Materials and Structures / Matrriaux et Constructions, Vol.

36, November 2003, pp 609-620

~ RILEMTC 162-TDF:'Testand designmethodsfor steel fibre reinforcedconcrete'

Round-robin analysis of the R I L E M TC 162-TDF beam-

bending test: Part 1 - Test method evaluation

Prepared by B.I.G. Barr ~and M.K. Lee ~,with contributions from E.J. de Place Hansen 3,
D. Dupont 4, E. Erdem s, S. Schaerlaekens 2, B. Schntitgen 5, H. Stang 3 and L. Vandewalle 4
(1) CardiffUniversity, Wales
(2) Belgian Research Institute, Belgium
(3) Technical University of Demnark, Denmark
(4) Katholieke Universiteit Leuven, Belgium
(5) Ruhr-University of Bochum, Germany

ABSTRACT RESUMs
A round robin test programme was carried out on the Un programrne d'e.~ais comparatifi" entre laboratoires a dtO
beam-bending test recommended by the RILEM TC 162- rdalisdpour le test de poutre soumise h laflexion, commepreser#
TDF [1 ]. Plain concrete and steel fibre reinfbrced concrete par TC 162-TDF de la RILEM. Des poutres en b~ton normal et en
(SFRC) beams were included in the test programme. The b~ton de fibres m~talliqu~s" (BFM) sont inclues dans le
material variables for the SFRC beams consisted of two programme d'es~i~. Pcatr les poutres en BFM, on a con~Mdrd
concrete strengths, three fibre dosages and three types of comrne variables" de mat&'iaux: deux rdsi~tanc~es gl la
fibres. A comprehensive statistical analysis was carried out compression, tro~ dosages" atefibres et trois (vpes de fibres. Une
to determine the applicability and robustness of the test anaO,se stat~tique exkxtustive a Ot~ erAcmOel~mr determiner la
method. It was tbund that although inter-lab variations do pertinence c4 Ia solMitd de la m~thode d'essai. On a constatO que,
occur, this was relatively small compared to the inherent bien que des variations de r~,~ltats entre les laboratoires existent,
material variation. It is also possible that the high variations celles~'i sont relativement petites; surtout lorsque l',on compare
observed could be due to the relatively small cross sections avec lea' :mriations intrim~es des mat&iawc eux-m~mes. Ces
used for the test beams. Additionally, an investigation was derni~es variations peuvent ~re ~galement &4es cnec sections
carried out to evaluate the objectiveness of the calculation trans~ersales relativement petites qui sont utilis.~espour les essais
procedure proposed by RILEM TC 162-TDF to obtain the de poutre. Ensuite, une analyse a ~td rdalisOe en vu d'&,a.hter
necessary design parameters. It was found that the l'objectbeit~ de la mOthode de calcut q~d est proposd par TC 162-
prescribed calculation procedure was satisfactory, as the TDF de la R1LE]ff pour obtenir les paramfitres n~cessaires lots
variation between the design parameters calculated at du dimemionnement. On a cons,taM que la m~thode de calcul
different laboratories was generally within the range of prescrite" dtait satisftisante, parce que la variation des
• An alternative method of obtaining the design param~tres; calcut~e par les dij]~rents k~oratoires pour le
parameters, by considering residual strengths, is suggested dimensionnement; Otait d'un ordre de grandeur de +_5%. Une
as it simplifies the calculation procedure and the test mdthode alternatfi~ est ndanmoins proposOe pout" obtenir les
method. In general, the beam test was found to be a good para,'n~es de" dimem'ionnement en prenant en comid~ration la
robust test and relatively easy to carry out.. r~sistance rds'iduelle. Cette m~thode sinwlifie la mdttuMe de cak~l
et la m~thode d'essai. En gOndral, l 'essai de poutre soumise h la
flexion simple, comme pre~crit par le TC 162-TDF de la RILE?r
est un essai robuste et relativement simple ?texdcuter.

