Polymer Testing 31 (2012) 1026–1030

Contents lists available at SciVerse ScienceDirect

Polymer Testing
journal homepage: www.elsevier.com/locate/polytest

Test method

Evaluation of the influence of testing parameters on the melt flow index

of thermoplastics
Sérgio D.C. Guerreiro a, Isabel M. João a, L.E. Pimentel Real b, *
a
Área Departamental de Engenharia Química, Instituto Superior de Engenharia de Lisboa, IPL, Rua Conselheiro Emídio Navarro, 1 1959-007 Lisboa, Portugal
b
Departamento de Materiais, Núcleo de Orgânicos, Laboratório Nacional de Engenharia Civil, Av. do Brasil 101, 1700-066 Lisboa, Portugal

a r t i c l e i n f o a b s t r a c t

Article history: The main goals of the present work are the evaluation of the influence of several variables
Received 31 May 2012 and test parameters on the melt flow index (MFI) of thermoplastics, and the determination
Accepted 12 July 2012 of the uncertainty associated with the measurements. To evaluate the influence of test
parameters on the measurement of MFI the design of experiments (DOE) approach has
Keywords: been used. The uncertainty has been calculated using a “bottom-up” approach given in the
Melt flow index
“Guide to the Expression of the Uncertainty of Measurement” (GUM).
HDPE
Since an analytical expression relating the output response (MFI) with input parameters
Design of experiments
Analysis of variance does not exist, it has been necessary to build mathematical models by adjusting the
Uncertainty experimental observations of the response variable in accordance with each input
parameter. Subsequently, the determination of the uncertainty associated with the
measurement of MFI has been performed by applying the law of propagation of uncer-
tainty to the values of uncertainty of the input parameters. Finally, the activation energy
(Ea) of the melt flow at around 200
C and the respective uncertainty have also been
determined.
Ó 2012 Elsevier Ltd. All rights reserved.

1. Introduction material’s ease of flow [3], which depends on its molecular

Polyethylene (PE) is one of the most widely used ther-
moplastics [1]. Its main qualities are the desirable balance
between its physical properties in the solid state and its
chemical inertness. The combination of these qualities
together with its low cost and ready processability makes
PE the material of choice for a wide variety of uses [2].
Melt flow index (MFI) is a parameter that plays an
important role in the plastics industry and is likely to
remain a dominant tool for quality control [3], mostly due
to the ease of operation, good repeatability of results and
low cost [4]. It fulfills a requirement for rapid material
characterization, specifically for checking against specifi-
cation and for assessing processability in terms of the
* Corresponding author. Tel.: þ351 218443737; fax: þ351 218443023.
E-mail address: lpimentel@lnec.pt (L.E. Pimentel Real).
mass and viscosity.
MFI is basically quantified by the measurement of the
melt mass flow rate (MFR) or melt volume flow rate (MVR),
being defined as the weight (g) or the volume (cm3) of the
polymer extruded in a specified period of time (usually
10 min) through a capillary of specific diameter and length
by the pressure applied by a dead weight under prescribed
temperature conditions [5,6].
Over the years, it has been observed that MFI correlates
well with a number of useful parameters, such as reaction
temperature and catalyst activation temperature during
polymer manufacture [7].
However, when reporting the results of the measure-
ment of a physical quantity, it is necessary to provide
a quantitative indication of the quality of the results so as to
give some information on their reliability. Without such an
indication, measurement results cannot be compared,
either between themselves or with reference values given

