T~l~rrnoc~Irir?

ticcr~
Acta. 238 ( 19943 295-307 295
Elsevict Scicncc B.V., Am&d&n
SSD1 w40-6031(94)018 14-Y

Modulated differential scanning calorimetry

M. Reading :b, A. Luget and R. Wilson

Ahslract

Modulated diffcrcntial scanning calorimetry (MDSC) is a recently dcvcloped extension

of DSC that adds a new dimunsion to the conventional approach. The usually lfnkar
tcmpcrnturc programmc is modulated by a smell perturbation, in this cast a sine wave. and
B mnthcmatical Lrcatmcnt is applied to the rcsulkmt data tu dcconvolute the s~rrnplc
rcsponsc to the perturbation from its rcsponsc to the underlying bating programme. In
this way the rcvcrsihlc (within the time scale of the pcrturbntion) and irrcvcrsible nature of
a thermal cvcnt can lx probed. The advantages include discntsngling ovcrlap+g
phennmcna. improving resolution and cnhnncing scmsitivity. A further bcnclil is that
mctastahlc melting can bc Jctcctcd in CRSCSwhcrc it is not normally ohscrved, This new
technique offers considcrablc scope for further dcvclcpmcnt.

INTRODUCTION

Differential scanning calorimetry (DSC) is an invaluable technique in

polymer science for studying glass transitions, reaction kinetics, melting
behaviour, thermal history, degree of crystallinity and other aspects of
polymer morphology. Modulated differential scanning caIorimetry
(MDSC) is a recently developed extension of DSC in which the usualIj!
linear or isothermai heating or cooling programme is modulated by some
form of perturbation [l-4]. A mathematical treatment is then used to
deconvolute the calorimetric response to the perturbation from the.
calorimetric response to the underlying temperature profile. The original
idea was conceived and demonstrated by M. Reading. A practical
automated instrument was then developed by’ M. Reading, 3. Hahn and B.
Crowe (Hahn and Crowe arz from TA Instruments) IS]. Here we use a
sinusoidal modulation and the deconvohition was cariied out using a
discrete Fourier transform method.
Either heat flux or power compensation instruments can be used for
MDSC which can be achieved with no fundamental change’ to the
conventional apparatus. The principle of its operation does not depend on
.
_.
* Corresponding authdr.

Qo40-6C)31 /Y4/507.00 &j 1994 - Elsevier. Science l3,V. All rights rescryed
2!?6 M. Re~dirr~ cl ~rr.~Tlrcr~rrochEI?I.
Acm 233 iIYY4) .&5~.?07

any particular design of calorimeter provided it is able to accurately

measure heats of reactions, latent heats, etc., even when used in a
non-modulated mode.

THEORY

The response of an ideal DSC under conditions where no significant

temperature gradients exist in the sample, is a combination of a signal that
.is dependent on the rate of change of temperature and one that is
dependent on the value of the temperaiure. The former depends on heat
capacity. In contrast, the rate of a thermally activated or kinetic process is
determined by the temperature. A process of this type is often, thought not
always, irreversible. Mathematically this can be expressed as
-dQ/df = C,,.t dT/dr +f(r, T) (1)
Where Q is the amount of heat absorbed by the sample, C,,,I is the
thermodynamic heat capacity (defined here as that due to the energy stored
in the vibrational, rotational and translational motions of the sample), T is
the absolute temperature, t is time, and f(t, T) is some function of time and
temperature that governs the kinetic response of any physical or chemical
transfo,rmation.
We have chosen the tel.m thermodynamic heat capacity, here and in ref.
4, in order to convey that we are referring to some change in the energy of
the sample that is not kinetically hindered and can, therefore, occur
practically instantaneously. Clearly we are not wishing to suggest that there
is something ‘un-thermodynamic’ about energy changes that arise from
kinetically hindered processes.
Equation (1) is based on a somewhat idealised picture, but it serves as a
first approximation model that does, as we shall see, enable us to interpret
many of our rtisults. The simplest case to consider is that of a linear
macromoIecule that is functinalised to allow it to undergo a cure reaction. If
the sample is being heated from some low temperature at which no
chemical reaction is occurring then the energy it absorbs as the temperature
is increased is dictated by the heat capacity which is in turn determined
mainly by the vibrational modes available to the sample. Wunderlich [63 has
demonstrated. such heat capacities can be calculated from. approximate
.vibrational spectra. If we assume that, (a) the sample is sufficiently thin to
avoid any significant temperature gradients occurring within it as it is
heated and cooled, and (b) the rate at which heat is supplied to the sample
is rrieas.ured by a perfect device,that responds instantaneously, then the rate
of change of sample temperature multiplied by C,,,, will perfectly match the
rate of heat flow. The process, of heat loss on cooling or heat gain on,heating
Cl1 be completely reversible and contain no hysteresis. This witt be true
regardless of the. nature of the temperature programme.
M. hwtlirtg EC a~./T~2cr?~~rrcC~i~?r. Accn 2.?N (i994) 2SS-307 297

