Tensile properties of jute fibres

A. S. Virk, W. Hall* and J. Summerscales

One hundred tensile tests were undertaken at each of five distinct fibre lengths (6, 10, 20, 30 and
50 mm) on a single batch of jute fibres from South Asia. The Young’s moduli were found to be
independent of length. The ultimate stress (fracture strength) and fracture strains were found to
decrease with increasing fibre length. The variation in mechanical properties at each fibre length
was characterised using Weibull statistics based on a maximum likelihood estimate; referred to as
point estimates. Two empirical based models (a linear and a natural logarithmic interpolation
model) have been developed to estimate the fracture properties at any length between 6 and
50 mm. These two interpolation models were also developed based on maximum likelihood
estimates. The point estimates were used to benchmark the performance of the two models. The
natural logarithmic model was found to be superior to the linear model.
Keywords: Natural fibres, Jute, Tensile properties, Composites, Weibull statistics

Introduction tensile strength (on average).6 This physical model of the

fibre is often referred to as the principle of ‘weak link
Natural fibres can be divided into three groups, scaling’. In contrast to these physical based models,
vegetable, animal or mineral fibres. Vegetable fibres empirical or phenomenological models can be developed
can be wood (further subdivided into softwood or from a (limited) database of experimental results. A
hardwood) or non-wood (bast, leaf or seed hair) fibres.1 number of authors have used the principle of weak link
Bast fibres are those ‘obtained from the cell layers scaling in an attempt to fully characterise the statistical
surrounding the stems of various plants’.2 These fibres distribution of synthetic and natural fibre properties but
are used in textile applications and are increasingly being have found limited success.7–9
considered as reinforcements for polymer matrix com- The present paper considers jute fibres: 500 fibres are
posites as they are perceived to be ‘sustainable’. Initial characterised (five different lengths with 100 fibres at
results from a quantitative life cycle assessment have each length). The variation in mechanical properties at
been presented by Dissanayake et. al.3,4 Natural bast each fibre length is established using Weibull statistics
fibres are composed primarily of cellulose. Cellulose based on a maximum likelihood estimate (MLE). Two
microfibrils have a potential Young’s modulus of empirical based models (a linear and a natural logarith-
y140 GPa5 which is comparable to that of man made mic interpolation model) are then developed to estimate
aramid (Kevlar/Twaron) fibres at y125 GPa. The bast the fibre properties (ultimate strength and fracture
fibres which are currently attracting the most interest are strain) at any length between 6 and 50 mm. The
flax and hemp (in temperate climates) or jute and kenaf investigation therefore aims to enhance understanding
(in tropical climates). of the stochastic nature of jute fibres with reference to
The widespread use of natural fibres as the reinforce- the authors’ focus on natural fibre reinforced polymer
ment in polymer matrix composites has been con- matrix composite structures.
strained by the wider ‘natural’ variation in their
mechanical properties compared to their man made
counterparts. Thus, an in depth understanding of their
natural variation is necessary for these fibres to emerge
Physical characterisation of fibres
as a realistic alternative to synthetic fibre reinforcements Jute technical fibres from a single source grown in South
for structural composites. The fibre tensile properties are Asia were mounted on slotted cards (based on Grafil
limited by the presence of critical flaws. If a fibre Test 101?13) with Devcon 2 Ton epoxy adhesive. A total
comprises a series of elements (or links) then the strength of 500 fibre tests were carried out (100 fibre tests at each
of that fibre is governed by the weakest link. A longer of the five gauge lengths: 6, 10, 20, 30 and 50 mm to an
fibre contains more links than a shorter one and the accuracy of ¡1 mm at each end). Before testing,
probability of a critical flaw therefore increases with measurements of fibre ‘diameter’ were taken at 1 mm
fibre length, resulting in longer fibres having a lower intervals along the length of the fibre using an Olympus
BX60MF optical microscope (serial number: 5M04733)
and analySIS image analysis software. This is necessary
Advanced Composites Manufacturing Centre, School of Engineering/ as the fibre diameter varies along its length. Table 1
Reynolds Building, University of Plymouth, Plymouth PL4 8AA, UK shows the mean fibre diameters and standard deviations
*Corresponding author, email [email protected] at each fibre length and the overall measurements for all