0142-9418/$ – see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.polymertesting.2012.07.008
 S.D.C. Guerreiro et al. / Polymer Testing 31 (2012) 1026–1030
in a specification. Therefore, it is necessary to obtain
a readily implemented, easy to understand and generally
accepted procedure for characterizing the quality of a result
of a measurement, that is, for evaluating and expressing its
uncertainty, which is no more than the doubt about the
accuracy of the measurement result [8].
From among several usually applied methods for eval-
uating uncertainty, the one indicated in GUM [8] is the
most common. The implementation of GUM starts with the
analysis of a mathematical model, which should include all
the relevant contributions to the test or calibration. The
overall uncertainty is then estimated by the law of propa-
gation of uncertainty, following the identification and
quantification of the uncertainty of individual factors of
influence [9].
2. Experimental
For conducting this study, a high density polyethylene
(HDPE), Borealis HE2470, has been used. All measurements
have been performed following a standard procedure [6]
and using a similar mass of HDPE (near 2 g). All work has
been carried out using a Melt Flow Tester CEAST P/N 6941.
The validation of the results has been performed using
Microsoft Excel and Minitab software.
In order to test the effects of the test parameters on the
response measured by the plastometer, a factorial design has
been used. Table 1 presents the factor levels used in the design.
As the number of factors in a factorial experiment grows,
the number of effects to be estimated also grows rapidly [10].
The study has been performed in three steps. In the first step,
a one half fraction of the factorial design 25 has been used with
replication (r ¼ 2) corresponding to 32 experiments, followed
by two factorial designs 23, also with replication (r ¼ 2), cor-
responding to a total of 16 experiments in each.
For the determination of the uncertainty associated
with the measurements, 36 experiments have been carried
out in a random order, 12 for each statistical significant
parameter. The conditions applied are indicated in Table 2.
Finally, for determining the activation energy and its
associated uncertainty, 12 experiments have been carried out
in a random order at temperatures of 180,190, 200 and 210
C.
3. Results and discussion
3.1. Evaluation of the influence of several variables and
testing parameters
In the first step, the combined effect of the five param-
eters described above, in particular parameters A, B, D, E
and F, has been studied using a 251 fractional factorial
Table 1
Factor levels.
Level Temperature Load Diameter
of plastometer, weight, of die, C
A (
C) B (kg) (mm)
Low () 190 2.16 2.1
High (þ) 200 5.00 2.3
a
FLC – First and last cut.
b
MC – Measurement cut.
1027
design with a replicate (r ¼ 2) in order to verify which
factors and or interactions could have a significant effect on
the fluidity of the polymer. Parameter C, diameter of die,
was not used since at the time of the experiments only one
die was available.
For performing a full factorial design 25 with two repli-
cates, the number of experiments to test all the combina-
tions would be very high, corresponding to 64 experiments,
which would involve excessive time. Hence, half of the
combinations with I ¼ ABDEF have been performed, the
defining relation of the design and the complete defining
relation for the design being shown in Table 3.
The design is of resolution V, and in this type of design
no main effect or two factor interaction is aliased with any
other main effect or two factor interaction, making high
resolution designs easy to interpret. Hence, these designs
are particularly useful in factor screening experiments,
providing very good information about main effects and
two factor interactions. Table 4 presents the observed MVR
data for this experiment.
The analysis of variance is summarized in Table 5.
The computed F-ratios to a 5% upper critical value of the
F distribution, in particular F0.05;1.16 ¼ 4.49, have been
compared. Thus, any factor or interaction for which the
value of F0 is greater than the critical value will be
considered as significant. Under these conditions, three
cases can be pointed out, the effects of the main factors A
and B, in particular temperature of plastometer and load
weight, and the effect of the interaction AB, with respect to
the interaction between both factors.
Fig. 1 shows the normal probability plot of the effect
estimates based on the MFR experiment. Clearly, the main
effects B and A and AB interaction are significant as they fall
far from the line passing through the other points. The
analysis of variance summarized in Table 5 confirms these
findings.
In the second step a 23 factorial design (r ¼ 2) has been
used to study parameters A and B, plus C, since other dies
were then available. The results are presented in Table 6.
The analysis of variance is summarized in Table 7.
Once again, the computed F-ratios to a 5% upper critical
value of the F distribution, in particular F0.05;1.8 ¼ 5.32, have
been compared, and it has been observed that all the
factors and interactions are statistically significant, factors
B and C being the most significant.
In the third step, the parameters considered as signifi-
cant have been established, proceeding to the study of the
parameters and interactions regarded as non-significant, to
verify if the effect of these latter ones might not be hidden
by the manipulation of the former ones, a 23 full factorial
design having been used once again.
Length of Type of Number of samples. F
measurement, cut, E
D (mm)
6.35 FLCa 1
25.4 MCb 10
1028 S.D.C. Guerreiro et al. / Polymer Testing 31 (2012) 1026–1030

