University of Cambridge, Materials Science & Metallurgy

H. K. D. H. Bhadeshia

Thermal Analysis Techniques

Thermal analysis comprises a group of techniques in which a physical property of a substance is measured as a function of temperature, while the substance is subjected to a controlled temperature programme. In dierential thermal analysis, the temperature dierence that develops between a sample and an inert reference material is measured, when both are subjected to identical heattreatments. The related technique of dierential scanning calorimetry relies on dierences in energy required to maintain the sample and reference at an identical temperature. Length or volume changes that occur on subjecting materials to heat treatment are detected in dilatometry; Xray or neutron diraction can also be used to measure dimensional changes. Both thermogravimetry and evolvedgas analysis are techniques which rely on samples which decompose at elevated temperatures. The former monitors changes in the mass of the specimen on heating, whereas the latter is based on the gases evolved on heating the sample. Electrical conductivity measurements can be related to changes in the defect density of materials or to study phase transitions.

Differential Thermal Analysis (DTA)

Introduction DTA involves heating or cooling a test sample and an inert reference under identical conditions, while recording any temperature dierence between the sample and reference. This dierential temperature is then plotted against time, or against temperature. Changes in the sample which lead to the absorption or evolution of heat can be detected relative to the inert reference. Dierential temperatures can also arise between two inert samples when their response to the applied heattreatment is not identical. DTA can therefore be used to study thermal properties and phase changes which do not lead to a change in enthalpy. The baseline of the DTA curve should then exhibit discontinuities at the transition temperatures and the slope of the curve at any point will depend on the microstructural constitution at that temperature. A DTA curve can be used as a nger print for identication purposes, for example, in the study of clays where the structural similarity of dierent forms renders diraction experiments dicult to interpret. The area under a DTA peak can be to the enthalpy change and is not aected by the heat capacity of the sample. DTA may be dened formally as a technique for recording the dierence in temperature between a substance and a reference material against either time or temperature as the two specimens are subjected to identical temperature regimes in an environment heated or cooled at a controlled rate.

Apparatus The key features of a dierential thermal analysis kit are as follows (Fig. 1): 1. Sample holder comprising thermocouples, sample containers and a ceramic or metallic block. 2. Furnace. 3. Temperature programmer. 4. Recording system. The last three items come in a variety of commercially available forms and are not be discussed in any detail. The essential requirements of the furnace are that it should provide a stable and suciently large hotzone and must be able to respond rapidly to commands from the temperature programmer. A temperature programmer is essential in order to obtain constant heating rates. The recording system must have a low inertia to faithfully reproduce variations in the experimental setup.

Fig. 1: Schematic illustration of a DTA cell.

The sample holder assembly consists of a thermocouple each for the sample and reference, surrounded by a block to ensure an even heat distribution. The sample is contained in a small crucible designed with an indentation on the base to ensure a snug t over the thermocouple bead. The crucible may be made of materials such as Pyrex, silica, nickel or platinum, depending on the temperature and nature of the tests involved. The thermocouples should not be placed in direct contact with the sample to avoid contamination and degradation, although sensitivity may be compromised.