When the temperature is increased the system will start curing. The
energy flow due to the bond breaking and forming process of the cure
reaction is fundamentally different from that due to the .sample’s
thermodynamic heat capacity. The rate of this chemicP1 reaction is
governed by the fraction of molecules that have sufficient energy to
overcome the activation barrier for the reaction which is, in turn;
determined by the temperature at any instant. Furthermore, this process is
completely irreversible. As the temperature of the sample increases and the
rate of the chemical reaction increases correspondingly, both it and the
thermodynamic heat capacity will contribute to the heat flow signal but they
will respond in different ways to changes in the heating programme.
Because of this, it is possible to separsite out the two different contributions.
As mentioned above, in MDSC the temperature piogramtie consists of
an underlying trend that is mqdulated by a small perturbation. The
underlying trend is usually the conventional linear temperature rise with
time. Here we have used the simplest modulation to achieve in practice
which is the superimposition of a sine wave. The temperature programme
tail be expressed as
Temperature = 7;) + bt + B sin(wt) (2)
Where z;, is the starting temperature, o is the frequency, I, is the heating
rate and B is the amplitude of temperature modulation.
If we assume that the temperature modulation is small and that over the
interval of the modulation the response of the rate of the kinetic process to
temperature can be approximated as linear, we can rewr-ite eqn. (1) as
dQ/dr = C,,,,(b + Bw cos(wf)) + f ‘(t,T) + C sin(of) (3)
where f ‘(I, T) is the average underlying kinetic function once the effect of
the sine wave modulation has been subtracted, C is the amplitude of the
kinetic response to the sine -wave modulation and (b + Bw cos(wt)) is the
measured quantity dT/dt. It may be thought that the assumption of a linear
response for the kinetic process is inadvisablk given the normally
exponential relationship between reaction rate and - l/T. However, over a
small temperature interval of a few tenths of a degree, this is a excellent
approximation for most commonly encountered chemical reactions.
It can be seen from eqn. (3) that the heat flow signal will contain a cyclic
component that will dependan the value of B, w and C. Below we illustrate
a method of separating this Cyclic response of the sample from its response
to the underlying heating rate and quantifying these twti signals-separately.
Another technique that is used to study responses to periodic tempera-
ture perturbations is AC calorimetry [7-g]. It consists of heating a sample
in a cyclic manner using a controlled ,amount- of electrical or electromag-
netic energy and measuring the corresponding increase in temperature.
Although the background temperature of the sample and measuring
system is often ramped, to date these instruments have been designed and
used to study the effects of the cyclic perturbation only, without taking any
account of the underlying signal. Most often the heat source is placed on
one side of the sample and the temperature sensor is on the opposite side.
This approach can confer some advantages when investigating the cyclic
response of liquid samples. However, this, and other features of their
design, make them unsuitable for making quantitative measurements of
heats of reaction, latent heats of melting, etc., even when a differential
configuration is used (this is especially true when the sample changes it
physical form during a transition, e.g. from a powder to a liquid). MDSC
provides all the usual calorimetric information obtainable with conven-
tional DSC and it is able to cope with changes in physical form. This makes
it a more appropriate technique for studying polymer systems and
irreversible phenomena generally.
The arguments presented so far do not account for sharp first order
transitions. Under the ideal conditions posited above the melting of a pure
metal would occur instantaneously and appear as a sharp spike that would
not be affected by the temperature modulation (though recrystallisation
might occur if the transiGon were reversible and the heating programme
included periods when the healing rate was negative). However, real
calorimeters cannot respond instantaneously and polymers rarely give
sharp melting peaks. We have shown that the underlying signal from
MDSC cpn give equivalent data to a conventional DSC for polymer melts
[4], thus latent heats, etc., can be measured in the normal way. A qualitative
interpretation of the cyclic signal obtained during polymer melting is siven
below. However, a full quantitative interpretation is not straightforward
and must include some consideration of the non-ideal behaviour of the
calorimeter cell. Work on this topic is being carried out in our laboratory
tind will be the subject of a future publication.
Clearly, real samples and real calorimeters do not conform to the
ideal&d specifications given above. Similarly the distinction between heat
flows arising from heat capacity and those arising from kinetic events is not
as clear cut in many systems as it is in the example given. However, we have
found this simple model provides a reasonable first approximation for
many thermal transitions and [cads to a method of analysis that yields
considerable practical benefits [4].