Table 2 Table 4
Applied conditions. MFR data for the 251 design.

Temperature, A (
C) Load weight, B (kg) Die diameter, C (mm) Run A B D E F MFR, g/10 min
180/190/200/210 2.16/3.04/4.04/5.00 2.0/2.1/2.2/2.3 r1 r2
1 1 1 1 1 þ1 0.122 0.127
2 þ1 1 1 1 1 0.144 0.144
After comparing the computed F-ratios with the 5% 3 1 þ1 1 1 1 0.556 0.549
upper critical value of the F distribution, F0.05;1.8 ¼ 5.32, it 4 þ1 þ1 1 1 þ1 0.630 0.639
seems that only the effect of factor D, length of measure- 5 1 1 þ1 1 1 0.124 0.123
6 þ1 1 þ1 1 þ1 0.140 0.139
ment, has statistical significance in the determination of 7 1 þ1 þ1 1 þ1 0.541 0.547
the fluidity of the polymer, since for the remaining factors 8 þ1 þ1 þ1 1 1 0.635 0.631
and interactions F0 was always smaller than the critical 9 1 1 1 þ1 1 0.127 0.127
value obtained. However, when comparing the obtained 10 þ1 1 1 þ1 þ1 0.141 0.146
11 1 þ1 1 þ1 þ1 0.544 0.550
value for this factor with the obtained values in the
12 þ1 þ1 1 þ1 1 0.622 0.638
previous step, its contribution is only marginally signifi- 13 1 1 þ1 þ1 þ1 0.121 0.122
cant, and thus it can be neglected. 14 þ1 1 þ1 þ1 1 0.140 0.137
15 1 þ1 þ1 þ1 1 0.549 0.550
16 þ1 þ1 þ1 þ1 þ1 0.630 0.633
3.2. Development of the mathematical model

The graphical representation of the measurement data that, for any of the parameters used, the best choice falls on
in Excel seems to be well fitted by a second order poly- the quadratic functions, because they have shown more
nomial function. Indeed, the use of the software Table curve reliable results.
confirms that the quadratic polynomial functions satisfac- The selected equation for each parameter is presented
torily represent the data, the analysis being based on the in Table 8.
value of the determination coefficient, r2, all around 0.99.
All better functions are complex and difficult to work with 3.3. Determination of the uncertainty associated to the
and have been discarded. However, we have decided to measurements
proceed in parallel with the study of linear functions, since
this fitting seems also to be a good choice. To determinate the uncertainty associated with the
For calculation purposes, a matrix model has been used measurement, the recommended GUM method has been
[11]. The regression coefficients have been obtained from used, considering (MVR) as response (output variable).
the determination of the coefficient vector b: The expanded uncertainty, U, associated with the

1
T  measurement has been obtained by multiplying the
b ¼ XT  X X Y (1) coverage factor, k, by the combined standard uncertainty, uc
(MVR):
The matrix of variances and co-variances, needed for the
uncertainty calculations, has been determined by the U ¼ k  uc ðyÞ (3)
following expression (Eq. (2)):
Table 5