Metallic blocks are less prone to baseline drift when compared with ceramics which contain porosity. On the other hand, their high thermal conductivity leads to smaller DTA peaks. The sample assembly is isolated against electrical interference from the furnace wiring with an earthed sheath, often made of platinumcoated ceramic material. The sheath can also be used to contain the sample region within a controlled atmosphere or a vacuum. During experiments at temperatures in the range 200 to 500 C, problems are encountered in transferring heat uniformly away from the specimen. These may be mitigated by using thermocouples in the form of at discs to ensure optimum thermal contact with the now at bottomed sample container, made of aluminium or platinum foil. To ensure reproducibility, it is then necessary to ensure that the thermocouple and container are consistently located with respect to each other. Experimental Factors Care is necessary in selecting the experimental parameters. For example, the eects of specimen environment, composition, size and surfacetovolume ratio all aect powder decomposition reactions, whereas these particular variables may not aect solidstate phase changes. Experiments are frequently performed on powders so the resulting data may not be representative of bulk samples, where transformations may be controlled by the build up of strain energy. The packing state of any powder sample becomes important in decomposition reactions and can lead to large variations between apparently identical samples. In some circumstances, the rate of heat evolution may be high enough to saturate the response capability of the measuring system; it is better then to dilute the test sample with inert material. For the measurement of phase transformation temperatures, it is advisable to ensure that the peak temperature does not vary with sample size. The shape of a DTA peak does depend on sample weight and the heating rate used. Lowering the heating rate is roughly equivalent to reducing the sample weight; both lead to sharper peaks with improved resolution, although this is only useful if the signal to noise ratio is not compromised. The inuence of heating rate on the peak shape and disposition can be used to advantage in the study of decomposition reactions, but for kinetic analysis it is important to minimise thermal gradients by reducing specimen size or heating rate. Interpretation and Presentation of Data A simple DTA curve may consist of linear portions displaced from the abscissa because the heat capacities and thermal conductivities of the test and reference samples are not identical, and of peaks corresponding to the evolution or absorption of heat following physical or chemical changes in the test sample. There are diculties with the measurement of transition temperatures using DTA curves. The onset of the DTA peak in principle gives the starttemperature, but there may be temperature lags depending on the location of the thermocouple with respect to the reference and test samples or the DTA block. It is wise to calibrate the apparatus with materials of precisely known melting points. The peak area (A), which is related to enthalpy changes in the test sample, is that enclosed between the peak and the interpolated baseline. When the dierential thermocouples are in thermal, but not in physical contact with the test and reference materials,

it can be shown that A is given by A=

mq gK

where m is the sample mass, q is the enthalpy change per unit mass, g is a measured shape factor and K is the thermal conductivity of sample. With porous, compacted or heaped samples, the gas lling the pores can alter the thermal conductivity of the atmosphere surrounding the DTA container and lead to large errors in the peak area. The situation is made worse when gases are evolved from the sample, making the thermal conductivity of the DTAcell environment dierent from that used in calibration experiments. The DTA apparatus is calibrated for enthalpy by measuring peak areas on standard samples over specied temperature ranges. The calibration should be based upon at least two dierent samples, conducting both heating and cooling experiments. It is possible to measure the heat capacity CP at constant pressure using DTA: CP = K T 2 T1 mH

where T1 and T2 are the dierential temperatures generated when the apparatus is rst run without any sample at all and then with the test sample in position. H is the heating rate and the constant K is determined by calibration against standard substances.

University of Cambridge, Materials Science & Metallurgy

H. K. D. H. Bhadeshia

Dierential Scanning Calorimetry

Introduction Dierential scanning calorimetry (DSC) is a technique for measuring the energy necessary to establish a nearly zero temperature dierence between a substance and an inert reference material, as the two specimens are subjected to identical temperature regimes in an environment heated or cooled at a controlled rate. There are two types of DSC systems in common use (Fig. 1). In powercompensation DSC the temperatures of the sample and reference are controlled independently using separate, identical furnaces. The temperatures of the sample and reference are made identical by varying the power input to the two furnaces; the energy required to do this is a measure of the enthalpy or heat capacity changes in the sample relative to the reference. In heatux DSC, the sample and reference are connected by a lowresistance heatow path (a metal disc). The assembly is enclosed in a single furnace. Enthalpy or heat capacity changes in the sample cause a dierence in its temperature relative to the reference; the resulting heat ow is small compared with that in dierential thermal analysis (DTA) because the sample and reference are in good thermal contact. The temperature dierence is recorded and related to enthalpy change in the sample using calibration experiments.