ß 2009 Institute of Materials, Minerals and Mining

Published by Maney on behalf of the Institute
Received 5 September 2008; accepted 20 September 2008
DOI 10.1179/174328408X385818 Materials Science and Technology 2009 VOL 25 NO 10 1289
Virk et al. Tensile properties of jute fibres

2 Typical jute fibre stress–strain curves (at strain rate of

1 Confocal microscope image of cross-section of jute
0?01 min21)
fibre

load cell (serial number 52967). The load cell was

fibres. The mean fibre diameters vary from 53?9 to
calibrated with dead weights up to 2 N force: a
61?5 mm with the overall average of 58?8 mm.
consistent error of ,5% of the target force at each
The cross-sectional area of a fibre was calculated from
level was observed. Figure 2 shows the tensile stress–
its mean fibre diameter assuming a circular cross-section
strain curves for the specimens with the median
as is normal in the textile industry. To analyse the cross-
modulus at each of the five fibre lengths. The Young’s
section, jute fibres were cast into a potting compound
modulus, ultimate strength and fracture strain were
(epoxy resin). After curing, the epoxy resin containing
individually calculated for all 500 specimens. These
the jute fibres was sequentially polished using 300, 400,
results are presented in Table 2.
600, 800, 1200, 2500 grit paper then finally to 6 mm
The jute fibres show an expected natural variation in
diamond. An Olympus LEXT Confocal microscope
mechanical properties. As will be shown later, the
(serial no.: 6E23013) and analySIS image analysis
Young’s moduli of the fibres do not change significantly
software was used to study the cross-section of the jute
over the range of fibre lengths tested. The ultimate
fibres. A typical section of a fibre is shown in Fig. 1.
strength and fracture strain distributions (for 10 mm
Clearly the fibre is not circular and thus the assumption
long fibres) are shown in Fig. 3. Similar distributions
of a circular cross-section is therefore one error source in
can be observed for the other four fibre lengths (i.e. 6,
the calculation of the mechanical properties below. The
20, 30 and 50 mm).
potential error is likely to scale with the measured fibre
diameter and hence the stochastic variation in fibre The Young’s modulus was estimated from the
cross-sections warrants further study. strength and fracture strain data, using BS ISO
11566:1996,10 assuming a linear relationship. The
experimental ultimate strength and fracture strain
Young’s modulus, ultimate strength and distributions were analysed using Weibull statistics.
fracture strain The application of Weibull fracture statistics for the
probabilistic strength of materials (including fracture
Tensile tests at a constant strain rate of 0?01 min21 statistics for a unidirectional composite of ductile fibres
were carried out on an Instron 3340 single column in a brittle matrix) have been reviewed by Kittl and
test machine with an Instron 2519-104 S-beam 500 N Diáz,11 and their use in relation to filament strengths by
van der Zwaag.12 The two parameter Weibull prob-
Table 1 Mean fibre diameters (standard deviations in
parentheses)
ability density function (PDF) is


Fibre length, mm Diameter, mm b s b{1 {ðsgÞb
f(s)~ e (1)
g g
6 53.9 (14.0)
10 57.7 (11.0) where b is the shape parameter (Weibull modulus) and
20 61.1 (13.5) g is the scale parameter (characteristic strength or
30 59.6 (13.7) strain).
50 61.5 (18.6) Figure 4 shows the Young’s moduli, ultimate
All 58.8 (14.6)
strengths and fracture strains respectively. The data is

Table 2 Mean Young’s moduli, ultimate strength and fracture strain (standard deviations in parentheses)

Fibre length, mm Elastic modulus, GPa Strength, MPa Fracture strain, %

6 32.5 (10.1) 558 (194) 1.79 (0.33)

10 26.3 (9.0) 464 (165) 1.70 (0.41)
20 31.0 (7.0) 403.3 (140) 1.29 (0.30)
30 31.0 (7.1) 392.4 (131) 1.25 (0.31)
50 28.3 (8.8) 336.3 (132) 1.11 (0.34)