1359-5997/03 9 RILEM 609

TC 162-TDF

1. I N T R O D U C T I O N the second phase where a wider range of materials would be

tested. For the first phase, UvWC cast all of the specimens
In the beginning of 1999, a research eonsortitml cortsisting involved. For the second phase, three other labs i.e. CSTC,
of eleven academic and industrial partners was created DTU and RUB prepared the specimens. Table 1 gives the
tbllowing an award of a Contract under the Industrial and overall round robin test programme.
Materials Technologies Programme (Brite-EuRam HI) to carry
out a project entitled "Test and Design methods ;for Steel Fibre
3. M A T E R I A L S A N D S P E C I M E N
Reintbrced Concrete (Design of SFRC)". One of the objectives
of the project was to investigate appropriate test methods to PREPARATION
characterise the basic material properties of SFRC materials.
As part of the effort to fulfil that objective, a round robin test
3.1 T y p e o f fibres
programme was carried out on two test configurations i.e.
notched beams under three-point bending and uni-axial tensile Three types of fibres, supplied by N.V. Bekaert S.A.,
test on cylinders. This paper presents the results for the round were used for the round robin test programme. All the
robin programme carried out on the three-point beam-bending fibres were hooked and collated. Details of their usage can
test. The set-up chosen for evaluation was based upon the be seen from Table 1 (from the superscripts). Details of the
recommendations of the RILEM TC 162-TDF [1] with some fibres are given in Table 2.
minor modifications as detailed later. The main aim was to
investigate the strengths, limitations and appropriateness of the 3.2 F i r s t p h a s e
proposals. UWC cast all of the specimens, which were NSC, {br the
Two concrete strengths were considered in the test first phase. The mix composition, based upon earlier work
programme. The concrete grades considered were C25/30, at UWC [2], of cement: fine aggregate: coarse aggregate:
termed as normal strength concrete (NSC) and C70/85, termed water is equal to 1: 2: 2.5:0.56 by weight.
as high strength cono~ete (tlSC). The letter "C" indicates a For the first phase, control specimens in the form of three
characteristic compressive strength and the numerals indicate 150rmn cubes and three cylinders (~bl00x200 mm long) were
the compressive strengths in N/mm2 measured fi'om cylinders cast along with the actual beam specimens (150x 150x600 ram)
and cubes respectively. Three different fibre dosages were (N.B. ~qe recommended PdLEM beam has dimensions of
employed in the test programme i.e. 25 kg/m3, 50 kg/m 3 and 150x150x550mm as s_hown in Fig. 1), Two of the control
75 kg/m3. In addition, plain concrete beams were also included specimens were retained at UWC, whereas the third was
in the test progranmqe. The plain concrete beanas essentially transported with the beams to the other testing labs. All
play the role of control specimens and were a means of specimens were cured under water at a temperature of 20-t2~
investigating the strengths, limitations and
sensitivity of the proposed test :method, as
Table 1 - Overall round robin test programme
they do not contain variations introduced
by fibre distribution and orientation.
Number of specimens
Five testing laboratories were
involved in this part of the round robin Testing First phase Second phase
programme, with Cardiff University lab C25/30 C25/30 C70/80
(UWC) as the task co-ordinator. The
laboratories are (in alphabetical order): 0 kg/m 3 501 kg/m 3 252 kg/m3 752 kg/m3 0 kg/m3 253 kg/m3
CSTC - Belgian Building Research CSTC 8 8 6 6 6 6
Institute DTU 8 8 6 6 6 6
D T U - Technical University of Derunark KUL 8 8 6 6 6 6
K U L - Katholieke Universiteit Leuven RUB 8 8 6 6 6 6
RUB - Ruhr-University of Bochum UWC 8 8 6 6 6 6
UWC - Cardiff' University (task co- Total 40 40 30 30 30 30
ordinator)
The above acron3qns for each Note: Superscript indicates Dramix 80/60 BN was used
Superscript 2 indicates Dramix 65/60 BN was used
participating laboratory will be used
Superscript 3 indicates Dramix 80/60 BP was used
henceforth. During the statistical analysis,
each individual laboratory was assigned a
nmnber (known only to the task co-ordinator) to maintain the
Table 2 - Characteristics of the fibres used
objectiveness of the statistical study.
Dramix Dramix Dramix
Property 65(60 BN 80/60 BN 80/60 BP
2. T E S T P R O G R A M M E Carbon
content Low Low High
Tensile
It was decided at an early stage that it would be wise to strength 1000 1000 2500
divide the foreseen round robin test progranune into two (MPa)
phases. The first phase was considered a pilot phase to Length (mm) 60 60 60
investigate the strengtks and weaknesses of the proposed beam Aspect ratio
test. Subsequently, this valuable experience would be fed into (Length / 65 80 80
Diameter)

610
Materials and Structures /MatOriaux et Constructions, Vol. 36, November 2003

At 21 days, half of the beam specimens from each mix

were taken out of the water and notched. The other half
were left in the water as they were to be notched at their
~F
respective testing labs at approximately 7 weeks. After DEVICE
notching, the specimens were immediately placed back
under water fbr further ctuing. Subsequently, at 28 days, all ' ~ ~
the specimens (whether notched or un-notched) were
removed from the water and wrapped in good quality cling
PPORT
film. They were then stored in a curing room (RH 65J:5%,
T=20~:2~ Approximately one week before testing, the
specimens were taken out of the curing room and sent to
their respective testing labs. One contoct hos to be f~xed
(two degrees of freedom)
3.3 Second phase
For the second phase, three other laboratories were
responsible for specimen preparation. RUB prepared the A

specimens with 75kg/m 3 of fibres, whereas DTU and CSTC

prepared the specimens with 25kg/m 3 o f fibres. CSTC also
fabricated the plain HSC test specimens.
DTU prepared the NSC with 25kg/m 3 of fibres using a 6 j ~ ~1 m
mix composition of 1: 2.07: 2.52: 0.51. RUB prepared the
NSC with 75kg/m 3 of fibres using a mix composition of 1: I-
IT ~ 1"1 L 50 I
3.89: 3.05:0.65 (NSC mix composition given in the fbrm of
cement: fine aggregate: coarse aggregate: water). CSTC, on
the other hand, cast the plain HSC and the HSC specimens
with 25kg/m 3 of steel fibres using a mix composition of 1:
1.93: 2.66: 0.0075:0.35 (cement: tine aggregate: coarse
aggregate: silica fume: water). In addition, 7.75 litre/m3 o f
super-plasticizer was added. Ftc - a '1 cr..o.~
For the second phase, a similar procedure to the first phase
was implemented. However, a slightly different approach
was used regarding notching of the specimens. All the - \ gscas!om~
specimens were notched at their respective testing labs just
before testing took place i.e. 7 weeks aider casting. The Fig. 1 - Schematic diagram showing the beam bending test set-
preparing labs also prepared control specimens, but these up (after [1]).
were not transported but tested at the preparing lab itself.
relatively shallow notch. Furthermore, HSC beams were
3.4 Notching procedure involved in the test programme which could result in unstable
For both the first and second phases, all beams were cut cracking for plain concrete specimens. By conU'olling the tests
via rotating diamond blades to maintain a ligament height via CMOD control a more stable test could be a,ttained.
o f 125mm. Whilst the notches were cut, a jet of water was
used to cool the surfaces during the notching procedure. All 4.2 Strain rates
notches were cut perpendicular to the top surthce at casting. In the first phase of the rotmd robin test programme, the
CMOD strain rate was not stated explicitly. Instead, the
RILEM recommendations were used as a rough guide fior the
4. TEST SET-UP SFRC specimens. Therefbre, it was fotmd that the variou~s
testing labs used different strain rates. Some of the labs used
4.1 Description of test set-up varying strain rates whereas the others used a constant strain
rate. Tables 3 and 4 give the strain rates adopted for the plain
The test set-up used is illustrated in Fig. 1 with some slight
I Table 3 - Strain rate regimes used in the first
modifications from the PdLEM recommendations. All tests
were conducted under closed looped conditions. The
Constant/ "CMOD at which Strain
measurements which were taken during testing were the load Testing
Varying strain rate was rate
applied, mid-span displacements on both sides of the beam (~St lab
strain rate applied (Era) ...... (#m/min)
and 82) and the crack mouth opening displacement (CMOD). 9 0 to 100 9 20
In the RILEM recommendations, it is stated that the tests 9 100 to 250 9 50
should be conducted in a manner such that the average mid-span CSTC Varying 9 250 to 500 9 250
deflection increases at a constant rate of 0.2 mrrgmin. A slight 9 500 to end of test 9 500
modification was made with respect to this requirement. Instead DTU Constant Whole of test 120
of carrying out the tests under average mid-span deflection KUL Constant Whole of test 50
control, the tests were carried out under CMOD control. It was * 0to 200'...... *30
UWC Varying 9 200 to end of test 9 60
anticipated that the beams would be stiff due to their size and the