Fig. 1: (a) Heat ux DSC; (b) powercompensation DSC

Heatux DSC This section is based largely on a description of the Dupont DSC system by Baxter and Greer. The system is a subtle modication of DTA, diering only by the fact that the sample and

reference crucibles are linked by good heatow path. The sample and reference are enclosed in the same furnace. The dierence in energy required to maintain them at a nearly identical temperature is provided by the heat changes in the sample. Any excess energy is conducted between the sample and reference through the connecting metallic disc, a feature absent in DTA. As in modern DTA equipment, the thermocouples are not embedded in either of the specimens; the small temperature dierence that may develop between the sample and the inert reference (usually an empty sample pan and lid) is proportional to the heat ow between the two. The fact that the temperature dierence is small is important to ensure that both containers are exposed to essentially the same temperature programme. The main assembly of the DSC cell is enclosed in a cylindrical, silver heating black, which dissipates heat to the specimens via a constantan disc which is attached to the silver block. The disc has two raised platforms on which the sample and reference pans are placed. A chromel disc and connecting wire are attached to the underside of each platform, and the resulting chromelconstantan thermocouples are used to determine the dierential temperatures of interest. Alumel wires attached to the chromel discs provide the chromelalumel junctions for independently measuring the sample and reference temperature. A separate thermocouple embedded in the silver block serves a temperature controller for the programmed heating cycle. An inert gas is passed through the cell at a constant ow rate of about 40 ml min1 ). The thermal resistances of the system vary with temperature, but the instruments can be used in the calibrated mode, where the amplication is automatically varied with temperature to give a nearly constant calorimetric sensitivity. Heat Flow in HeatFlux DSC Systems A variety of temperature lags develop between the specimens and thermocouples, since the latter are not in direct contact with the samples. The measured T is not equal to TS TR where TS and TR are the sample and reference temperatures respectively. TS TR may be deduced by considering the heat ow paths in the system. The following additional notation (due to Greer and Baxter) is relevant (Fig. 2): TSP , TRP = Temperature of the sample and reference platforms, respectively, as measured by the thermocouples. TSP is normally plotted as the abscissa of a DSC curve. TF = Temperature of the silver heating block. RD =Thermal resistance between the furnace wall and the sample or reference platforms (units C min J1 ). RS , RR = Thermal resistances between the sample (or reference) platform and the sample (or reference). CS , CR = Heat capacity of the sample (or reference) and its container. H = Imposed heating rate. TR = Temperature lag of the reference platform relative to furnace. TS = Temperature lag of the sample platform relative to furnace. TL = Temperature lag of the sample relative to the sample thermocouple.

The following equations then hold: TR = HRD CR TS = HRD CS T = HRD (CS CR ) TL = HRS CS TS = TR + T TL = RS /RD TS (1) (2) (3) (4) (5) (6)

Fig. 2: Thermal resistance diagram representing a heatux DSC

Calibration: The Temperature Lag TL TL is nonzero because the thermocouple is not in direct contact with the sample. When the transition temperature T does not vary with heating rate, equation 4 indicates that a plot of the apparent T versus H keeping the other quantities xed, would at zero H extrapolate to the true value of T ; the apparent T is the true value plus the lag. A plot of the apparent T versus CS would also extrapolate to the true T at CS = 0, when H and RS are kept constant. Alternatively, the sample may be allowed to reach the temperature of the sampleplatform by holding at a temperature just beyond T , and recording a DSC curve corresponding to the equilibration event. The area of this curve can then be used to deduce the temperature lag; this kind of an analysis requires more sophisticated equipment than is normally available. Another method, due to Greer, is based on equation 6, and involves the evaluation of RS /RD . TR is measured for a particular reference, usually just an empty pan and lid. A heating run is rst performed with an empty pan on both the sample and reference platforms. This provides a baseline, from which measurements of T can be carried out. A second run is then performed, with two pans on the sample side, and one on the reference side. The dierence