1290 Materials Science and Technology 2009 VOL 25 NO 10

 Virk et al. Tensile properties of jute fibres

a ultimate strength; b fracture strain

3 Frequency distributions for 10 mm fibres

presented with a horizontal line bisecting the box as the

median value, the box is bounded by the second and
third quartiles, the whiskers indicate the 10th and 90th
percentiles while the dots show the fifth and 95th
percentile points. The mean fibre modulus is y30 GPa
(29?9¡8?7 GPa) and is sensibly independent of fibre
length for the range measured. This is indicated by the
solid line in Fig. 4a. Figure 4b shows that the fibre
ultimate strength decreases with an increase in fibre
length and thus provides clear support for the Griffith
concept9 that longer fibres are more likely to have a
critical flaw than their shorter counterparts. Figure 4c
shows that the fracture strain for longer fibres is also
seen to decrease (as would be expected given the
quasilinear stress–strain curves).

a modulus; b ultimate strength; c fracture strain

Prediction of tensile properties using 4 Fibre property against fibre length
Weibull statistics
A designer should have a high level of confidence in the properties have high variability (as evidenced by the
mechanical property data if the selected materials are to above Tables). To fully understand the performance of
be used in structural applications, but natural fibre natural fibre composites, extensive knowledge of the

Table 3 Two parameter Weibull model for fibre tensile Table 4 Two parameter Weibull model for fibre fracture
strength strain

Fibre length, Weibull Characteristic strength Fibre length, Weibull Characteristic strain
mm modulus b g, MPa mm modulus b g, %

6 3.08 624 6 6.36 1.92

10 3.04 519 10 4.60 1.86
20 3.18 451 20 5.05 1.41
30 3.31 437 30 4.80 1.36
50 2.84 378 50 3.64 1.23

Materials Science and Technology 2009 VOL 25 NO 10 1291

Virk et al. Tensile properties of jute fibres

a Weibull modulus versus fibre length; b characteristic

a Weibull modulus versus fibre length; b characteristic
strain versus fibre length
stress versus fibre length
6 Two parameter Weibull fracture strain
5 Two parameter Weibull ultimate tensile strength

The 90% two sided confidence bounds (i.e. from 5%

fibre reinforcement properties is essential. Weibull lower bound to 95% upper bound) for tensile strength
statistics offer a means of characterising the variations and fracture strain parameters with a two parameter
in fibre ultimate strength and fracture strain. Weibull model are given in Tables 5 and 6 respectively.
For each of the five fibre lengths (i.e. 6, 10, 20, 30 and Further details of the mathematical approach for these
50 mm) the Weibull moduli, characteristic strengths and confidence bounds are given in Appendix II.
fracture strains were calculated using a MLE (see In Fig. 4b and c, the Weibull analysis confirms that
Appendix I). The computed ultimate tensile strength both ultimate strength and fracture strain decrease with
and fracture strain parameters for two parameter increasing fibre length (and hence, the Weibull para-
Weibull model are given in Tables 3 and 4 respectively. meters also decrease with increasing fibre length). It is

Table 5 Two parameter Weibull model: confidence bound for tensile strength

Fibre length, mm Upper bound bU Lower bound bL Upper bound gU Lower bound gL

6 3.47 2.73 660 590

10 3.45 2.68 549 490
20 3.60 2.81 476 428
30 3.75 2.92 460 415
50 3.23 2.49 402 356

Table 6 Two parameter Weibull model: confidence bound for fracture strain

Fibre length, mm Upper bound bU Lower bound bL Upper bound gU Lower bound gL

6 7.23 5.60 1.97% 1.87%

10 5.21 4.06 1.93% 1.79%
20 5.73 4.44 1.46% 1.36%
30 5.46 4.21 1.41% 1.32%
50 4.13 3.20 1.29% 1.17%

