurements
of both the friction coefficient and the structural build-up kinetic law.
All friction experiments (Figure 10) show a peak stress at the start of the pull-out with a higher
resistance than later when a plateau stress is ultimately reached. Consequently, the initiation of slipping
is substantially more difficult than its continuation, thus intermediate pauses in slipping should be
avoided, as had already been recognized in the Ph.D. Thesis of Lloret [9]. Results from this work
underline that this comes from a difference between static and dynamic friction. As both increase
with hydration, the need to avoid stop-go is not caused by hydration but is in the very nature of the
difference between both those friction modes. Our results also express this by demonstrating the
process window for SDC, this implies that the process window at the start of slipping is narrower than
it is for the running process.
The underlying reason for the higher initial stress is a structural breakdown of the interface
material [32]. In partial support of this argument, it is observed that the factor between the initial peak
pull-out stresses and the plateau stresses are similar in all tests: 3.6–4.8 for direct filling and 2.2–3.0 for
gradual filling (see Table A2 in Appendix A). The fact that the stress reaches a plateau implies that the
interfacial material reaches a steady state of structural breakdown at some point.

5. Part 3: A Theoretical Model for the SDC Process Window

5.1. Reformulating the Process Window with the SDC Number’

The slip-forming process model is based on the force balance in the formwork during slipping
(see Figure 14). SDC can be understood as gravitational extrusion through a formwork-size nozzle
where the weight of the concrete in the formwork is the driving force that has to overcome the friction
acting along the formwork walls. Thus, the ratio of the surface and the volume in the formwork plays
a significant role in successful slipping and will be referred to as hydrodynamic radius rhy = 2·Volume
Sur f ace .
 For the dimensions used for different SDC cases, we find the following values of 𝑟ℎ𝑦 :
 Column with a square cross-section of 150 mm side length: 𝑟ℎ𝑦 = 75 𝑚𝑚 [9];
 Column with a circular cross-section of 212 mm diameter: 𝑟ℎ𝑦 = 106 𝑚𝑚 [9,10,13];
 Column with an elliptical cross-section of 125 mm and 80 mm radii: 𝑟ℎ𝑦 = 96 𝑚𝑚 [9];
 2020,
Materials Façade mullion with rectangular cross section of 100 mm and 70 mm: 𝑟ℎ𝑦 = 41 𝑚𝑚 [14];
13, 2084 15 of 24
 Folded structures with 25 mm thin 500 mm long sides: 𝑟ℎ𝑦 = 25 𝑚𝑚 [11].

14. Force
Figure 14.
Figure Forcebalance
balanceduring
duringslipping with
slipping a formwork
with for producing
a formwork thin folded
for producing structures
thin folded [11,12].
structures
[11,12].
For the dimensions used for different SDC cases, we find the following values of rhy :

• Column with a square cross-section of 150 mm side length: rhy = 75 mm [9];

• Column with a circular cross-section of 212 mm diameter: rhy = 106 mm [9,10,13];
• Column with an elliptical cross-section of 125 mm and 80 mm radii: rhy = 96 mm [9];
• Façade mullion with rectangular cross section of 100 mm and 70 mm: rhy = 41 mm [14];
• Folded structures with 25 mm thin 500 mm long sides: rhy = 25 mm [11].

The vertical stress at the extrusion point can be expressed as the difference between the hydrostatic
pressure ρgH of concrete in the formwork, expressed with the formwork of height H, and the frictional
stress τFr acting along the formwork walls (see Figure 14). As τFr varies along with the formwork
height, we write:
Z H
2
σVE = ρgH − τFr (z)dz. (1)
rhy 0
This process is constrained by two failure types: the tear-off and the flow-out [10]. First, concerning
the tear-off, the vertical stress σVE has to be positive. Second, to avoid flow-out, the Tesca criterion
must be satisfied, which implies that σVE must be lower or equal to twice the yield stress of the concrete
leaving the formwork τ0 (tExtr ). Both of these conditions can be summarized as:

0 ≤ σVE /2 ≤ τ0 (tExtr ). (2)

From Section 3 and in particular Figures 4–6, the yield stress can be considered to evolve as a
power-law, scaling to the time t after acceleration. Further, from Section 4 and in particular Figure 12,
the friction stress can be considered as proportional to the yield stress:

τ0 (t) = αC tβC (3)

τFr (t) = αFr τ0 (t) = αFr αC tβC . (4)

where αFr is a friction coefficient, αC is the prefactor of the power law for yield stress effective and βC is
its exponent.
For a constant vertical building rate V, we can rewrite this in relation to any position z along with
the formwork height:
 βC
z
τFr (z) = αFr αC . (5)
V
Materials 2020, 13, 2084 16 of 24

Substituting this into (1) we get:

H  βC
2 αFr αC
Z
2 z
σVE = ρgH − αFr αC dz = ρgH − HβC +1 . (6)
rhy 0 V rhy (1 + βC )V βC

Equation (5) can also be used to write conditions at the level of extrusion. There, the paste is the
oldest in the formwork and the yield stress at the time of extrusion τ0, Extr can be written as:
βC
H

τ0, Extr = αC , (7)
V

substituting this equation into (6) gives:

2 αFr τ0, Extr

σVE = ρgH − H (8)
rhy (1 + βC )

which in turn can be substituted into the expression for the operating window given in (2):

HαFr ρgH HαFr

≤ ≤ 1+ . (9)
rhy (1 + βC ) 2 τ0,Extr rhy (1 + βC )

As proposed by Szabo et al [11], this process window can be more conveniently formulated with
Hα
a dimensionless parameter ξ given by r (1+Frβ ) , giving:
hy C

ρgH
ξ≤ ≤ 1 + ξ. (10)
2 τ0,Extr

In this paper, we use a slightly different notation, by introducing a dimensionless yield stress:
τ0,Extr
T0,Extr = , (11)
ρgH

and a dimensionless number that we call the SDC number, given by:

1
SDC = . (12)
2ξ

The meaning of this number is best understood by rewriting it as:

( 1 + βC )
SDC = . (13)
rhy αFr
H

By doing this, the numerator depends only on strength gain, which is captured by the exponent
of the yield stress growth βC . It can be understood to represent a characteristic cohesion gain. In the
denominator, the term rH represents the ratio between the height of fresh concrete over a characteristic
hy
thickness of the formwork section. It can be thought of as a friction equivalent of a slenderness
ratio, whereby high values of the ratio lead to high friction and vice versa. Together with the friction
coefficient αFr , this implies that the denominator relates to a characteristic friction of the system. All in
all, it implies that the dimensionless SDC number expresses a ratio between a characteristic cohesion
and a characteristic friction.
With this in mind, we are now in a better position to discuss the process window that we rewrite as:

SDC
≤ T0,Extr ≤ SDC. (14)
2 SDC + 1
Materials 2020, 13, 2084 17 of 24

This indicates that for the process to work, the dimensionless yield stress should be lower than the
SDC number. Otherwise, the material is too cohesive, friction is too high and there will be a pull-off.
SDC
It also indicates that to avoid flow out, the dimensionless yield stress should be larger than 2 SDC +1 .
For large SDC numbers, where cohesion gain is fast and/or friction is low, the dimensionless yield
ρgH
stress should be larger than 12 (meaning τ0,Extr > 2 based on Equation (11)).
With respect to the main objective of this paper, which is to examine the feasibility of producing
thin folded elements by SDC, the effect of a change in hydrodynamic radius Rhy can be quantified if
the other parameters remain unchanged:

SDC1 r
Rrhy = = 1. (15)
SDC2 r2

In the case of formworks for thin folded geometries, the increase in surface area compared to the
change in volume results in a hydrodynamic radius three times smaller than the one previously used
in SDC formworks for columns. Thus, the process window for slipping is three times narrower for
thin folded structures than for columns. With the narrow process window for thin folded forms, it is
therefore even more important to have consistent yield stress evolution over time than it is for columns.

5.2. Practical Considerations

Friction force measurements from Section 4 and Equation (4) make it possible to determine αFr .
However, we should recall that the friction is substantially higher at the start of slipping and then
decreases. Thus, we describe each material sample with two αFr parameters, reported in Figure 15
along with their standard deviation. In this regard, we note that the αFr parameters show only a limited
variability within a given data series, and this most likely originates from limitations in the friction
measurement
Materials 2020, 13,setup.
x FOR PEER REVIEW 17 of 24