between the rst and second DSC curves is a measure of TR , as a function of temperature. This becomes evident from equations 1 and 3; for the rst run, CS and CR are identical and hence T = 0, while for the second run CS = 2CR , so that T = TR . By repeating this procedure, TR can be obtained as a function of heating rate. To obtain the temperature lag TL , more tests are performed, bearing in mind that TR + T = TS Tests are conducted at a variety of heating rates, using a sample with a known transition temperature which is independent of heating rate, placed in the sample pan, with an empty pan on the reference side. These experiments give values of T , and hence TS , as a function of heating rate; the gradient g1 of the graph of TS versus heating rate is equal to RD CS , equation 2. Another set of experiments, based on equation 4 then gives a plot of the apparent transition temperature as a function of heating rate, and extrapolation to zero H yields the true transition temperature hence a graph of (TL versus H can be plotted, whose gradient g2 is equal to RS CS , equation 4. Hence, g1 /g2 = RD /RS . The temperature lag may be calculated (since RS /RD and TR are known) for a given reference and at any heating rate or CS , using equation 6. Temperature Calibration The temperature plotted on the abscissa of a DSC record is related to the emf generated at the thermocouple located under the sample. For standard thermocouple conditions, the emf may be reliably converted to temperature units using established calibration charts, but a variety of eects can cause the thermocouple to age and shift calibration. It is advisable to calibrate the abscissa using substances with precisely known melting points; most DSC instruments have facilities which allow calibration over limited temperature ranges. In changing the abscissa scale to a true temperature reading, allowances have to be made for the thermal lag eect (TL ), but this can be avoided by using very low heating rates for the purposes of calibration. Calorimetric Calibration Calibration is carried out by measuring the changes in specic heat or in enthalpy content of samples for which these quantities are known. When the DuPont instrument is used in the calibration mode, the procedure related to equation 2 may be used to measure specic heat changes. The heat balance equation for the heatux DSC system can be shown to be as follows: T TRP R + RS d(TSP TRP ) dH = SP (7) + (CS CR )H + CS D dt RD RD dt dH /dt refers to the heat evolution of an exothermic transition; the rst term on the right hand side is the area under the DSC peak, after correcting for the baseline. The second term on the right refers to the actual baseline, and it is this which is used in specic heat determinations. The last term takes account of the fact that some of the evolved heat will be consumed by the specimen to heat itself, and does not aect the are under the DSC peak, but may distort the peak shape. From equation 7 it is clear that when dH /dt can be arranged to be zero, the second term can be used to determine specic heat. The method is involves a comparison of the thermal lag between the sample and reference; the system is rst calibrated with a sapphire specimen, so that Csapphire = EqY /HM

where M is the mass of the specimen, E is a calibration constant, Csapphire , the specic heat capacity of the sapphire, q Y axis range (J s mm1 ) and Y the dierence in Y axis deection between sample (or sapphire) and blank curves at the temperature of interest. Enthalpy changes can be determined by measuring the areas under peaks on the DSC curve, when the latter is a plot of T versus time. A relationship of the form indicated in equation 1 then applies, again when the instrument is in the calibrated mode. The Baseline and the Transformation Curve In DTA or DSC, it is expedient to conduct experiments either isothermally or with the temperature changing at a constant rate. In the former case, the ordinate value would be plotted against time at isothermal temperature, whereas in the latter case it could be plotted against time or temperature. The following discussion is based on the abscissa being a time axis; the height referred to is that beyond the baseline. For DTA the height of the curve at any particular time t is a measure of the dierence in temperature, T , between the sample and the reference. For power compensated DSC, the height of the curve at some particular time t is a measure of the heat evolving from the sample per unit time, dH /dt (this also applies to heat ux DSC, after suitable calibration). For either DTA or DSC, one can assume that T is proportional to dx/dt or dH /dt to dx/dt, respectively. Here, x refers to the volume fraction of transformation, t to the time measured from the point where the appropriate curve departs from the baseline, and H to the enthalpy change. The constants of proportionality follow from the condition that the total area under the DTA or DSC corresponds to either x = 1, or to x equal to some constant value if the transformation terminates prematurely. This assumes that a reliable baseline can be obtained from the experimental information. The baseline can be visually estimated for sharp peaks without entailing large errors; for broad peaks it is dicult to qualitatively establish the baseline. The problem is complicated by the fact that the DSC instrumental baseline on either side of the peak is not a nosignal line. Even in the absence of a transition, the instrument measures the eect of the heat capacity of the sample, which may vary with temperature. This variation is usually nearly linear, but the curvature becomes noticeable over wide temperature ranges. One approximation to the baseline is a straight line connecting the start and nish of the transformation. Other methods involve the use of stepped baselines; the parent and product parts of the experimental curve are linearly extrapolated towards the centre of the experimental prole, and are connected by a vertical step at the position of the peak. Again, this method has no fundamental basis. The most reliable way of constructing the baseline is an iterative technique due to Scott and Ramachandrarao. The fractions transformed are rst calculated approximately, using a linear baseline between the initial and nal points of the reaction. The baselines of the parent and product are then extrapolated under the peak; this gives two separate baselines, since the heat capacities of the parent and pure product dier. The true baseline at any t is taken to be at a position between the extrapolated baselines. The exact positioning of the new baseline between the extrapolated parent and product baselines depends on an estimated value of the amount of product at any time t, using a lever rule type of a calculation. The new baseline generated in this manner can then be used as the starting point of another iteration and the process can be repeated to the desired accuracy. One iteration seems good enough for most purposes. A subtle correction which has to be taken into account when constructing transformation