EXPERIMENTAL

All of tGe fotl6wing results wcrc obtained on a TA Instruments DSC

2910 modified to allow the MDSC mode of operation. In all these
experiments an underlying heating rate of O.OYCs” was used with a
modulation with a period of 50 s at an amplitude of 0.2”C controlled at the
reference (this is currently a non-standard implementation). Wunderlich
[lOI has observed that control of the reference is advantageous when
studying melting behaviour as the sample temperature itself cannot be
properly &ntrolIed.
The samples were all ICI poIy(ethylene terepthalate) (PET) and were
treated by heating the sample rapidIy to 270°C in the DSC, The
temperature was held for 10min before the sample was removed and
rapidly placed on a metal surface at room temperature. All of the heat
capacity measurements, including the cyclic heat capacity .(see below) were
performed by llrst running an empty pan (mass about I5 mg), then a
60.638 mg sapphire standard, then the PET sample. The sapphire and PET
runs were corrected for the empty pan offset. and then the sample c,, was
calcuIated by reference to the known c,, values for sapphire (these were
fitted to a third order polynomial to provide a continuous reference over
the whole temperature range). Sample pans were all flattened using a
specially made steeI former and-lids were used but not crimped so that the
same crucible could be used for the calibrations and PET runs.

RESULTS

Figure 1 shows the raw data for a MDSC response for a sample of
quenched PET using an underlying heating rate of 0.5”C s-l. The two basic
measurements are the heat flow, shown in Fig. l(a), and the, phase lag,
shown in Fig. 1 (b). It should be noted that in the experiments presented
here, the heating rate is never negative. This avoids possible asymmetries in
the heat-flow response that might arise from supercooling. The deconvolu-
tion algorithm consists of an averaging process that effectively subtracts the
effects of the modulation. When applied to the temperature programme
and the heat-flow response the average signals can be plotted against each
other as shown in Fig, 2. This process effectiveIy reconstitutes the results
that would have been obtained had a non-modulated experiment been
performed. We refer to these outputs as the underlying heat flow and
temperature. As mentioned above, we have found these curves to be
virtually indistinguishable from traditional DSC results obtained at. the
same underlying heating rate provided the rate at which a transition occurs
is not too rapid compared to the frequency of the perturbation [4]. A rough
guide is that there should be at Ieast six modulations over th.e course of any
single thermal event.
A discrete Fourier transform (DFT) algorithm is then used to evaluate
the magnitude of the cyclic component. The amplitude of the cyclic
component of the temperature signal can be compared with that from.the
dQ/dr signal, thus enabling a value for heat capacity to be determined
which we refer to as the cyclic Ci,_ It is important to realise that this cyclic
mtiasuremcnt comprises a contribution from the thermodynamic heat
capacity as defined above plus a contribution from any kinetic event, such
300

SsmoLa: PET QUENCHED ~UHCFWPD 23.4591

srro: 16.8460 mg
Method: -60 ta 393 a3 P!iO AQ.2
Comment: -60 TO 35O’C AT 3’C/MIN PEFIIOO-50s AMPLITUDE-0.2-C

dO0
(al
I

-100 .I
0 2000 4000 6000 8000 ia

samp 1e: PET OLIENCflED 122. El64l

RiZQ: 16.6670 mg
Huthod: --so 1’0 350 R3 P50 &O. 2
comnlcn t: .-SD TO 350% AT 3*C,‘m:n Period-SOS Ainplltude-0.2’C
0.4-
l-
0.2.