611
TC 162-TDF

concrete and SFRC beams during the first phase of the round
Table 4 - Strain rate regimes used in the first
robin programme respectively.
phase (.for the SFRC beams)
Constant/ CMOD at which In the second phase, it was agreed that a more stringent
Testing Strain rate test characterisation in terms o f the strain rate should be
Varying strain rate was
lab (~tm/min) employed for both the plain and SFRC beam specimcms.
strain rate applied (~tm)
* 0 to 100 9 20 These rates are given in Table 5.
e 100 to 250 ,,, 50
CSTC Varying 9 250 to 500 9 250
e 500 to 4000 9 500
5. RESULTS AND DISCUSSION
,4000 t o end o f t e s t 9 2000
DTU Constant Whole of test 250
KUL Constant Whole of test 200
5.1 Control specimens
RUB Constant Whole of test 250 Table 6 shows the strength and density test results tbr the
9 0 to 200 * 30 control specimens. In general, the variability for the control
UWC Varying
9 200 to end of test * 240 specimens is low thus indicating that the specimen
preparation procedure was satisfactory.

Table 5 - Strain rate regimes used in the second 5.2 Typical results
phase
From the raw data obtained from the round robin test
Type of concrete CMOD strain rates
programme, the load-average mid-span deflection (P-6) curves
Plain 50 ~tm/min until rupture
and the load-CMOD (P-CMOD) curves can be obtained.
9 200 gm/min until CMOD reaches 3.5 mm.
Steel fibre Typical P-5 and P-CMOD curves for the plain NSC and HSC
9 After CMOD of 3.5 mm, speed can
reinforced concrete are given in Figs. 2 and 3 respectively. On the other
optionally be increased to 2000 ~tm/min
hand, Figs. 4 and 5 show typical plots o f P-5 and P-CMOD Ibr
25 25-

20 "~*NSC 20-
-4-HSC

15 t5-
Z Z
"13
t@
,a 10,

--~- C25/30(25) " -

-.~- C25J3f~50)
- a - C25130(75)
-,i~ C70/85(25)
0 0
0 0,1 0,2 0,3 0,4 0,5 0,6 o 0,5 1 1,5 2 2.5
Average mid-span deflection, ;5 (ram) Average mid-span deflection, ;5 (ram)

Fig. 2 - Typical P-6 curves for the plain concrete beams. Fig. 4 - Typical P-6 curves for the SFRC beams.
25

20 "-,,~NSC 20
",m-HSC

Z
!

-~.- C25t30(25>
,.~-C25t30(50)
,.~-C25/30(75)
-~*C70185(25)

O 0.1 0,2 0,3 0.4 0,5 0.6 o o.5 1 ~5 2 Z5 3

Crack mouth opening displacement, CMOD (ram) Crack mouth ~ n l n g d t s p l a c e ~ , CMOD (mm)

Fig, 3 - Typical P-CMOD curves for the plain concrete beams, Fig, 5 - Typical P-CMOD curves for the SFRC beams.

612
Materials and Structures / Matdriaux et Constructions, Vol. 36, November 2003

Table 6 - Results for the control s p e c i m e n s the SFRC beam specimens. It can be observed that the
Indirect P-5 and P-CMOD curves have shnailar shapes. The
Fibre Density Cube tensile
Cylinder
relationship between 6 and CMOD has been
Concrete strength strength
dosage P strength investigated in detail during the course of the round
grade
(kg/m~ (k~/m3) (N/ram2) f, robin testing programme [3, 4]. However, this
(N/mm2) (N/am5
relationship will not be presented here, as it is reported
C25/30 0 2367 44.5 3.89 35.3 in detail in a companion paper.
(0.45) (8.58) (8.77) (8.68)
C25/30 25~ 2400 N/A N/A 39.5 5.3 Intra-lab variation
(0.45) (3.56)
C25/30 50~ 2370 42.6 N/A N/A In the investigation of intra-lab variation, the
(0.88) (10:28) coefficient of variation, V, of the load measured at
C25/30 75~ 2407 45.0 N/A N/A several mid-span deflections for each testing lamb is
(0.55)
,, (2.50) calculated. This has been done for all types of concrete
C70/85 0 2395 91.8 N/A N/A beams tested for all the testing labs.
(0,66).......
(4.69) Figs, 6 and 7 show a plot o f V of the load at
C70/85 253 2401 89.6 N/A ...... N/A
increasing values of mid-span deflection for the
(0.32) (2.45)
NSC and HSC beams respectively. For plain
Note: Superscript l indicates Dramix 80/60 BN was used
Superscript 2 indicates Dramix 65/60 BN was used concrete, the most important value is the maximum
Superscript 3 indicates Dramix 80/60 BP was used load and the coefficient of variation for this value is
Coefficients of variation, V, given in brackets (%) within 10% indicating that the test set-ups are
N/A - not available robust where intra-lab testing is concerned.
It may be observed that the variation increases with
5O
increasing mid-span deflection. For the NSC, the overall V
g * Lab 1
> D increases from 7.87% to 33.2% tbr the range considcq~xl.
O Lab 2
Similarly, tbr the HSC, V inc~.ases ti,om 9.81% to 32.8%. This
40 tn increasing value of V can be explained by the thct that, once
m
D & Lab 3 well into the post-w-~_k region tbr plain concrele beams, the