curves from DSC curves is that the peak shape (rather than peak area) can be expected to be distorted, because some of any energy evolved may serve to the heat sample itself. In continuous heating experiments, the magnitude of this eect can be shown to be proportional to the heat capacity of the sample and to the rate of change of the dierential temperature with time. Autocatalysis and Recalescence Calorimetric experiments can be adiabatic or isothermal. The temperature is maintained constant in an isothermal experiment, whereas heat is neither added nor removed from the system during an adiabatic experiment. In practice, experiments fall somewhere between the ideal isothermal and adiabatic conditions. In an experiment where the rate of heat evolution is large relative to the capacity of the calorimeter to maintain isothermal conditions, the specimen temperature rises beyond the desired level, until a steady state is reached. This adiabatic rise in temperature will aect the rate of reaction, which may in term exaggerate the evolution of heat. This eect is known as autocatalysis. Recalescence describes the case where the release of heat reduces the transformation rate. Kinetics of Glass Crystallisation Both DSC and DTA have been used to study of the crystallisation of glasses. With few exceptions, the results have been analysed using JohnsonMehlAvrami equations with little attention to the mechanism of crystallisation. The general form of the equations is: x = 1 exp{ktn } (8)

where x is the volume fraction of transformation at time t, k is a function of transformation temperature, and n is a parameter which can in special cases give an indication of the mechanisms involved. The equation applies to isothermal transformations with the following assumptions: 1. It is assumed that the growth rate is constant, i.e. there is no composition change during transformation. 2. Modern calorimetric experiments use small quantities of samples; it is assumed the free surfaces of these samples do not aect the kinetics of transformation. 3. The extended volume concept on which the Avrami equation is based relies on random nucleation. Activation Energy The term k is temperature dependent since it is a function of the nucleation and growth rates of the transformation product; for most solidstate transformations both of these processes can be expected to be thermally activated. Consider a transformation in which nucleation is random, the nucleation and growth rates are constant and where growth is isotropic. Equation 8 becomes: x = 1 exp{Y 3 It4 /3} (9)

where Y is the growth rate and Iis the nucleation rate per unit volume. Hence, k = Y 3 I/3 = C1 (C2 exp{GY /RT })3 (C3 (exp{GI/RT }) = C4 (exp((3GY GI )/RT }) (10)

where GY and GI are the activation free energies for growth and nucleation, respectively, and both are assumed to be independent of temperature (R is the gas constant). A further assumption is that the growth and nucleation events are both singly activated processes. The activation energies of equation 10 may be lumped together into a single eective activation energy given by G , which is the term really obtained from an analysis using equation 9. G cannot be isolated using this analysis since x depends on more than just the growth rate. For isothermal transformation experiments, G can be obtained plotting the time taken to achieve a xed amount of transformation (i.e. tx ) versus 1/T , a plot based on equation 11 below, which is derived from equation 10: tx = C5 exp{G /nRT } (11)