Figure 15. Frictional parameter calculated for each mix and filling method at the start and during
Figure 15.The
slipping. Frictional parameter
error bars representcalculated for deviation
the standard each mixof and
thefilling method atmeasured
αFr parameters the startover
and time
during
on
slipping. The error bars represent
the different friction strips. the standard deviation of the 𝛼 𝐹𝑟 parameters measured over time
on the different friction strips.
With such estimates of αFr , the boundaries of the process window for each material can be
With
determined.suchToestimates 𝛼𝐹𝑟we
illustrateofthis, , theconsider
boundaries of theformwork
a typical process window
for thinfor each geometries
folded material can be
with
determined.
rhy = 0.025mToandillustrate
H = 0.3m. this, we consider
Moreover, a typical
we use of βC determined
formwork
the values for thin folded geometries
for our mixes𝑟ℎ𝑦
differentwith =
(see
0.025𝑚 and 𝐻 = 0.3𝑚. Moreover, we use the values of
Table A3 in the Appendix A). First, we determine the dimensionless 𝛽 𝐶 determined for our different mixes
yield stress from Equation (11). (see
Table
The A3 in the
operation Appendix).
window is thenFirst, we determine
expressed the dimensionless
in the series of plots showing yield
the stress from Equation
dimensionless (11).
yield stress
The operation window SDCis then expressed in the series of plots showing the dimensionless yield stress
between the ratio 2 SDC+1 and SDC. Values are reported in Figure 16 with the example of ‘S_C_1’ and
𝑆𝐷𝐶
betweengradual’
‘U_A_1 the ratiofor both values of αFr
and SDC. Values are reported
(the plots with the in Figure
other 16 and
mixes withfilling
the example
methods ofcan
‘S_C_1’ and
be found
2 𝑆𝐷𝐶+1
‘U_A_1 gradual’ for both values of 𝛼𝐹𝑟 (the plots with the other mixes and filling methods canand
in the Appendix A in Figures A2–A5). The operation window is shown to be narrower at the start be
found in the Appendix in Figures A2–A5). The operation window is shown to be narrower at the
start and wider during slipping for all material and filling types. Further, S_C_1 mixes filled either
all at once or gradually provide the tightest process window of the three compositions, which makes
it the least favourable mix design for producing thin folded structures. U_A_1 filled gradually
provides the best conditions for formworks for thin folded structures based on the width of its process
determined. To illustrate this, we consider a typical formwork for thin folded geometries with 𝑟ℎ𝑦 =
0.025𝑚 and 𝐻 = 0.3𝑚. Moreover, we use the values of 𝛽𝐶 determined for our different mixes (see
Table A3 in the Appendix). First, we determine the dimensionless yield stress from Equation (11).
The operation window is then expressed in the series of plots showing the dimensionless yield stress
𝑆𝐷𝐶
between the ratio and SDC. Values are reported in Figure 16 with the example of ‘S_C_1’ and
Materials 2020, 13, 20842 𝑆𝐷𝐶+1 18 of 24
‘U_A_1 gradual’ for both values of 𝛼𝐹𝑟 (the plots with the other mixes and filling methods can be
found in the Appendix in Figures A2–A5). The operation window is shown to be narrower at the
widerand
start during
wider slipping
duringfor all material
slipping and
for all filling types.
material Further,
and filling S_C_1
types. mixesS_C_1
Further, filled either
mixes all at once
filled or
either
gradually provide the tightest process window of the three compositions, which makes
all at once or gradually provide the tightest process window of the three compositions, which makes it the least
itfavourable
the least mix design for
favourable mix producing thinproducing
design for folded structures. U_A_1
thin folded filled gradually
structures. U_A_1 provides the best
filled gradually
conditions
provides thefor formworks
best conditionsforfor
thin folded structures
formworks based structures
for thin folded on the width of its
based onprocess
the widthwindow for the
of its process
start-up and continuous slipping.
window for the start-up and continuous slipping.

Figure 16.
Figure Comparisonof
16. Comparison ofmeasured
measuredextrusion
extrusion yield
yield stress
stress shown
shown with
with full
full circles
circles for
for ‘S_C_1’
‘S_C_1’ material
material
(left) and with empty squares for ‘U_A_1 gradual’ material (right), along with the operation
(left) and with empty squares for ‘U_A_1 gradual’ material (right), along with the operation window window
SDC
between 2 𝑆𝐷𝐶
between andSDC
SDC+1and SDCwith
withaaformwork
formworkforfor thin
thin folded
folded structures
structures (continuous
(continuous horizontal
horizontal lines
lines
2 𝑆𝐷𝐶+1
indicate the start-up and dashed the slipping).
indicate the start-up and dashed the slipping).