1 Y T  Y  BT  X T  Y
XT  X  (2) Analysis of variance for the MFR experiment (first step).
nef
Source of variation Sum of Degrees of Mean F0
where nef is number of effective degrees of freedom, X squares freedom square
matrix concerns the various input parameters, Y matrix Main effects
corresponds to the measured response (melt volume flow A 0.0205 1 0.0205 1201.65
B 1.6745 1 1.6745 98,136.26
rate MVR) and signs “T” and “1” refer to the transposition
D 0.0001 1 <0.0001 3.55
and inversion of the respective matrices. E <0.0001 1 <0.0001 0.36
For the statistical analysis of the data, an analysis of 2nd order interactions
variance has been used, which made it possible to confirm AB 0.0089 1 0.0089 522.26
AD <0.0001 1 <0.0001 0.06
Table 3 AE <0.0001 1 <0.0001 0.47
Confounded effects for the fractional factorial design BD <0.0001 1 <0.0001 0.73
251. BE <0.0001 1 <0.0001 0.18
DE <0.0001 1 <0.0001 0.59
Confounded effects 3rd order interactions
A ¼ BDEF BE ¼ ADF ABD <0.0001 1 <0.0001 <0.0001
B ¼ ADEF DE ¼ ABF ABE <0.0001 1 <0.0001 <0.0001
D ¼ ABEF ABD ¼ EF ADE <0.0001 1 <0.0001 <0.0001
E ¼ ABDF ABE ¼ DF BDE <0.0001 1 <0.0001 <0.0001
AB ¼ DEF ADE ¼ BF 4th order interaction
AD ¼ BEF BDE ¼ AF ABDE <0.0001 1 <0.0001 1.05
AE ¼ BDF ABDE ¼ F Error 0.0003 16 <0.0001 –
BD ¼ AEF I ¼ ABDEF Total 1.7043 31 – –
 S.D.C. Guerreiro et al. / Polymer Testing 31 (2012) 1026–1030 1029

100 Table 7
Analysis of variance for the MFR experiment (second step).
90
80 A Source of variation Sum of Degrees of Mean F0
B
70 squares freedom square
Percentage

AB
60 Main effects
50 A 0.0149 1 0.0149 1860.50
40 B 1.3248 1 1.3248 165,600.12
C 0.1482 1 0.1482 18,528.12
30
2nd order interactions
20 AB 0.0092 1 0.0092 1152.00
10 AC 0.0005 1 0.0005 60.50
0 BC 0.0586 1 0.0586 7320.50
0 70 140 210 280 350 3rd order interaction
ABC 0.0004 1 0.0004 55.12
Normalized effects Error <0.0001 8 <0.0001 –
Total 1.5567 15 – –
Fig. 1. Normalized distribution of effects.

The expanded uncertainty, for a confidence level of 95%,

To determine the combined standard uncertainty, it was
has been eventually obtained by manipulating the param-
critical to ascertain the existence of correlation between
eters described above. Thus, using Eq. (3), the value of
parameters. Once this correlation has been proven, Eq. (4)
U95% ¼ 0:12 is obtained and it is possible to assure, for
represents the expression appropriate for its
the given standard conditions, in particular temperature of
determination:
190
C, load weight of 5 kg and a diameter of die of 2.1 mm,
Xn  df 2 Xn1 Xn  df  a value of MVR of (0.73  0.12) cm3/10 min or, alternatively,
u2c ðyÞ ¼ i¼1 dx
u 2
ðx iÞ þ 2 
i¼1 j¼iþ1 dx
a value of MFR of (0.55  0.09) g/10 min.
i i
 