0.0’

a -0.2

i
d
-0.4

-0.6

_...I
v---Q-k-
-1.0 !
-60 0 50
L
100 150 200
I
250
I
300 i0
Temperature/T

Fig. I. Raw MDSC data for quenched PET. (a) Temperature. d7’/dr and heat flow data (no
baseline correction applied); endo, is downward and exo is upward. (b) The phase lag.
 Samplix PET IYUENCHED [LINCRHPD 23.0591
Sire: 16.8450 mS
HBthpd: -50 -0 .S50 R3 PSO A0.2
Commant: -60 TO 350-C AT 3-C,‘MIN PERIOD~5Os AMPLITtJDE~b.2-C

-SD
I
0 50
I
100 :50 200 250 300
I
350
TempereturePC

Fig. 2. Undertying heat flow (no baseline correction applied) vs. underlying temperature.
Endo is downward and cxo is upward.

as a cure reaction, arising from C sin(wc). A kinetic contribution can only

occur when there is an enthalpic process (such as an exotherm from a
crystallisation or cure reaction or an endotherm arising from a melting or a
relaxation event} whereas the heat capacity contribution is always present.
In principle the phase lag measurement could be used to separate the cyclic
C,, into the in- and out-of-phase components of the response to the
perturbation. However, before doing this, ways must be found of properly
calibrating and testing this signal. We are currently working on this topic
and this will be the subject of a future article. Here we use only the
amplitude of the cyclic component of the heat flow to calculate an overall
value for the cyclic heat capacity.
As stated above, the cosine wave (C,,_,) contribution is always present
and is, in the case of the PET sample shown in Fig. l(a), by far-the major
part of the cyclic component in the heat flow signal for aI1 of the transitions’
except the melting. This can be seen by reference to Fig. l(b) where the
peaks in the phase lag seen at the TS and the crystallisation are small;
consequently ih& kinetic contribution must also be small. We have
generally found this to be the case for all the polymer samples. we
have looked at when the amplitude of the modulation is a few tenths of a
degree.
 Underlyingi beet cepecilies of PET
At 3’Ymln.period n 5Os, amplitude - 0.2 *C

-1

-2

--3
-50 50 150 250

TsmperaturepC

Fig. 3. Underlying L;, for PET from three indcpcndcnt MDSC cxpcrimcnls.

One way of presenting MDSC data .is in terms of apparent specific heat
capacity which we denote here as c,,* The underlying heat flow can be
converted to this form simply by dividing by the underlying heating rate.
Figure 3 gives the results expressed in this fashion for three independent
experiments on PET. Figure 4 gives the corresponding cyclic heat capacity
measurements. There are minor differences in the sizes of the peaks
between the runs due to the difficulty of exactly duplicating the quenching
conditions.
It is appropriate to make a few observations regarding the accuracy of
the cyclic heat capacity values. Some very encouraging initial results [2,,4]
led us to the conclusion that heat capacities could be measured from the
cyclic component without even the need for empty pan corrections. It was
observed that most of the empty pan curvature typically found even in
modern instruments was expressed in the underlying signal but not in the
‘cyclic signal and, in our first instrument, the deviation from zero for the
cyclic measurement when using ctirefully ma’tched pans was very small. As
the number of MDSC cells has grown, further experience has shown that
it Is advistible to carry out a full calibration involving an empty pan
and sapphire run if very accurate results arc required. Furthermore, the
thermal concluctivity of the sample plays a far more important role in the
cyclic measurement than in the underlying one and great care must be
taken to erwure good thermal contact belween the pan and ttie conducting
plate. A lid must. be used to ensure that the sample is completely
:
 Cyclic heel capacities ol PET
Al 3Wmlqpsrhd - 60 s. amplitude = 2%

-1

-2

--3
-SO 50 150 250

Tempemtu r&C

Fig. 4. Cyclic q, for PET From three independent MDSC cxpcrimcnts.