1292 Materials Science and Technology 2009 VOL 25 NO 10

 Virk et al. Tensile properties of jute fibres

therefore prudent to now consider the relationship the fit of the two interpolation models (see Figs. 5 and 6)
connecting the Weibull parameters (b and g) to the to the point estimates at the five fibre lengths suggests
fibre length. the logarithmic model (even at fibre lengths ,50 mm) is
superior to its simpler linear counterpart, an issue that is
Weibull parameter interpolation – MLE reinforced (and quantified) using the summation of
GOFNs. The GOFNs suggest a significant improvement
Plots of b and g (with confidence bounds) versus fibre
in the overall fit using the logarithmic model as opposed
length (with an error bound of ¡1 mm at each end) for
the fibre ultimate tensile strength and fracture strain to its linear counterpart. Both the linear and logarithmic
characteristics are shown in Figs. 5 and 6 respectively. models show a better fit to the strength data than to the
Weibull parameters were interpolated (based on a linear fracture strain data.
and natural logarithmic relationship) for the ultimate
strength and fracture strain by maximising the log Conclusion
likelihood function (using an MLE) for all of the
One hundred tensile tests were undertaken at each of five
experimental observations at all of the fibre lengths (see
Appendix III). It is considered likely that as fibre length distinct fibre lengths (6, 10, 20, 30 and 50 mm) on a
increases the ultimate strength and fracture strains (and single batch of jute fibres from South Asia. For the jute
hence the Weibull moduli and characteristic strength fibres tested, the Young’s modulus was found to be
and strain) will tend to a constant value, suggesting that y30 GPa while the ultimate strength and fracture
the logarithmic relationship will be the most relevant, strain fall from y558 to y336 MPa and y1?79 to
particularly for longer fibres. At fibre lengths up to y1?11% respectively as the length increases from 6 to
50 mm, however, the performance of the simpler linear 50 mm. Weibull parameters for the jute fibres have
model is still worthy of consideration. The Weibull been estimated for each fibre length using an MLE, i.e.
parameters for any fibre length (between 6 and 50 mm) point estimates. Based on the point estimates, two
can be approximated from either the simpler linear or empirical models (a linear and a natural logarithmic
the natural logarithmic trend lines. Based on the interpolation model) have been successfully developed
interpolated parameters, the ultimate strength and to characterise the ultimate strength and fracture strain
the fracture strain PDF can be calculated, and the across the entire range of the fibre lengths tested (i.e. 6–
associated fibre ultimate strength and fracture strain 50 mm). The logarithmic interpolation model for
data can be generated without additional mechanical ultimate strength and fracture strain was found to
testing, using (for example) the Monte Carlo method. produce a better fit to the point estimates (i.e. at the five
distinct fibre lengths) than the linear model, but both
Goodness of fit models produce a better estimation for ultimate
strength than for fracture strain.
The goodness of fit technique (see Appendix IV) was
used to examine how well the experimental data agrees
with the assumed distribution.13 Herein, an Anderson- Appendix I
Darling goodness of fit number (GOFN) is calculated to The two parameter Weibull PDF is
compare the experimental measurements to the Weibull

parameter point estimates at each fibre length (see b s b{1 {ðsgÞb
Tables 2 and 3 for strength and fracture strain para- f(s)~ e (1)
g g
meters respectively). The interpolated Weibull para-
meters used to characterise the assumed linear and where b is the shape parameter (Weibull modulus) and
logarithmic models are also evaluated at the measured g is the scale parameter (characteristic strength or
fibre lengths. The sum of the GOFN for the linear and strain).
logarithmic models at each fibre length is used to Experimental results show some observed values are
identify the best empirical model: the lowest sum is more likely to occur than other values. The PDF
considered to represent the ‘best fit’ model for the entire parameters are therefore estimated to maximise the
dataset. The summation of the point estimates highlights likelihood of producing the observed experimental
the minimum possible GOFN that can be attained using data.14
either the linear or logarithmic MLE interpolation The Likelihood function for the two parameter
functions, i.e. the target value. A visual inspection of Weibull PDF is

Table 7 Anderson Darling GOFN for Weibull ultimate Table 8 Anderson Darling GOFN for Weibull fracture
tensile strength at different fibre lengths strain at different fibre lengths

Interpolation Interpolation

Fibre Point Linear Logarithmic Fibre Point Linear Logarithmic

length estimate model model length estimate model model

6 0.90 4.97 1.31 6 0.24 4.62 0.86

10 0.71 1.29 1.44 10 0.33 2.69 3.14
20 0.33 3.63 1.02 20 0.36 13.62 4.30
30 0.27 0.95 0.70 30 0.70 3.08 1.04
50 0.48 1.32 0.94 50 0.78 4.25 1.46
Sum 2.69 12.16 5.41 Sum 2.41 28.26 10.80