It can be observed that a minimum waiting time is needed before the start of slipping to avoid
It can be observed that a minimum waiting time is needed before the start of slipping to avoid
collapse, for example 55 min for ‘U_A_1 gradual’ (see Figure 16, right). A minimum waiting time
collapse, for example 55 min for ‘U_A_1 gradual’ (see Figure 16, right). A minimum waiting time also
also exists for S_C_1, but cannot be seen in the measured data. From Figure 16 (left) it can, however,
exists for S_C_1, but cannot be seen in the measured data. From Figure 16 (left) it can, however, be
be inferred that this value would lie around or slightly below 45 min. For this mix, we can additionally
inferred that this value would lie around or slightly below 45 min. For this mix, we can additionally
see in the same figure that the slipping should not start later than 70 min. For mix U_A_1, an upper
see in the same figure that the slipping should not start later than 70 min. For mix U_A_1, an upper
waiting time must also exist but is not observable in the data reported in Figure 16 (right). Concerning
mix S_C_1, it is important to note that the slipping initiation has a relatively narrow operation
window. However, once the slip-forming process has successfully begun, it should continue with fewer
problems due to the larger process window that is obtained after initiation owing to lower dynamic
than static friction.
A previous study [13] found αFr = 0.12 determined with friction measurements along a rectangular
formwork during continuous slipping with a similar mix composition as ‘S_C_1 gradual’ assuming
βC = 2, which is a typical yield stress evolution in such mixes. Compared to that report, here βC
is determined from penetration tests and αFr from friction tests where the integrated friction force
neglects confinement effects including thin walls or corners. Therefore, in this study αFr is significantly
lower, which causes the possibility of overestimating the upper bounds of the process window for
both the start and the slipping. The difference between these measurements emphasizes the effect of
formwork geometry. The process window could be refined numerically, however, it would require
separate computations for each different case. The confinement effects could also be considered by
determining the upper bound of the operation window for slipping as the same as that at the start.
This conservative approach provides a safety coefficient of about 3, which should largely compensate
for the higher friction in confined spaces. The effect of different formwork materials would require
further investigation to study the interaction between the fresh concrete and the pull-out strip.
In Figure 17, the effect of formwork geometry with different rhy shows a larger process window
for columns with larger rhy than for thin folded structures with smaller rhy . The slipping strategy can
be defined based on the narrowness of the process window. For folded structures (Figure 17, Left),
the slipping should start close to the lower bound of the operation window, because this can be well
defined. Indeed, the lower bound only implies material failure and flow out, so its definition is subject
to less uncertainty. By starting the system under such conditions, we make sure to keep it away from
the pull-out strip.
In Figure 17, the effect of formwork geometry with different 𝑟ℎ𝑦 shows a larger process window
for columns with larger 𝑟ℎ𝑦 than for thin folded structures with smaller 𝑟ℎ𝑦 . The slipping strategy
can be defined based on the narrowness of the process window. For folded structures (Figure 17,
Left), the slipping should start close to the lower bound of the operation window, because this can be
Materials 2020, 13, 2084 19 of 24
well defined. Indeed, the lower bound only implies material failure and flow out, so its definition is
subject to less uncertainty. By starting the system under such conditions, we make sure to keep it
“pull-off”
away from failure due tofailure
“pull-off” friction (upper
due bound(upper
to friction of the operation
bound of window). The window).
the operation “pull-off” regime for thin
The “pull-off”
folded structures
regime not only
for thin folded is uncertain
structures butisalso
not only starts very
uncertain low, starts
but also thus, the
veryrisk of thus,
low, goingthe
into it is
risk ofhigher
going
thanitinisthe
into case than
higher of columns.
in the case of columns.

SDC
𝑆𝐷𝐶
Figure 17.
Figure The process
17. The process window
window between
between 2 SDC andSDC
+1 and SDC for
for the
the start
start of
of slipping
slipping with
with aa typical
typical
2 𝑆𝐷𝐶+1
formwork for thin folded structures (left) and with a cylindrical formwork to produce columns (right).
formwork for thin folded structures (left) and with a cylindrical formwork to produce columns (light).
The red arrows highlight the slipping strategy for the start time and increased or decreased slipping
The red arrows highlight the slipping strategy for the start time and increased or decreased slipping
time for later.
time for later.
As an example, in Figure 17 (left), the starting point for slipping folded structures is shown with a
As an example, in Figure 17 (left), the starting point for slipping folded structures is shown with
triangle at 55 min marking the time interval the concrete needs to spend in the formwork before it slips
a triangle at 55 min marking the time interval the concrete needs to spend in the formwork before it
out. Provided the slipping is successfully initiated, the slipping rate could then be reduced and the
slips out. Provided the slipping is successfully initiated, the slipping rate could then be reduced and
acceleration time increased. In the same figure, this is indicated by the square at 65 min, with the arrow
the acceleration time increased. In the same figure, this is indicated by the square at 65 min, with the
highlighting the shifting of the operation condition. Indeed, it can also be seen on Figure 17 left that
arrow highlighting the shifting of the operation condition. Indeed, it can also be seen on Figure 17
once the slipping has started the “pull off” limit (upper bound) goes to much higher levels. Therefore,
left that once the slipping has started the “pull off” limit (upper bound) goes to much higher levels.
it becomes much safer to reduce the slipping speed, raising the operation conditions more distinctly
Therefore, it becomes much safer to reduce the slipping speed, raising the operation conditions more
above the lower bound and thus reducing the risk of material failure (flow out). There is, however,
a balance to find here since too much speed reduction will have a negative impact on production rates.
In contrast, in Figure 17 (right), for elements with much larger hydrodynamic radii, such as
column, the opposite procedure is probably best. Indeed, in this case the upper bound of the operation
window for slipping initiation is much higher. It is, therefore, safe to wait longer and stay away from
the lower bound. In Figure 17 (right) this is shown by the triangle at 75 min. In this case, once the
slipping has started it is advantageous to move faster, reducing the acceleration time. This should,
however, be done without moving too close to the lower bound. In Figure 17 (left) this is shown by
the square representing the same steady-state operation condition at 65 min (as in the previous case).
The arrow once again underlines the change in operation conditions, which in this case is advantageous
because it increases the production speed without compromising material stability too much.
Significantly, the most suitable mix for producing a desired geometry with a formwork of a specific
rhy is based not only on the extents of the process window, but also on the later strength evolution in
the section below the formwork. Otherwise, material failure or buckling may occur [2,4,6]. Results
presented in the calorimetry Section 3.1.3 show that the aluminate sulfate solution does not cause
sulfate depletion (the main hydration peaks are not inverted), and therefore represents a good option
to avoid unexpected delays of the tricalcium silicate hydration present in cement.