df

  u xi ; xj 3.4. Determination of the activation energy and its associated
d xj uncertainty
(4)
The relation making it possible to calculate the activa-
For each parameter, there are different factors that
tion energy (Ea) of the melting flow is explicit in Eq. (5):
influence the final uncertainty. In this study, the maximum
error of the equipment, the expanded uncertainty of the Ea
equipment and the uncertainty associated with the lnðMFRÞ ¼ lnðkÞ  (5)
RT
regression are the factors to be considered, on the basis of
the results presented in Table 9. Thus, by plotting MFR as a function of temperature
After determining all factors, it has been possible to (Fig. 2), it is possible to obtain a straight line. The slope of
calculate the ones that are related to the standard uncer- the linear regression gives the value of Ea/R, where R is the
tainty of each parameter, on the basis of the results pre- universal gas constant. The value obtained corresponds to
sented in Table 10. Ea ¼ 28:8 kJ=mol
Proceeding to the calculation of the correlation coeffi- The uncertainty associated with the regression used for
cient, the existence of a positive linear correlation between Ea determination is calculated from Eq. (6) [12,13], where xi
each pair of parameters has been observed. The respective represents 1/Ti, 3 2i represents the residual sum of squares
correlation coefficients are near 1 (Table 11). and t is the coverage factor for 95% confidence, considering
The combined standard uncertainty, accounting for all a Student’s t distribution:
the values obtained in each of the previous steps, has vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
!
u P 2
presented a value of 0.06 cm3/10 min. u
¼ t n2  t
3i
UðEa Þ95% P  R2 (6)
The number of effective degrees of freedom calculated is ðn  2Þ ðXi  XÞ2
nef ¼ 50. Finally, based on a Student’s t distribution,
a coverage factor k ¼ 2 is obtained. After replacing all parameters by their values, it was
possible to determine the uncertainty associated to the
Table 6
MFR data for the 23 design (second step).
determination of Ea. In order to validate the results obtained

Run A B C MFR, g/10 min

Table 8
r1 r2
Determined functions for each parameter.
1 1 1 1 0.126 0.129
2 þ1 1 1 0.140 0.140 Factor Polynomial function
3 1 þ1 1 0.546 0.543 MVR (T), MVR ¼ 3.4167  105  T2  9.2167  104
4 þ1 þ1 1 0.628 0.636 cm3/10 min  T  3.2898  101
5 1 1 þ1 0.197 0.200 MVR (P), MVR ¼ 2.9193  102  P2  8.8161  103
6 þ1 1 þ1 0.212 0.212 cm3/10 min  P þ 4.4734  102
7 1 þ1 þ1 0.836 0.837 MVR (D), MVR ¼ 7.0833  101  D2  1.2688
8 þ1 þ1 þ1 0.970 0.964 cm3/10 min  D þ 3.0795  101
1030 S.D.C. Guerreiro et al. / Polymer Testing 31 (2012) 1026–1030

Table 9 The transformation of MVR into MFR made it possible to

Factors that influence the final result of the uncertainty. obtain MFR ¼ 0.55 g/10 min. A comparison of this value
Factors influencing Temperature Load Diameter with the value indicated in the product data sheet shows
final uncertainty of plastometer weight (g) of die (mm) that the measurements taken have been performed
(
C) correctly, since the values in question are similar, and the
Maximum error 0a 0a 0a product used indeed corresponded to what was included in
of the equipment
the data sheet.
Uncertainty of the 0.31 0.00007 0.0005
calibration In addition to the proposed objectives, it has also been
(expanded for 95%) possible to determine the Ea of viscous flow, through the
Uncertainty of the 0.8278b 0.0469b 0.0214b correlation between it and MFI, and the uncertainty asso-
regression ciated with this measurement, hence proving the existence
a
The values have been adjusted. of a relationship between MFI and another important
b
For standard conditions namely, T ¼ 190
C, P ¼ 5 kg and D ¼ 2.1 mm. parameter.
The uncertainty associated with the measurement of
MFI has a calculated value of 0.12 cm3/10 min for MVR
Table 10
Standard uncertainty associated to each parameter. and 0.09 g/10 min for MFR, which represents 16% of the
measured value. These values are considered too high,
Component of uncertainty Standard uncertainty
because they are significantly higher than the typical
u (T) 0.8839 (
C)
standard deviations of interlaboratory testing (up to 8%).
u (P) 0.0469 (g)
u (D) 0.0214 (mm)
The main reasons for such high values of uncertainty
are the heterogeneity of the HDPE used, which is not
a typical reference material, and the errors associated with
the fitting of parameter C. To reduce these problems
Table 11
Correlation coefficients.
a reference material for melt flow index measurements
would be welcome. The fitting regressions may also be
Component of uncertainty Correlation coefficient improved, using a multivariable regression or alternative
u (T,P) 0.9092 methods of uncertainty estimation, such as a Monte Carlo
u (T,D) 0.9099
approach.
u (P,D) 0.9050