~ 30 Variation for
peak loads Qr
~o
X Lab 5
applied load is very, small. Cortsequ~tly, a small change in
applied load will result in a relatively large value for V. The
"6 range of variation in the NSC beam_s seems to be much larger
A - O -Overall
/ A
x
than Oat observ'ed tbr the HSC beams. This is because for the
20 O /~ & & X
NSC beams, the beams received by each lab come from two
x x mixes, each contributing four beams. On the other hand, for the
t~ '
HSC beams, each testing lab received 6 beam,s, all coming from
different mLxes. Thus, the probability of a laboratory receiving a
set of beams of a particular nature (be it higher or lower in
x strength) is increased in the case of the plain NSC beams.
o
o Figs. 8 to 11 show similar plots of V for the load
o 0.1 0,2 0.3 0,4 0,5 0,6 measurements at increasing mid-span deflections for the
Average mid-span deflection, <5(mm) SFRC beams. In general, it is clear that V for the SFRC
beams is relatively higher in comparison with the plain
Fig, 6 - Intra-lab variation for plain NSC beams. concrete beams. Unlike the plain concrete beams, the
5o overall V does not increase significantly as the mid-
g deflection increases. This is because, in contrast with the
> plain concrete beams, the load resisted by the SFRC beams
& are generally much higher than those resisted by the plain
o 40
concrete beams and thus the error introduced will be
correspondingly lower. The overall V is the highest tbr the
m Variation for /O beams with a fibre content of 25 kg/m 3 regardless of
30 3eak loads /
E / concrete strength. The lowest overall V is recorded for the
II/
"Z u beams with a fibre content of 75 kg/m 3.

20
* | * Lab1 Fig. 12 shows the variation of loads measured at specific
_.0. / * * mid-span deflections for all the SF'RC types involved in the
[ ] Lab 2 round robin test programme. It is apparent that the
"6 /o &
& Lab 3
variability of the low fibre content beams is significantly

i]"
o higher compared to the beams with 75 kg/m 3. The variation
- O - Overall of the beams with 50 kg/m 3 fibre content is in between
8 these two extremes. It is interesting to note that both the
0 NSC and HSC beams with low fibre content of 25 kg/m 3
0 0,1 0,2 0.3 0,4 0,5 0.6 exhibit variation of a similar order. This would indicate that
Average mid-span deflection, 8 (mml the fibre content is the dominant factor in determining the
variation in the beam test results. As the fibre content
Fig. 7 - Intra-lab variation for plain HSC beams.

613
T C 162-TDF

40 40

X X > o o
X •
<> s o
0
30 X 30 0
.....O .............O ...............O ............ O • •
......... A A A A A

J E O Q
A
O
1~, A
~ ZX

20 20
c A
O
"C D
> O Lab 1 I
O Lab 1
"6 O
10 O Lab 2 10 n Lab2 I
A Lab 3
o A Lab 3 I
x Lab 4
x Lab 5 • Lab 4
O ....o Overall O

0.5 ~ ~.5 2 2.5 3 3.5 0 0,5 1 t.5 2 2.5 3 35

Average mid-span deflection, 5 (mm) Average mid-span deflection, S ( m m )

Fig. 8 - lntra-lab variation fbr the C25/30 (with 25 kg/m 3 Fig. 11 - lntra~lab variation for the C70/85 (with 25 kgfm3
of Dramix 65/60 BN fibres) SFRC beams. of Dramix 80/60 BP fibres) SFRC beams.
40. 40
@ Lab 1
El Lab 2
& Lab3
X Lab 4 X X X X
~ 30. Lab 5 ~ x x
--@- Overall

i g -I! - -It - -tl

o
E o [3 o D o o
~ 20. z "~ 20
0
A A
@ 0 0 I~, II, I> '~ A A A A

~0 10 o C25/30{25)
9 o C25/3o(5o)
A C25/30(75)
O X C70/85(25)

0.5 ~ ~.5 2 2.5 a 3.5 0.5 1 1.5 2 2,5 3 3.5

Average mid-span deflection, 6 (ram) Average mid-span deflection, ~5( r a m )

Fig. 9 - Intra-lab variation for the C25/30 (with 50 kg/m 3 Fig. t2 - Compilation of overall variation for all the SFRC
of Dramix 65/60 BN fibres) SFRC beams. beam specimens.
30 accommodate is approached, the probability of achieving a
more uniform mix is increased. This would reduce the
>
variability in the fibre dis~bution and thus test results with
_o lower variability can be expected.
"o
o r O O In general, for the NSC and HSC beams with the lowest
20 x
x X fibre content of 25 kg/m 3, the overall variation is in the
region of 24% to 38%. The variation within each lab on the
"s .......0 ....... [] 0 ........ ....0D .........
0'" A A ~ other hand can vary from approximately 12% to 35%.
.j 13
B A As the fibre content is increased, the variation is
X generally decreased. For the NSC beams with 50 kg/m 3 of
A •
O Lab 1 fibres, the overall variation is approximately 23% for the
D Lab 2 displacements considered. The variation within each lab
A Lab 3
x Lab 4
ranges approximately in the order of 13% to 25%.
,o Overall For the NSC beams with the highest fibre content of
0 75 kg/m 3, the overall average variation is very much
0.5 1 1.5 2 2.5 3 3,5 improved. Generally, the overall variation is in the region
Average mid-span deflection, ~ ( m m ) of 15% to 17% for the loads considered. The variations
Fig. 10 - Intra-lab variation for the C25/30 (with 75 kg/m 3 within labs vary from 10% to 23%. This is a significant
of Dramix 65/60 BN fibres) SFRC beams. improvement in relation to the beams with the lowest fibre
content of 25 kg/m 3.
increases, the variability is reduced. This is because as the In summaxy, if a 10% variation level is considered, the
limit of the amount of fibre that the concrete mix can variations observed are quite high for all the concrete types