It is dicult to determine the activation energy from anisothermal experiments. For any thermally activated process, the DTA or DSC peaks will shift with heating rate; Kissinger derived a relationship between the peak shift and the eective activation energy, assuming homogeneous transformation: dx/dt = C6 (1 x)m exp{G /RT } (12)

where m is the order of the reaction, and the other terms have their usual meanings. Kissinger showed that 2 d(ln{H/Tp }) G = (13) d(1/Tp ) R where H is the heating rate used and Tp is the sample temperature at which the maximum deection in the DTA or DSC curve is recorded. The equation requires that Tp equals the temperature at which the maximum reaction rate occurs. Most solidstate reactions are not homogeneous, but proceed by nucleation and growth events. Hence the G value obtained through equation 13 must not be compared with that obtained from isothermal experiments which obey the JohnsonMehlAvrami equation. Henderson has 2 shown that for reactions that obey equation 8, a plot of ln{H/Tp } versus 1/Tp should have a slope of G /nR rather than the G /R of equation 13. Marseglia has suggested that the activation energy G for anisothermal experiments can be deduced from a plot of ln{H/Tp } versus 1/Tp . The dierence between Marseglia and Henderson arises because the former takes account of the variation of k with time, whereas the latter does not. However, the manner in which the dependence of k on time is taken into account is not rigourous: dk dT dk dk = = H dt dT dt dT Thus, the variation in growth rate with time is not fully accounted for. Phase Transitions Thermal analysis techniques have the advantage that only a small amount of material is necessary. This ensures uniform temperature distribution and high resolution. The sample can be

encapsulated in an inert atmosphere to prevent oxidation, and low heating rates lead to higher accuracies. The reproducibility of the transition temperature can be checked by heating and cooling through the critical temperature range. During a rst order transformation, a latent heat is evolved, and the transformation obeys the classical ClausiusClapeyron equation. Second order transitions do not have accompanying latent heats, but like rst order changes, can be detected by abrupt variations in compressibility, heat capacity, thermal expansion coecients and the like. It is these variations that reveal phase transformations using thermal analysis techniques. Because of the sensitivity of liquidvapour transitions to pressure, additional precautions are called for when testing for boiling points or enthalpy changes. The ambient pressure is required; the peak area no longer corresponds to the latent heat of vaporisation in any simple way. The transition temperature T is related to the pressure P by the ClausiusClapeyron equation ln{P } = L/RT + C where L is the molar heat of vaporisation and C is an integration constant. L can be obtained using the Clausius-Clapeyron equation and a set of measured P, T values, assuming L is independent of temperature, that the volume of the vapour phase far exceeds that of the liquid, and that the vapour behaves as an ideal gas. Greater care is needed when studying solidsolid transitions where the enthalpy changes are much smaller than those associated with vaporisation. Stored energy in the form of elastic strains and defects can contribute to the energy balance, so that the physical state of the initial solid, and the nal state of the product, become important. This stored energy reduces the observed enthalpy change. Polymer Crystallinity It is assumed that a volume fraction V of the polymer consists of perfectly crystalline material which melts i.e. becomes amorphous, over the course of the experiment. The matrix which is not crystalline is assumed to be perfectly amorphous. The transition from the crystalline to the amorphous state is accompanied by a heat of fusion, written HF O when it occurs at the pure crystal melting point T0 . The fraction V of crystalline phase can be determined for a partially crystalline specimen by comparing the measured heat of fusion with HF O . Imagine a DSC experiment in which a partially crystalline polymer is heated from a temperature T1 to T2 where the polymer becomes completely amorphous (T1 < T0 < T2 ). The enthalpy changes can be analysed in the following phenomenological sequence (Fig. 3): a) Both the crystalline and amorphous phases are rst heated, without transformation to T0 . The enthalpy change for this process is Ha = V (HC,10 ) + (1 V )(HA,10 ) where the last two terms simply represent the change in heat content of the crystalline and amorphous components, respectively on heating form T1 to T0 . Ha can be deduced from the DSC curve by measuring the area between the section of the DSC curve obtained before any change in V , linearly extrapolated over the range T1 to T0 , and the instrumental baseline (i.e. the no-sample baseline). b) At T0 the crystalline component is allowed to become amorphous. The enthalpy of fusion for this is H b = V HF 0

c) The now completely amorphous material is permitted to rise in temperature from T0 to T2 , so that the enthalpy change is Hc = HA,02