References
-0.2500 y = -3464.6304x + 6.8720 [1] D.B. Malpass, Introduction to Industrial Polyethylene – Properties,
R² = 0.9970 Catalysts, Processes, Wiley, Massachusetts, 2010.
[2] A.J. Peacock, Handbook of Polyethylene – Structures, Properties and
Applications, Marcel Dekker, Inc., New York, 2000.
-0.4500
ln (MVR)

[3] M. Rides, C. Allen, H. Omloo, K. Nakayama, G. Cancelli, Interlabor-

-0.6500
-0.8500
0.0021 0.0022
1/T
Fig. 2. Determination of Ea.
above two complementary methods have been used [12,13],
leading to equal final results of UðEaÞ95% ¼ 1:1 kJ=mol.
4. Conclusions
The complete analysis of factorial design shows that the
temperature of the plastometer, the load weight and the
diameter of die are of important statistical significance in
the determination of melt flow index. Parameter D (length
of measurement), despite being statistically significant, has
contributed little to the determination of MFR, and
parameters E and F, the type of cut and the number of
samples, neither of them significantly affect the determi-
nation of MFI.
atory comparison of melt flow rate testing of moisture sensitive
plastics, Polymer Testing 28 (2009) 572–591.
[4] C.A. Cáceres, S.V. Canevarolo, Correlação entre o Índice de Fluxo à
Fusão e a Função da Distribuição de Cisão de Cadeia durante
a Degradação Termo-Mecânica do Polipropileno, Polímeros: Ciência
e Tecnologia 16 (4) (2006) 294–298.
[5] A.V. Shenoy, D.R. Saini, Melt flow index: more than just a quality
control rheological parameter, part 1, Advances in Polymer Tech-
0.0023 nology 6 (1) (1986a) 1–58.
[6] EN ISO 1133-1, Plastics – Determination of the Melt Mass-flow Rate
(MFR) and Melt Volume-flow Rate (MVR) of Thermoplastics – Part
1: Standard Method (ISO 1133-1:2011) (2011).
[7] A.V. Shenoy, D.R. Saini, Melt flow index: more than just a quality
control rheological parameter, part 2, Advances in Polymer Tech-
nology 6 (2) (1986b) 125–145.
[8] JCGM 100, Evaluation of Measurement Data: Guide to the Expres-
sion of Uncertainty in Measurement (2008).
[9] V.J. Barwick, S.L.R. Ellison, Estimating Measurement Uncertainty
Using a Cause and Effect and Reconciliation Approach, Part 2:
Measurement Uncertainty Estimates Compared with Collabo-
rative Trial Expectation, Analytical communication 35 (1998)
377–383.
[10] D.C. Montgomery, G.C. Runger, Applied Statistics and Probability for
Engineers, third ed., John Wiley and Sons, Inc., United States of
America, 2003.
[11] J.O. Rawlings, S.G. Pantula, D.A. Dickey, Applied Regression Analysis:
a Research Tool, second ed., Springer, United States of America,
1998.
[12] E.L. Crow, F.A. Davis, M.W. Maxfield, Statistics Manual, Dover
Publications, Inc., New York, 1960.
[13] D.A. Skoog, D.M. West, F.J. Holler, Fundamentals of Analytical
Chemistry. International Edition, seventh ed., Saunders College
Publishing, 1996.