614
Materials and Structures / MatOriaux et Constructions, Vol. 36, November 2003

considered within the round robin li'~unework. The next two us to pin-point the sources o f the diffbrences. Similar to the
sections will try to identit~r the source or sources of variation ANOVA test, the post-hoe tests relies on the t~...... statistical
observed within the test results. In particular, attention will be tables. The Scheff6 test differs from the Bonferroni tests in
given to the inter-lab and within-mix variations. that it is more conservative in the sense that the probability
of a Type I error is less.
5.4 Inter-lab variation
It is acknowledged here that there are several limitations
In any test programme conducted at different places, to the statistical tests carried out. The statistical tests
there are bound to be differences introduced by the test set- assume that the samples come from an identical source (a
ups used. However, more importantly, one would be population) and thus any significant difference must be
interested to know the extent and severity of this variability. caused only by the difference in treatment, the treatment in
To determine whether there was any significant this case being the test set-up. However, in the case of the
difference in the test results due to any variation in the round robin test programme, samples were mixed in
different lab set-ups, statistical analyses were carried out batches. Moreover, no consideration has been given to
using the plain concrete beam specimens. In the round variation introduced by transportation of tile concrete
robin test programme, two different types of plain concrete cylinders, notching procedure etc. A complete
beams were tested; one was the NSC, C25/30, and the other consideration of all these other factors is not possible.
was the HSC, C70/85. The plain concrete beam results were Detailed mathematical basis of these methods can be
used for this part of the statistical study because this would obtained fi'om books on statistics [5-7].
remove any variation due to the introduction of fibres into Table 7 shows the results of the statistical analysis of the
the concrete matrix. The analysis of variation (commonly test results. A ratio of more than unity for tTt~ntis obtained tbr
knows as ANOVA) and post-hoc tests (two such tests were two values of ma_ximum load i.e. tbr both the NSC and HSC
used i.e. the Scheff6 and the Bonferroni tests) were carried beams. As the maximum load is the most important
out in this investigation. characterising paxameter for plain concrete, this indicates that
The ANOVA method is able to indicate whether there are inter-lab variations do occur. However, when considering the
any significant differences in the test results at a particular post-peak regime, all the f / ~ t values fbr the loads at the
confidence level. An f-value can be computed using the prescribed 6 points considered arc below the critical level.
different sets of results. This t=value is then compared with the Table 8 shows the results of the post-hoc tests conducted.
~t-value obtained ii-om statistical tables. These statistical The table shows which labs differ significantly from each
tables are given in terms of f~,~v. The value of ot here other at a level of et = 0.05. Both the Scheft? and Bonferroni
represents the level of confidence, whereas u and v are the tests indicate that there is a significant difi%rence in the value
degrees of fi~edom associated with the test results. The of Pm,~ for the plain NSC beams mad the post-hoe tests
degrees of freedom can be said to be factors which take into indicate that the significant difference is dominated by Lab 5.
account the number of treatments and samples. Normally, in Apart from these results, no other significant difference is
statistical inferences, a value of a=0.05 is adopted. This value detected. It is interesting to note that the post-hoc tests do not
of ct has been used in this study. The value of ~ is the identify any significant difference for the value of Pm~x for
probability of making a Type I (or a error) ie. when we accept the plain HSC beam results due to the ratio f/l~rit being very
that there is a difference when in reality there is none. A ratio close to the value o f unity.
off/f.ri, larger than unity would indicate the null hypothesis (i. e. Upon further investigation, it was found that Labs 1 to 3
that there is no significant difference) has to be rejected at the had average P ~ values of 11.4 ix.) 11.8 kN whereas Lab 5 had
particular level ofct. an average value of 13.5 kN. The specimens tested by Lab 5
In addition to ANOVA, two other statistical tests were were from the first mix and there was a tendency for the mix to
carried out. These tests were the Bont~rroni and Scheffd be a little dry thereby causing an increase in strength [8].
tests, collectively known as post-hoc tests. ANOVA can However, this increase in strength is not significant in absolute
give an indication of the existence of significant differences terms i.e. the average "value of the maximum load tbr Lab 5
between various sets of results but it does not tell us where was found to be approximately 2 kN (~15% increase) more
the difference or differences occur. Post-hoc tests enables than the other testing lab averages. This difference shows up in

Table 7 - Compilation of A N O V A analysis results for plain concrete beams

Concrete Load, P, at prescribed average ~5values ,,, Area
grade Statistical
under the
(Fibre parameters Po.1 P0a5 P0.2 Poa5 V0.3 P0.4 P-5 curve
dosage)
u 3 3 3 3 3 3 3 3
v 17 17 17 17 17 17 17 t4
r2 f-ratio 7.56 3. t 4 2.43 1.95 1.46 1.30 1.42 0.851
eq
fcrit 3.20 3.20 3.20 3.20 3.20 3.20 3.20 3.34
fffcrit 2.36 0.981 0.760 0.609 0,455 0.405 0.443 0.255
u 2 2 2 2 2
tt3
v II 11 11 11 11 ll 11 9
O0
f-ratio 4.08 3.48 2.05 1.32 129 1.51 2.95 3.07
t~ f-crit 3.98 3.98 3.98 3.98 3.98 3.98 3.98 4.46
t7f~t 1.02 0.873 0.516 t 0.333 0.325 0.379 0.74i 0.690
Note: Figures in bold indicate tTt~itratios greater than unity
P• indicates the load at the prescribed 6 value ofx mm