Materials Science and Technology 2009 VOL 25 NO 10 1293

Virk et al. Tensile properties of jute fibres

n
  
b{1 b
 Fisher matrix,15,18 namely
{ðsgr Þ
L(sjb,g)~ P bg sr
g e (2)
r~1 0 1{1
  2
{ LLbL2
2
L L
{ LgLb
var(b) cov(g,b)
For computational convenience the log likelihood ~@ 2 2
A (8)
function is used14,15 cov(b,g) var(g) {LL { LLgL2
LbLg
 P n  b{1 Pn  b
L~n|ln bg z ln sgr { sr
g (3)
r~1 r~1

The Weibull parameters are estimated by maximising Appendix III

the log likelihood function using Newton’s method.16,17
The partial derivatives of equation (3) are set equal to
Weibull parameter interpolation –MLE
zero, namely Based on linear and logarithmic interpolation models
  Weibull parameter estimates for the strength and
LL fracture strain are established by maximising the log
~0 xj ~b,g (4)
Lxj likelihood function for all of the experimental observa-
tions at all of the fibre lengths.
Using Taylor’s series expansion the authors get equa-
For linear relations between the shape parameter b,
tion (5),16 which can be iteratively solved for shape and
scale parameter until convergence criterion is met the scale parameter g and the fibre length are
 
LLi b~aLi zb
½Ji
fXiz1 g~{ z½Ji
fXi g xj ~b,g (5) (9)
Lxj g~cLi zd
where, [Ji] is the matrix of second partial derivatives where a and c are the slope, and b and d are the intercept
(Hessian matrix) of the log likelihood function. {Xi} and for the shape parameter and scale parameter respec-
{Xiz1} are the initial and final vectors of the Weibull tively. Li is given fibre length.
parameters.
Replacing the shape parameter and the scale para-
 
LLi meter in PDF by the linear relationship yields
Lxj

aLi zb{1  s aLi zb
aLi zb s { cL zd
is a vector of the values of partial derivatives of the log f(s)~ e i
(10)
cLi zd cLi zd
likelihood function for the initial value of Weibull
parameters. Thus, the Weibull Likelihood function is
L(sja,b,c,d)~
2 3
Appendix II X k

aLi zb{1  sr aLi zb
aL i zb s r {
Pnr~1 4 5 (11)
cLi zd
e
Confidence bounds on Weibull parameters i~i
cLi zd cLi zd
Confidence intervals characterise a range within which
the point estimates of the Weibull parameters (see And the log likelihood function is
Tables 3 and 4) are likely to occur a given percentage of (
 n "
aLi zb{1#
Xk
aLi zb X sr
time.18,19 The uncertainty about the parameters is given L~ ni |ln z ln
by the confidence width (upper–lower estimate) of the i~1
cLi zd r~1
cLi zd
parameters. A wide interval may suggest that more )
Xn
aLi zb
samples need to be tested to get improved estimates of sr
{ (12)
the parameters.19 Herein, the Weibull Fisher matrix r~1
cLi zd
method15,18 is used to estimate the confidence bounds.
The lower and upper bounds on the point estimates of The parameters for the linear relationship are estimated
the Weibull parameters are calculated using by maximising the log likelihood function using
( ) Newton’s method16,17 (see Appendix I).
z½var(b)
1=2 b Similarly parameters for a logarithmic relation between
bU ~b|exp bL ~ n o the shape parameter or the scale parameter and the fibre
b z½ var(b)
1=2
exp b length can be optimised. The linear relations between
( ) (6) shape and scale parameter and fibre length are simply
z½var(g)
1=2 g replaced with the logarithmic functions
gU ~g|exp gL ~ n o
g z½var(g)
1=2
exp g
b~aln(Li )zb
For two sided confidence bound z is given by (13)
ð? g~cln(Li )zd
1{d 1 {t2
~ 1=2
e 2 dt~1{W(z) (7) It can be readily shown that the MLE interpolation
2 (2p) z
equations of the Weibull strength parameters are
where d is the confidence level, subscripts U and L
identify the upper and lower bounds for the Weibull bLinear ~0:0002(fibre length)z2:9726
parameters respectively, b and g are estimated b ~{0:0053ln(fibre length)z3:0731 (14)
Log
Weibull parameters. The variance and covariance of
Weibull parameters are estimated from the inverse of and