surrounded by a good thermal conductor and samples must not be too

thick. Further, work is required to optimise all the parameters including the
underlying heating rate, the amplitude and period of the perturbation, the
sample mass and configuration and the crucible type. We have found the
parameters we have used here (see above) give acceptable results but better
results will probabIy be achieved with further experience.
In Fig. 5 we have taken the average of the three experiments for bo’th the
underlying and cyclic C,, and coplotted them with the values for the
amorphous and glassy Cl, for PET taken from the.ATHAS database [IO]
which provides carefully verified reference C,, data. It can be seen that there
is reasonably good agreement between the literature values and the cycl,ic
C{, below the glass transition and above the melting transition. The
agreement with the underlying values is not as good. ‘but this’ is not
surprising given the slowness for the heating rate. Much higher vaIues are
gencrtil.ly preferred for accurate meas:rrements of C,, using conventional
DSC (to which the underlying MDSC signal is equivalent). However, using
a higher underlying heating rate would have degraded the quality ‘of the
cyclic data; thus a compromise has been reached. The underlying and cyclic
Cr, should be the same below the glass transition and above the melting
transiton as no kinetic events are occurring. Following the procedures we
have previously used (41, a linear baseline correction, shown in Fig. 5, has
been applied to the underlying data to bring them into alignment-with the
cyclic values. These resuIts. are shbwn in Fig. 6 which also shows the
difference’between the two signals.
 Hoat cepackieo of PET
showing literature vek~es and baseline correction

--5 , I
-50 50 t54 250
TemperelurePC
4 Ut (cryntalllno~ d Ut gemorphous$

Fig. 5. Averaged results from Figs. 3 and 4 plus the literature values for the c,, of PET from
the ATHAS datahasc. Also shown is the basclinc used to correct the underlying plot.

Heat capacirles of PET

with corrsaod LC,and difference plot

1 A

-1

-2

-3
-50 250

TwmperalurWC

Fig. 6. Average cyclic c,, and baseline corrected average underlying c,, for PET plus thu
diffcrencc between these ?wo curves.
M; Reading et uL/T~~cr~?tr~cl~ir?~. Acta 2.W (1994) 29%3U7 305

Inspection of Fig. 6. gives rise to some interesting observations.

Instantaneously (on the time scale of the modulation) reversible pherioti:
ena such as glass transitions appear in the cyclic measurement and in the
underlying measurement but not in the difference plot. When the Tg occurs
the difference plot appears to show a small relaxation endotherm. Simple
irreversible (on the time/temperature scale of the moduiation) kinetic
events such as the crystallisation appear in the underlying measurement and
in the difference signai but not in the cyclic data. This is intuitively
satisfactory when one considers eqn. (3). The modulation of the tempera-
ture is comparatively small (of the order of tenths of a degree) whereas the
modulation in heating rate. is comparatively Iarge, ranging from 0.5 to
5S”C min-‘. Thus, for many kinetically controlled phenomena, C can be
approximated to zero and the cyclic measurement ‘sees’ only the
thermodynamic heat capacity. The melting event presents a complex
mixture of thermodynamic and kinetic effects and appears in all three pIots.
These observations are in agreement with those made above with respect to
the phase lag.
Another way of presenting the data is in terms of heat flow. This is
achieved simply by multiplying the cyclic C,,, by the underlying heating rate&
The underlying signal is then termed the total heat flow, the cyclic signal is
called the reversing heat flow and the difference is the non-reversing heat
flow. These last two signals are illustrated in Fig. 7. MDSC has now been
commercialised and this is the basic form of the data given by the current
MDSC system manufactured by TA Instruments.
One feature that is apparent on inspection of Fig. 6 is that for the cyclic
measurements the onset of the melting peak starts immediately after the
cold recrystallisation. In PET it is known that a process of mehing and
subsequent recrystallisation occurs continuously from the point where a
distribution of metastable crystallites is formed during the cold crystallisa-
tion event untii complete melting occurs 161. This is not clearly dl:lected by
conventional DSC but is revealed by the cyclic measurement of MDSC.
This can be understood when the effect of heating rate on irreversible
melting of this kind is considered. It is well known that changing the heating
rate affects the balance between the melt and recrystallisation processes
and consequentIy the amount of crystalline and amorphous material &any
point in time [6]. It is, therefor 5, not surprising that the cyclic measurement,
which spans many heating rates, gives a different answer f&-n the
underlying measurement which is equivalent to single heating rate. This
type of behaviour is a powerful signature for metastable. behaviour and
equivalent results can only be obtained from conventional DSC by multiple
experiments at different heating rates.
The practical benefits of MDSC compared to conventional DSC have
been, already been illustrated [4]. They include, (1) making interpretation
clearer by defining an event as reversing (e.g.. a glass transition) .or
306 M.’ Riwiittg PI ctf./Tjterrtttrdtirtt* Acfn 2.38 (1994) 295-307