615
TC 162-TDF

the variability is in the material

Table 8 - Compilation of post-hoe test results on plain concrete beams due to fibre variation at the
Concrete Post-hoe Load, P, at prescribed average ~ values Area failure section. Another
grade (Fibre under the
test Pmax Poa Poas Po.2 Po.2s Po.3 Po.4 P-~5curve possibility is that the test
dosage)
specimens have relatively small
1-5
Scheff6 2-5 * * * * * * cross sections. Consequently, a
3-5 small variation or difference in
r
tt~ 1-5 the nmnber o f :fibres has a direct
Bonferroni 2-5 * * * * * * and a relatively large influence
3-5 on the toughness of the material
tested. This would be more
Scheff~ * * * * * * * pronounced in specimens with
low fibre dosages.
p..
L) Bonferroni * * * * * * *
5.5 Within-mix
Note: 1-5 means significant difference observed for results between labs 1 and 5 variation
* means no statistical difference identified
Another source o f variation
the inter-lab variation statistical analysis because the is the inherent variations
coefficient o f variation o f the maximum load is low and existing within any one mix. There were too many
subsequently magnifies the impact. If Lab 5 is disregarded for specimens involved within the round robin test programme
the plain NSC beams, it is obvious that the inter-lab difference "that they could not be prepared in one go. Instead, they had
in the maximum load is not as significant as betbre (averages to be fabricated in batches i.e. several concrete mixes were
o f 11.4 to 11.8 kN). prepared. To investigate the inherent material variation, the
Therefore, it is safe to assume that difl'erences in the test results were further considered by sepaxating the restdts
set-ups are minimal in the sense that it should not adversely according to the mix from which they were made. This,
affect the beam test results to the point where the variation however, was only done for the NSC beams and the NSC
due to different set-ups overshadow the variation due to the beams with 50 kg/m 3 o f fibres i.e. specimens from the first
material itself phase o f the round robin programme. To effectively
With respect to the SFRC beams, four types o f concrete were investigate the effects o f within-mix variation, it is logical
tested: the NSC beams with fibre dosages o f 25, 50 and that only the first phase results be used, as all the beam
75 kg/m 3, together with HSC beams with a single fibre dosage specimens fabricated from any particular mix were sent to
o f 25 kg/m 3. Tables 9 to 12 show a compilation o f the ANOVA one testing lab only (and not distributed amongst the testing
and post-hoc analysis results. Significant differences between labs). During the first phase, each testing lab received eight
the testing labs are detected in many o f the values considered for beams, with four beams coming from one mix. A similar
the NSC beams with fibre dosages of 25 and 50 kg/m3. Tables analysis could not be carried out on the second phase
10 and 12 show the results o f the post-hoe tests. Like the specimens because in the second phase, each testing lab
ANOVA technique, Table 9 - Compilation of A N O V A analysis results for SFRC beams
many differences are
observed for the above Concrete Statistical Load, P, at prescribed average 5 values Area to
grade 07ibre Pmax
mentioned concrete dosage) parameters P0.5 PL0 PLS P~.o P2.5 P3.O 8 = 3mm
types, u 4 4 4 4 4' '4' 4 4
There seems to be a ~ v 23 22 22 22 22 22 21 22
systematic introduction ~ f-ratio 6.97 2.73 4.13 5,03 5,42 6.12 7.37 3.83
o f variation due to the ~ bcritical 2.80 2.82 2.82 2.82 2.82 2.82 2.84 2.82
results from Lab 4 for ~ f/s 2.49 0.969 1.46 1.79 1.92 2.17 2.59 1.36
the C25/30 with 25 ,.-, u 4 4 4 4 4 4 3 3
O
kg/m 3 of fibres. ~ v 29 29 30 29 28 27 24 24
However, this is not a ~ f-ratio 2.62 1.96 2.59 2.91 3.15 3.53 3.78 3.80
cause for concern, as ~ f-critical 2.70 2.70 2.70 2.70 2.71 2.73 3.01 3.01
Lab 4 does not appear f/refit 0.969 0.726 0.957 1.08 1.16 1.29 1.26 1.26
in the post-hoc analysis ~" u 3 3 3 3 3 3 3 3
t-- v 15 15 15 15 15 14 13 13
for the other SFRC
types. Theretbre, there r f-ratio 0.735 0.682 0.457 0.509 0.566 i .04 0.958 0.892
is no inter-lab ~ f-critical 3.29 3.29 3.29 3.29 3.29 3.34 3.41 3.41
f/t~ 0.223 0.207 0.139 O.155 O.172 0.312 0.281 0.262
systematic introduction u 3 3 3 3 3 3 3 3
o f variation hence cq v 16 16 t6 16 16 16 15 15
suggesting that the ~ f-ratio 4.11 2.25 2.06 1.81 1.82 1.79 2.12 2.34
differences do not f-critical 3.24 3,24 3.24 3.24 3.24 3.24 3.29 3.29
occur due to the test ~ f/refit 1.27 0.695 0.636 0558 0.561 0.552 0.645 0.712
set-ups alone. A more
likely possibility is that Note: Figures in bold indicate f/fr ratios greater than unity
Px indicates the load at the prescribed fi value ofx mm

616
Materials and Structures / M a t ~ r i a u x et Constructions, Vol. 36, November 2003

both lab and mix variations into the

T a b l e 10 - C o m p i l a t i o n o f p o s t - h o e results on S F R C b e a m s equation for the second phase of the
Concrete Load, P, at prescribed average 6 values Area
Post-hoe testing programme thus rendering
grade (Fibre test P~x to fi =
dosage) Po.5 P;.o P~.s P2,0 P2~ P3.0 3ram within-mix variation studies for the
4-I 4-1 second phase specimens
4-1 meaningless. Loads at several points
Scheff6 3-4 * * 4-2 4-2 4-5 *
4-2 of mid-span deflections were taken
4-5 4-5
C25/30(25)
4-1 4-1 4-1 into consideration as has been done
4-1
Bonferroni 3-4 * 2-4 4-2 4-2 4-2 2-4 previously. These results are shown
4-2
4-5 4-5 4-5 in Figs. 13 and 14.
Scheff6 * * * * * * 1-5 * Figs. 13 and 14 indicate that the
C25/30(50)
Bonf?rroni * * * * l-5 1-5 1-5 1-5 variability is still significant even
* * * * * * *
Scheffd * when the results were separated
C25/30(75) * * * * * * *
Bonferroni * according to mixes. For the plain NSC
, * * * * * *
Scheff6 /-2 beams, the variability of the maximum
C70/85(25) * * * * , * *
Bonferroni 1-2
load is below 10%. On the other hand,
Note: 3-4 means significant difference observed between labs 3 and 4
the post-peak load variation increases
* means no statistical difference identified
from 1.20% to 13.2% at a mid-span
deflection of 0.1 turn to 9.94% to
T a b l e 11 - C o m p i l a t i o n o f A N O V A a n a l y s i s 50
for RILEM )arameters • N-0-1
Concrete
grade (Fibre Statistics fr~, f~q2 f~3 r~ N-0-3
dosage) A N-0-6
u 4 4 4
== - N-0-15
v 23 23 23
W
C25/30(25) f-ratio 6.97 1.85 3.07 30 84 x N-0-14
f-critical 2.80 2.80 2.80 E Variationfor O O