1294 Materials Science and Technology 2009 VOL 25 NO 10

 Virk et al. Tensile properties of jute fibres

gLinear ~{4:3307(fibre length)z582:0750 References

1. P. A. Fowler, J. M. Hughes and R. M. Elias: J. Sci. Food Agric.,
g ~{103:283ln(fibre length)z779:285
Log (15) 2006, 86, 1781–1789.
2. C. A. Farnfield and P. J. Alvey: ‘Textile terms and definitions’, 7th
The MLE interpolation equations of the Weibull para- edn; 1975, Manchester, The Textile Institute.
meters for fracture strain are, 3. N. P. J. Dissanayake, J. Summerscales, S. M. Grove and M. M.
Singh: Proc. Conf. on ‘Composites innovation 2007 – improved
bLinear ~{0:0378(fibre length)z5:3086 sustainability and environmental performance’, Barcelona, Spain,
October 2007, NetComposites, proceedings on CD-ROM.
b ~{0:817ln(fibre length)z6:955
Log (16) 4. N. P. J. Dissanayake, H. N. Dhakal, S. M. Grove, M. M. Singh
and J. Summerscales: Proc. Int. Conf. on ‘Flax and other bast
and plants (fiber foundations – transportation, clothing and shelter in
the bioeconomy)’, Saskatoon (Saskatchewan), Canada, July 2008,
gLinear ~{0:00016(fibre length)z0:0193 Saskatchewan Flax Development Commission, 47–58.
5. J. F. V. Vincent: ‘Structural biomaterials’, London, Macmillan,
g ~{0:0036ln(fibre length)z0:0258
Log (17) 1982.
6. A. A. Griffith: Philos. Trans. Roy. Soc. Lond. A, 1921, 221A, 163–198.
7. K. L. Pickering and T. L. Murray: Composites Part A, 1999, 30, (8),
1017–1021.
Appendix IV 8. J. Andersons, R. Joffe, M. Hojo and S. Ochiai: Compos. Sci.
Technol., 2002, 62, (1), 131–145.
Anderson Darling goodness of fit test 9. K. L. Pickering, G. W. Beckerman, S. N. Alam and N. J. Foreman:
The Anderson Darling GOFN is used to examine the fit Composites Part A, 2007, 39, (2), 461–468.
of experimental data to the calculated Weibull distribu- 10. ‘Carbon fibre – determination of the tensile properties of single-
filament specimen’, BS ISO 11566,1996.
tions at each fibre length (point estimates).
11. P. Kittl and G. Diáz: Res. Mech., 1998, 24, (2), 99–207.
The Anderson Darling GOFN number is calculated 12. S. van der Zwaag: J. Test. Eval., 1989, 17, (5), 292–298.
using equation (18),13 namely 13. M. A. Stephens: in ‘Goodness-of-fit techniques’, (ed. M. A. S.
Ralph and B. D’Agostino), 97–185; 1986, New York, Marcel
X
n
Dekker, Inc.
A2 ~{n{(1=n) (2i{1)½ln(Zi )zln(1{Znz1{i )
(18) 14. I. J. Myung: J. Math. Psychol., 2003, 47, 90–100.
i~1 15. ‘Life data analysis reference’; 2005, Tucson, AZ, ReliaSoft
Publishing. Available online at: http://www.weibull.com/lifedata-
where, n is the number of samples and Zi (for a Weibull webcontents.htm. (Accessed 22 August 2008).
distribution) is 16. S. S. Rao: ‘Engineering optimization: theory and practice’; 1996,
"
 # New York, John Wiley and Sons.
si b 17. J. Nocedal and S. J. Wright: ‘Numerical optimization’; 2006, New
Zi ~1{exp { (19) York, Springer.
g 18. E. A. Elsayed: ‘Reliability engineering’; 1996, New York, Addison
Wesley Longman.
where s is the vector of experimental observation in 19. M. Smithson: ‘Statistics and confidence’; 2000, London, SAGE
ascending order. Publication.

Materials Science and Technology 2009 VOL 25 NO 10 1295