6. Conclusions
For thin folded geometries produced with smart dynamic casting, consistent strength evolution
and extended time during which the material meets the requirements of the narrow process window
are necessary. This work shows that these goals can be achieved with different set-on-demand material
systems with different base mixes and accelerators. By exchanging the initially used C100 accelerator
with aluminum sulfate, the limitations for dosage could be overcome due to the different mode of
Materials 2020, 13, 2084 20 of 24

action. Aluminum sulfate and its working mechanism are known in the concrete industry, as it is used
as one of the main components in various shotcrete accelerators [33]. However, its potential for digital
fabrication could be represented best as a set-on-demand system with controlled yield stress evolution
and a long open time that allows large-scale architectural elements to be produced.
Quantifying friction here provided another approach to assess different material systems.
This work shows that friction is proportional to yield stress, with a coefficient of proportionality
that depends on material composition and filling history. As friction is the highest in the beginning,
continuous slipping is safer than start-up. The theoretical framework suggests that materials with a
high exponent for strength evolution and a low frictional coefficient are beneficial. However, it also
shows that all the material systems investigated have advantages and drawbacks and must be selected
according to the component height and formwork hydrodynamic radius of the individual case.
As a result of this work, we propose a revised formulation of the operation window for SDC in
which a dimensionless SDC number is introduced. The dimensionless yield stress at the exit of the
SDC
formwork should be lower than this SDC number and larger than the ratio 2 SDC +1 , which for thin
1
folded structure has a limit value of 2 . The model presented slightly overestimates the upper bound
as it neglects confinement effects. This, together with the already rather narrow operation windows,
underscores the sensitivity of SDC for producing thin folded geometries. It further emphasizes the
importance of proper material design to achieve consistent yield stress evolution. The model presented
in this paper offers a first approximation and the friction tests proposed help to define and test
material requirements. More refined solutions will require either large scale SDC production testing,
computational simulations, or both.

Author Contributions: Conceptualization, methodology, formal analysis: A.S., L.R. and R.J.F.; investigation,
validation, data curation, visualization, writing—original draft preparation: A.S.; writing—review and editing:
L.R., E.L.-F., F.G., M.K. and R.J.F.; project administration: A.S. and E.L.-F.; supervision and funding acquisition:
F.G., M.K. and R.J.F. All authors have read and agreed to the published version of the manuscript.
Funding: The research is pursued in the interdisciplinary framework of the National Competence Centre of
Research (NCCR) Digital Fabrication funded by the SNSF at ETH Zürich.
Acknowledgments: The authors thank Heinz Richner and Andi Reusser for technical assistance, and Lukas Esser
and Pascal Küng for their work on the friction test setup during their Bachelor thesis.
Conflicts of Interest: The authors declare no conflict of interest. The funders had no role in the design of the
study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to
publish the results.

Appendix A
Table A1. Spread flow measurements of the different retarded and accelerated compositions.

Mix for Uniform Mix for Adapted t d1 d2 h

Base Mix τ0 (Pa)
Acceleration Acceleration (hh:mm) (cm) (cm) (cm)
S 16.2 16.5 2.1 128
S_C_1 S_C_1 01:00 26 25.6 13
S_C_1 03:15 23.5 24.5 19
S_C_1 05:05 21.8 22.2 29
S_C_3 03:10 26.4 26.9 11
S_C_5 05:00 29.6 30.1 6.3
S_A* 01:05 13.2 12.8 3.8 403
S_A_1 S_A_1 01:00 24.2 23.9 19
S_A_1 03:05 19.7 20.3 47
S_A_1 05:05 19.1 19.7 54
S_A_3 03:00 24.3 24.4 17
S_A_5 05:00 25.3 25 15
U 21.2 21.7 31
U_A_1 U_A_1 01:00 34.7 35.7 2.6
U_A_1 03:05 33.5 35.2 3
U_A_1 05:05 30 30.7 5.5
U_A_3 03:00 33.4 36.7 2.7
U_A_5 05:00 32.5 34.1 3.5
Materials 2020, 13, 2084 21 of 24

Table A2. The average ratio of peak and plateau stresses measured on the strips of each mix composition.