f/f~it 2.49 0.660 1.10 "6 peak loads

o N-0-5
P O
e
u 4 4 4 + N-0-0
v 28 28 28 o - +
C25/30(50) f-ratio 1.50 1,92 3.11 _ + § X

f-critical 2.71 2.71 2.71 "6 0

+ X
f/fcrit 0.554 0.708 1.15 @ <> 0
u 3 3 3
v 15 15 14
C25/30(75) f-ratio 2.71 0.678 0.956 8 I
0
f-critical 3.29 3.29 3.34 o
fff~rit 0.824 0.206 0.286 0.1 0.2 0.3 0.4 0,5 0.6
u 3 3 3 Average mid-span deflection, 5 (mm)
v 16 16 16
Fig, 13 -Within-mix variations for the plain NSC beams.
C70/85(25) f-ratio 3,78 1.99 1,91
f-critical 3.24 3.24 3.24 40
f/f~ 1.17 0.613 0.588
Note: Figures in bold indicate f/~:ritratio greater >
than unity
30 A w !
T a b l e 12 - C o m p i l a t i o n o f p o s t - h o c test results on A
+
v~ + + +
RILEM
Concrete E " O
Q

grade (Fibre Post-hoe test tfct fcq2 f~q3 20

dosage) 0 0
0 X X
Scheff~ 3-4 * * to
s.. r
C25/30(25) Bonf~rroni 3-4 * *
> ,0
Scheff6 * * *
C25/30(50) Bonferroni * * * I0
Scheft~ * * *
C25/30(75) Bonferroni * * * oN-50-7 oN-50-9 /xN-50-11 • xN-50-22

Scheff~ * * * 8 oN-50-23 +N-50-8 -N-50-10 - N - 5 0 - ~ 7 r

C70/85(25) Bonferroni 2-4 * *
Note: 3-4 means significant difference observed for 0.5 1 1.5 2 2.5 3 3.5
restdts between labs 3 and 4 Average mid-span deflection, 5 ( m m )
9 means no statistical difference identified
Fig. 14 - Within-mix variations for the C25/30 (with 50 kg/m 3
received only one beam from any one mix. This introduces of Dramix 80/60 BN fibres) SFRC beams.

617
TC 162-TDF

32.0% at a mid-span deflection of 0.4 mm. Clearly, most of the protbrma to complete. The identities of the test beams were
values of V lie above 10%. However, arguably, the most not made known to the testing labs. Several quantities,
important coefficient of variation for the plain concrete which were relevant to the design parameters proposed by
specimens is that fbr the maximum load. Since this is within RILEM TC-TDF, were to be calculated. These parameters
the arbitrary benchmark of 10%, the test method is robust were calculated for both the P-8 and P-CMOD curves.
where intra-lab tests (with styzcimens fi'om a particular mix) Fig. 16 shows the coefficients of variation of the R1LEM
are concerned. design parameters obtained from the P-8 curves. Fig. 17
As tbr the results for the NSC beams with 50 kg/m3 of shows the corresponding coefficients of variation of the
fibres, the variations are much higher. The load varies parameters obtained from the P-CMOD curves.
approximately from 10% to 30% within the range of mid-
span deflections considered.

5.6 Interpretation of calculation procedure

The method used to calculate the design parameters are
shown in Fig, 15. In the analysis of the beam results from
the round robin test programme reported above, all the
RILEM TC 162-TDF beam design parameters were
calculated by the task co-ordinator. This was done to avoid
any variation in the calculated results being due to differing
interpretation of the procedure pertaining to the calculation
of the design parameters. However, since the beam test
recommendations are to be used by various parties, it is
important to determine if differing interpretations of the
calculation procedure will introduce significant differences
in the design parameters. Therefore a study was carried out
in collaboration with the other testing labs involved in the
round robin programme to investigate this issue. By
comparing results from all the tabs, difficulties with the
interpretation of the procedure may be identified.
Eight beams were selected, from the pool of beam Fig. 16 - Variations in the calculations from P-8 curves.
results, by the task-coordinator. This "blind" set of beam
results was then sent to the other testing labs with a

Fig. 17 - Variations in the calculations from P-CMOD curves.

In general, the variability of most of the design par.meters

calculated at the different labs is found to be wilhin 5%, and all
are below 10%. Only two beams had Fararneters with variation
exceeding 5%. These are ~a values fbr beams 2 and 7. Design
p~ers t ~ and t~q3 had low variability and hence, ~ and
f~a3are geneeally quite s~aighfforw-ardto calculate and do not
pose significant probk~ns. The parmnc§ f~ for beams 2 and 7
Fig. 15 - Diagrams showing method of calculating RILEM design had a coefficient of vafi~on of 8.5% and 9.2%, respectively.
parameters i.e. fat, f~q2and feq3(after [1]). In the case of f~.~,it is obvious that the vaaability is related to