Samp la: PET QUENCHEI3 0JNCRMPD 23.0591

Siif% 16.6460 mg
Uethod: -60 to 350 RS PtiO AO. 2
Comment: -60 TO 350eC AT 3’Cr’MIN PERiODUSQe AMPLiTUDE-0.2-C

Total heel flow

-Revcrring hrat flow

Nonreversing heat flow

I 0 50 100 150 200 250 300 350 400

TempemturePC

Fig. 7. Results from an MDSC cxpcrimcnt on qucnchcd PET cxprcssed in terms of heat
flow (no basclinc correction npplicd). Endo is downward and exo is upward.

non-reversing (e.g. loss of volatile material), (2) revealing hidden

transitions by disentangling overlapping processes, (3) improving resolution
for second order transitions by making changes in heat capacity easy to
detect at low underlying heating rates, (4) improving the detection limit for
changes in heat capacity, (5) detecting metastable behaviour, (6) making
possible quasi-isothermal heat capacity measurements thus enabling, for
example, vitrification under isothermal conditions to be studied.
Though further work must be done, there are some encouraging
indications that MDSC will ultimately make possible more accurate
rkeasurements of heat capacity [l l] and thermal conductivizy [12]. There is
even the interesting prospect of continuouSly measuring both of these
quantities simultaneously as a function of temperature.
The current deconvolution methodology is very simple and leaves
enormous scope for improvement. The key to progress is a proper
understanding of how real ‘calorimeters and samples behave when subjected
to a cyclic perturbation. We are currently engaged in modeliing and testing
the behaviotir of the heat-flux cell used by us as a first step to improving our
deconvolution methodology. Beyond this there is the question df interpre-
ting the results of frequency scanning experiments. At very low frequencies
the cyclic response will contain information on the kinetics of any
irreversible (or ‘&lowly reversible’) process occurring. As the frequency is
increased the signa will become dominated by the thermodynar& h&at
capacity. As it is increased still further the sample’s response will be
determined by its thermal condtictivity. By using MDSC concepts we have
in prospect a considerable increase in the amount of information that can
be obtained from scanning calorimetry and it mr,y well become ‘the
preferred calorimetric technique for most polymer applications.

REFERENCES

1 M. Reading. D. Elliot and V-L. Hill. Proc. 9th ICTA Meeting. Hatfield. UK. Augusl
IYY2:J. Therm. Anal.. 41) (lYY3) Y4Y-955.
2 P.S. Gill. S.R. Saurbrunn nnd M. Reading. Proc. Yth ICTA Meeting, Hatfield. UK.
August lYY2: J. Therm. Anal.. 40 (1003) 931-039.
3 J.C. Scfcris. 1.M. Snlin. P.S. Gill and M. Reading, Proc. Acad. Grcccc, 67 (1992)
3 1 l-346.
4 M. Rending. Trends Polym. Sci.. 8(l) (1993) 248-2.53.
5 M. Rending. B. Hahn and 5. Crowc. United States Patent 5,224,775. lY92.
6 B. Wundcrlich. Thermal Analysis. Academic Press. Boston. 1990.
7 1. Hatta and A.J. Ikushima, J. Appl. Phys., 20[11) (IYXI) 1951%2I111.
8 G.S. Dixon. S.G. Black, CT. Butler and A.K. Jain. Anal. Biochcm., 121 (1982) 55-61.
Y O.L. Mayorga. W.W. Osdol. J.1,. Lacombn and E. Frcirc. Proc. Natl. Acad. Sci. USA,
Biophysics. ,YS ( 1YXX) YS 14-45 18.
111 B. Wundcrlich. personal communication. lYY2.
11 Y. Jin. A. Bollcr. and B.‘Wundcrlich, Abwtr. 22nd NATAS Conf.. Denver. CO, 19-22
Scptcmbcr iY93. pp+ 56-64.
12 S.M. Marcus and R. Blainc. Ahstr. 22nd NATAS Conf.. Denver, CO, 10-22 September
1993, pp. 102- 1OX.