Mix Peak/Plateau
S_C_1 3.57
S_A_1 3.74
U_A_1 4.77
S_C_1 gradual 3.04
S_A_1 gradual 2.27
U_A_1 gradual 2.18

Table A3. Process window boundaries expressed in yield stress and dimensionless yield stress according
to Equations (9) and (14) for thin folded structures.

T0 Lower Limit (-) T0 Upper Limit (-)

αFr (-)
rhy (m) H (m) β (-) SDC SDC/(2*SDC+1)
Start-Up Slipping Start-Up Slipping Start-Up Slipping
Folded Structures
S_C_1 0.025 0.3 2.9 0.2 0.1 1.8 6.3 0.4 0.5
S_A_1 0.025 0.3 1.5 0.1 0.0 3.8 14.0 0.4 0.5
Materials 2020,
U_A_113, x FOR PEER
0.025 REVIEW
0.3 2.5 0.1 0.0 2.7 10.5 0.4 21 of 24
0.5
S_C_1 gradual 0.025 0.3 2.9 0.1 0.0 4.0 10.4 0.4 0.5
S_A_1 gradual 0.025 0.3 1.6 0.1 0.0 4.0 8.3 0.4 0.5
S_A_1 gradual 0.025 0.3 1.6 0.1 0.0 4.0 8.3 0.4 0.5
U_A_1 gradual 0.025 0.3 2.4 0.1 0.0 4.4 8.7 0.4 0.5
U_A_1 gradual 0.025 0.3 2.4 0.1 0.0 4.4 8.7 0.4 0.5
Mullions
Mullions
S_C_1 0.075
S_C_1 0.075 0.40.4 2.9 2.9
0.2 0.2
0.1 0.1
3.9 3.9
14.214.2 0.40.4 0.50.5
S_A_1 0.075 0.4 1.5 0.1 0.0 8.4 31.4 0.5 0.5
S_A_1
U_A_1 0.075
0.075 0.40.4 1.5 2.5 0.1 0.1 0.0 0.0 8.4 6.2 31.423.5 0.50.5 0.50.5
U_A_1
S_C_1 gradual 0.075
0.075 0.40.4 2.5 2.9 0.1 0.1 0.0 0.0 6.2 9.0 23.523.4 0.50.5 0.50.5
S_C_1
S_A_1 gradual
gradual 0.075
0.075 0.40.4 2.9 1.6 0.1 0.1 0.0 0.0 9.0 9.1 23.418.6 0.50.5 0.50.5
U_A_1 gradual
S_A_1 gradual 0.075
0.075 0.40.4 1.6 2.4 0.1 0.1 0.0 0.0 9.1 10.0 18.619.5 0.50.5 0.50.5
U_A_1 gradual
Columns 0.075 0.4 2.4 0.1 0.0 10.0 19.5 0.5 0.5
Columns
S_C_1 0.106 0.4 2.9 0.2 0.1 5.6 20.1 0.5 0.5
S_A_1 0.106 0.4 1.5 0.1 0.0 11.9 44.4 0.5 0.5
U_A_1 0.106 0.4 2.5 0.1 0.0 8.7 33.3 0.5 0.5
S_C_1 gradual 0.106 0.4 2.9 0.1 0.0 12.7 33.1 0.5 0.5
S_A_1 gradual 0.106 0.4 1.6 0.1 0.0 12.9 26.3 0.5 0.5
U_A_1 gradual 0.106 0.4 2.4 0.1 0.0 14.1 27.6 0.5 0.5

Heat flow
Figure A1. Heat flow over
over time
time from time of acceleration for the U mix accelerated with uniform
dosage of aluminium sulfate.
 Figure
Materials 2020,A1. Heat flow over time from time of acceleration for the U mix accelerated with uniform
13, 2084 22 of 24
dosage of aluminium sulfate.

Figure A2. The

The process
process window
window expressed
expressed with
with the
the dimensionless
dimensionless yield stress for the start of slipping
slipping
Materials 2020, 13, x FOR PEER REVIEW 22 of 24
Materials
with 2020, 13, x FOR
a typical PEER REVIEW
formwork structures when
for thin folded structures when ‘S_A_1’
‘S_A_1’ material
material is
is used
used for
for the process. 22 of 24
the process.

Figure A3. The process window expressed with the dimensionless yield stress for the start of slipping
Figure A3. The
The process
process window
window expressed with the dimensionless yield stress for the start of slipping
with a typical formwork for thin folded structures when ‘U_A_1’ material is used for the process.
thin folded
with a typical formwork for thin folded structures
structures when
when ‘U_A_1’
‘U_A_1’ material
material is
is used
usedfor
for the
theprocess.
process.