618
Materials and Structures / Matdriaux et Constr~wtions, Vol. 36, November 2003

difficulties in determining F~, because of their direct relationship. plain concrete beams). The variability of the measured load
Although the procedure tbr calculating the required within the range of mid-span deflections considered (for the
design parameters does not seem to have significant NSC beams with 50 kg/m 3 of fibres) is found to be 10% to
problems in interpretation, it is nevertheless quite a lengthy 30%.
and relatively difficult procedure to carry out. The From the inter-lab and within-mix studies, it can be
procedure in obtaining the necessary design parameters can concluded that the inherent material variability is the
be simplified if residual strengths at predetermined dominating factor in the relatively high variations observed
deflections (or CMODs) are used instead. This is desirable, in the round robin test programme.
as it will reduce the time needed to calculate the design A general study of the variation due to tim type of
parameters. In addition, it will theilitate the possibility of material shows that there is a decrease in the variability as
measuring CMOD alone to obtain the parameters as it has the fibre content is increased. A change in the concrete
been found that there is a simple relationship between 8 and strength matrix from a normal strength grade to a high
CMOD [3, 4]. Thus, if only residual strengths are needed strength grade does not seem to improve matters. It is
(instead of specific areas under the curve), the calculation therefore evident that the fft~re content is the overriding
procedure and the test method can be simplified. factor in influencing the variability in the beam test results.
The reason underlying this observation is that as the limit of
the amount of fibre the concrete mix can accommodate is
6. C O N C L U S I O N approached, the probability of getting a more uniform mix
is improved. This would reduce the variability in the fibre
To evaluate the reproducibility of the RILEM TC 162- distribution and thus test results with lower variability can
TDF proposed beam test, a study on the variation of the be expected.
round robin beam test results was performed. From the After a thorough analysis of the variation in the bem-n
study of the intra-lab variations, it is evident that most of results in t e ~ s of both intm-lab, inter-lab and within-mix
the variations in the post-peak regime for both the plain variation, it can be said that the possibility of achieving a
concrete and SFRC beams are significant. 10% convergence between results is not possible for the
tlowever, the intra-lab variations fi~rthe maximum load tbr fibre contents considered. Even within a given lab and mix,
the plain concrete beams were within 10%. As the most a convergence of 10% amongst beam results would be
important parameter for the plain concrete beam is the extremely difficult to achieve. This is due to the difficulty
maximum load, this suggests that the test is robust, at least in achieving a unitbrm fibre dis~bution especially with
where intra-lab variation is concerned. In addition, from the low fibre contents. Another possible reason for t_he high
intra-lab variations, there does not seem to be any systematic variation is that the test specimens have relatively small
large or small variation coming from any one particular lab i.e. cross sections. A small variation or difference in the
one particular lab does not have a large (or small) variation for number of fibres has a direct and a relatively large
all of the concrete types involved. This suggests that there influence on the toughness of the material tested. This
cotfld be more than one source of variation. Thus further phenomenon would be more pronounced in specimens with
investigations were carried out to identify the source or sources low fibre dosages.
of variations. More importantly, the investigations aim to To investigate the variation to results due to
identify the major source of variation. interpretation of the RILEM TC 162-TDF calculation
Therefore, it is safe to assume that differences in the test procedure, eight "blind" beam test results were sent to
set-ups are minimal in the sense that it should not adversely various testing labs. The variations in the calculations were
affect the beam test results to the point where the variation then analysed. It was found that, in general, the variation in
due to different set-ups overshadow the variation due to the the calculated values of the RILEM design parameters, t]-~t,
material itself. s and t;~q.~were within 5%. The design parameters fcq2 and
Similar tests on the SFRC beams, however, indicate that f~q3 had very low variability for all the beams and it is not
there seems to be significant differences in results from the anticipated that there will be any major problems in the
different testing labs. However, the statistical tests assume interpretation of the procedure to calculate them.
that the specimens all originate from a single population i.e. It is proposed that the procedure in obtaining the
from a single mix. This clearly is not the case and hence, necessary design parameters be simplified by using residual
the inherent material variation was studied. strengths at predetermined deflections (or CMODs). This
The within-mix variation study was based upon the plain will reduce the time needed to calculate the design
NSC beams and the NSC beams with 50 kg/m 3 of fibres. parameters. Moreover, it will fhcilitate the possibility of
This is because for the first phase, the beams received by measuring CMOD alone to obtain the parameters using a
each lab come from two mixes, each contributing four simple relationship between 6 and CMOD [3, 4]. Hence
beams. On the other hand, for the second phase, each with the use of residual strengths instead of specific areas
testing lab received 6 beanas, all coming from different under the curve, the calculation procedure and the test
mixes. Thus, the variations within the second phase results method can be simplified.
have both the influence of inter-lab and within-mix (or From the experience and analysis carried out with regards
inherent material) variations within them. to the recommended beam-bending test, it was found that the
From the within-mix studies, it is apparent that the test was a good and robust testing system. It was found that
variation within a mix itself is relatively high. Most of the although inter-lab variations do occur, this was relatively
V values at several prescribed average mid-span deflection small compared to the inherent material variation.
values are above 10% (except for the maximum load for the

619
TC 162-TDF

ACKNOWLEDGEMENTS [2] Taylor, M.R., Lydon, F.D. and Ban', B.LG., 'Mix
proportions tbr high strength concrete', Construction and
Building Materials 10 (6) (1996)445-450.
The work reported in this paper tbrms part of the Brite- [3] Barr, B.I.G. and Lee, M.K., 'Definition of round robin test.
Eurana project "Test and Design Methods for Steel Fibre Preparation of specimens. Execution and evaluation of
Reinforced Conca'ete", contract no. BRPR-CT98-0813. The round robin testing', Report for Test and Design Methods
partners in the project are: N.V. Bekaert S.A. (Belgium co- for Steel Fibre Reinforced Concrete, EU Contract-BRPR-
ordinator), Centre Scientifique et Technique de la Construction CT98-813 (2001) 105 p.
(Belgium), Katholieke Universiteit Leuven (Belgium), [4] Barr, B.I.G., Lee, M.K, de Place Hansen. E.J., Dupont, D.,
Technical University of Denmark (Denmark), Baltbur Beatty Erdem, E., Schaerlaekens, S., Schn[itgen, S., Stang, H. and
Rail Ltd (Great Britain), University of Wales Cardiff (Great Vandewalle, L., 'Rour_~l-robinanalysis of the RILEM TC 162-
Britain), Fertig-Decken-Union GmbH (Germany), Ruhx- TDF beam-bending test: Part 2 - Approximation of/5 from
University-Bochum (Germany), Technical University of CMOD response', Mater. Struct. 36 (263) (2003) 621-630.
[5] Cramer,D., 'Fundamental statistics for social research: step-
Braunschweig (Germany), FCC Construccion S,A. (Spain), by-step calculations and computer techniques using SPSS
Universitat Polyt6cnica de Catalunya (Spain). for Windows', Routledge, 1998.
[6] Wadsworth, H.M., 'Handbook of Statistical Methods For
Engineers and Scientists, 2ndedition', (McGraw-Hill, 1998).
REFERENCES [7] Keeping, E.S., 'Introduction to Statistical Inference',
(Nostrand, 1962),
[1] RILEM TC t62-TDF, 'R1LEM TC 162-TDF: Test and design [8] Neville, A.M., 'Properties of Concrete, 4~h edition',
methods for steel fibre reintbrced concrete. Bending test. (Longman Ltd, 1998).
Recommendations',Mater. Struct. 33 (225) (2000) 3-5.

620