Figure A4. The process window expressed with the dimensionless yield stress for the start of slipping
Figure A4. The process window expressed with the dimensionless yield stress for the start of slipping
Figure
with A4. The process
a formwork window
for thin folded expressed
structures with
whenthe dimensionless
‘S_C_1 yield stress
gradual’ material for the
is used for start of slipping
the process.
with a formwork for thin folded structures when ‘S_C_1 gradual’ material is used for the process.
with a formwork for thin folded structures when ‘S_C_1 gradual’ material is used for the process.
 Figure A4. The process window expressed with the dimensionless yield stress for the start of slipping
Materials 2020, 13, 2084 23 of 24
with a formwork for thin folded structures when ‘S_C_1 gradual’ material is used for the process.

The process
Figure A5. The process window
window expressed
expressed with
with the
the dimensionless
dimensionless yield stress for the start of slipping
slipping
with aatypical
typicalformwork
formwork forfor
thin folded
thin structures
folded when
structures ‘S_A_1
when gradual’
‘S_A_1 material
gradual’ is usedis
material forused
the process.
for the
process.
References
1. Buswell, R.A.; Leal de Silva, W.R.; Jones, S.Z.; Dirrenberger, J. 3D printing using concrete extrusion:
A roadmap for research. Cem. Concr. Res. 2018, 112, 37–49. [CrossRef]
2. Reiter, L.; Wangler, T.; Roussel, N.; Flatt, R.J. The role of early age structural build-up in digital fabrication
with concrete. Cem. Concr. Res. 2018, 112, 86–95. [CrossRef]
3. Hanehara, S.; Yamada, K. Rheology and early age properties of cement systems. Cem. Concr. Res. 2008, 38,
175–195. [CrossRef]
4. Wolfs, R.J.M.; Bos, F.P.; Salet, T.A.M. Early age mechanical behaviour of 3D printed concrete: Numerical
modelling and experimental testing. Cem. Concr. Res. 2018, 106, 103–116. [CrossRef]
5. Juilland, P. Early Hydration of Cementitious Systems; EPFL: Lausanne, Switzerland, 2009.
6. Roussel, N. Rheological requirements for printable concretes. Cem. Concr. Res. 2018, 112, 76–85. [CrossRef]
7. Feys, D.; Cepuritis, R.; Jacobsen, S.; Lesage, K.; Secrieru, E.; Yahia, A. Measuring Rheological Properties of
Cement Pastes: Most common Techniques, Procedures and Challenges. RILEM Tech. Lett. 2017, 2, 129–135.
[CrossRef]
8. Wangler, T.; Lloret, E.; Reiter, L.; Hack, N.; Gramazio, F.; Kohler, M.; Bernhard, M.; Dillenburger, B.; Buchli, J.;
Roussel, N.; et al. Digital Concrete: Opportunities and Challenges. RILEM Tech. Lett. 2016, 1, 67. [CrossRef]
9. Lloret Fritschi, E. Smart Dynamic Casting—A Digital Fabrication Method for Non-Standard Concrete Structures;
ETH: Zurich, Switzerland, 2016.
10. Lloret, E.; Shahab, A.R.; Linus, M.; Flatt, R.J.; Gramazio, F.; Kohler, M.; Langenberg, S. Complex concrete
structures: Merging existing casting techniques with digital fabrication. CAD Comput. Aided Des. 2015, 60,
40–49. [CrossRef]
11. Szabo, A.; Reiter, L.; Lloret-Fritschi, E.; Gramazio, F.; Kohler, M.; Flatt, R.J. Adapting Smart Dynamic Casting
to Thin Folded Geometries. In Proceedings of the First RILEM International Conference on Concrete and Digital
Fabrication—Digital Concrete 2018; Wangler, T., Flatt, R.J., Eds.; Springer International Publishing: Cham,
Switzerland, 2019; pp. 81–93.
12. Szabo, A.; Lloret-Fritschi, E.; Reiter, L.; Gramazio, F.; Kohler, M.; Flatt, R.J. Revisiting Folded Forms with
Digital Fabrication. In Proceedings of the eCAADe SIGraDi 2019, Architecture in the Age of the 4th Industrial
Revolution; Sousa, J.P., Castro Henriques, G., Xavier, J.P., Eds.; FAUP: Porto, Portugal, 2019; Volume 2,
pp. 191–200.
13. Lloret Fritschi, E.; Kohler, M.; Reiter, L.; Gramazio, F.; Flatt, R.J.; Wangler, T. Smart Dynamic Casting: Slipforming
with Flexible Formwork—Inline Measurement and Control; HPC/CIC: Tromsø, Norway, 2017.
Materials 2020, 13, 2084 24 of 24

14. Lloret-Fritschi, E.; Scotto, F.; Gramazio, F.; Kohler, M.; Graser, K.; Wangler, T.; Reiter, L.; Flatt, R.J.;
Mata-Falcón, J. Challenges of Real-Scale Production with Smart Dynamic Casting. In First RILEM International
Conference on Concrete and Digital Fabrication—Digital Concrete 2018; Wangler, T., Flatt, R.J., Eds.; Springer
International Publishing: Cham, Switzerland, 2019; Volume 19, pp. 299–310. ISBN 978-3-319-99518-2.
15. Scotto, F.; Lloret-Fritschi, E.; Gramazio, F.; Kohler, M.; Flatt, R.J. Adaptive Control System for Smart
Dynamic Casting. In Proceedings of Learning, Adapting and Prototyping-Proceedings of the 23rd CAADRIA
Conference, Beijing, China, 17–19 May 2018.
16. Lloret Fritschi, E.; Reiter, L.; Wangler, T.; Gramazio, F.; Kohler, M.; Flatt, R.J. Smart dynamic casting:
Slipforming with flexible formwork—inline measurement and control. HPCCIC Tromsø 2017, 2017, 27.
17. Reiter, L.; Palacios, M.; Wangler, T.; Flatt, R.J. Putting Concrete to Sleep and Waking It Up with Chemical
Admixtures. In Proceedings of the 11th International Conference on Superplasticizers and Other Chemical
Admixtures in Concrete, Ottawa, ON, Canada, 10–13 July 2015; Volume 302, pp. 145–154.
18. Le, T.T.; Austin, S.A.; Lim, S.; Buswell, R.A.; Gibb, A.G.F.; Thorpe, T. Mix design and fresh properties for
high-performance printing concrete. Mater. Struct. Constr. 2012, 45, 1221–1232. [CrossRef]
19. Agustí-Juan, I.; Müller, F.; Hack, N.; Wangler, T.; Habert, G. Potential benefits of digital fabrication for
complex structures: Environmental assessment of a robotically fabricated concrete wall. J. Clean. Prod. 2017,
154, 330–340.
20. Engel, H.; Rapson, R. Tragsysteme—Structure Systems; Hatje Cantz Verlag: Berlin, Germany, 2007; ISBN
978-3-3-7757-1876-9.
21. SikaRapid ®C-100. Available online: https://che.sika.com/dms/getdocument.get/2c0a5740-fc68-389d-acb5-
944e817deb27/441388_CH_D.PDF (accessed on 9 April 2020).
22. Hajiesmaeili, A.; Denarié, E. Next Generation UHPFRC for Sustainable Structural Applications. 2018.
Available online: http://infoscience.epfl.ch/record/253042 (accessed on 9 April 2020).
23. Wetzel, A.; Glotzbach, C.; Maryamh, K.; Middendorf, B. Microstructural investigations on the skinning of
ultra-high performance concrete. Cem. Concr. Compos. 2015, 57, 27–33.
24. Roussel, N.; Coussot, P. “Fifty-cent rheometer” for yield stress measurements: From slump to spreading flow.
J. Rheol. 2005, 49, 705–718. [CrossRef]
25. Mettler, L.K.; Wittel, F.K.; Flatt, R.J.; Herrmann, H.J. Evolution of strength and failure of SCC during early
hydration. Cem. Concr. Res. 2016, 89, 288–296. [CrossRef]
26. Szabo, A.; Reiter, L.; Lloret-Fritschi, E.; Gramazio, F.; Kohler, M.; Flatt, R.J. Processing of Set on Demand
Solutions for Digital Fabrication in Architecture. In Rheology and Processing of Construction Materials; Springer:
Cham, France, 2020; pp. 440–447.
27. Lootens, D.; Jousset, P.; Martinie, L.; Roussel, N.; Flatt, R.J. Yield stress during setting of cement pastes from
penetration tests. Cem. Concr. Res. 2009, 39, 401–408.
28. Mantellato, S. Flow Loss in Superplasticized Cement Pastes; ETH: Zürich, Switzerland, 2017.
29. Mantellato, S.; Palacios, M.; Flatt, R.J. Relating early hydration, specific surface and flow loss of cement
pastes. Mater. Struct. 2019, 52, 5. [CrossRef]
30. Reiter, L.; Wangler, T.; Anton, A.-M.; Flatt, R.J. Setting on Demand for Digital Concrete—Principles,
Measurements, Chemistry, Validation. In Cement and Concrete Research; Elsevier: Oxford, UK, 2020.
31. Tenoutasse, N. The hydration mechanism of C3A and C3S in the presence of calcium chloride and calcium
sulphate. Proc. Int. Symp. Chem. Cem. 1968, 2, 372–378.
32. Roussel, N. Rheology of fresh concrete: From measurements to predictions of casting processes. Mater. Struct.
2007, 40, 1001–1012. [CrossRef]
33. Aïtcin, P.-C. Accelerators. In Science and Technology of Concrete Admixtures; Woodhead Publishing: Cambridge,
UK, 2016; pp. 405–413. ISBN 978-0-08-100693-1.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).