Hc thus corresponds to the area between the DSC curve and the instrumental baseline, between the temperatures T0 to T2 . If the total enthalpy change calculated from the separation of the DSC curve from the instrumental baseline is given by H12 , then H12 = Ha + Hb + Hc and V = (H12 Ha Hc )/HF 0

Fig. 3: Analysis of a DSC peak.

University of Cambridge, Materials Science & Metallurgy

H. K. D. H. Bhadeshia

Dilatometry
The dilatometric method utilises either transformation strains or thermal strains; the basic data generated are in the form of curves of dimension against time and temperature. Thermal Expansion The coecient of linear expansion, also known as expansivity, is the ratio of the change in length per C to the length at 0 C. The coecient of volume expansion for solids is approximately three times the corresponding linear coecient. The coecient of volume expansion of a liquid is the ratio of the change in volume per degree, to the volume at 0 C. The value of the coecient varies with temperature. The coecient of volume expansion for a gas under constant pressure is nearly the same for all gases and temperatures, and is equal to 0.00367. If l0 is the length at 0 C, then the length at a temperature T can be written: lT = l0 (1 + e1 T + 22 T 2 + . . .) (1)

The expression is usually terminated after the second term; the same form of equation can be used to represent volume expansion. The thermal expansion is a consequence of the nature of interatomic forces, and solidstate theory predicts a simple relationship between specic heat capacity at constant volume (CV ) and the coecient of linear expansion. Thus, Gruneisen has demonstrated that e1 is proportional to CV , and this relationship can be exploited in dilatometry; Curie transitions in metals are associated with an anomaly in the specic heat capacity, and can therefore be detected by dilatometry through the accompanying change in e1 . The sample should be free to move, i.e. without any mechanical constraints which would limit accuracy. The length changes should not be transmitted through a second material upon whose expansion coecient depends the evaluation of the test material. These conditions rule out the use of most pushrod type dilatometers. A dilatometer capable of meeting the requirements is described in the Journal of Scientic Instruments, 2, 515 (1969). Study of Vacancies Ballu and coworkers rst suggested that dilatometry, combined with precision Xray lattice parameter measurements may be used to determine the concentrations of point defects in metals. They showed that the the ratio of vacancies to lattice sites, cv , is given by cv = 3 l a l a (2)

where l is the length and a the lattice parameter of a material with a cubic lattice. The presence of point defects must alter the volume of the sample, but the lattice parameter should not be aected, apart from certain changes due to relaxation, to be discussed later. In comparing lengths at dierent temperatures, subtracting a/a removes changes due to thermal expansion, leaving only the eect of point defects.

This subtraction is not necessary in isothermal experiments. The lattice parameter measured is a mean value which includes the eect of local relaxations around a point defect. Subtracting a/a from the length change therefore also removes the inuence of local relaxations, leaving only the contribution to l of new lattice sites formed or lost in creating the defects. During isothermal annealing of defects, lattice parameter measurements provide a measure of the lattice relaxation due to the defects and permit the determination of the absolute defect concentration. Vacancy clusters can lead to a misinterpretation of results. Fortunately, the concentration of such clusters is expected to be relatively small. Note that it is the excess concentration of vacancies that is revealed because interstitials can cancel some of the eect of the vacancies. PushRod Dilatometer The pushrod dilatometer can only be used for studies of length changes in solid materials. The sample rests between the tips of a xed quartz rod and a similar frictionless sensing rod in the centre of a highfrequency induction furnace. Length changes are transmitted through the frictionless rod to an electronic transducer which in turn drives the recording system. The thermocouple is spot welded to the sample, and referenced at 40 C by means of a constant temperature bath. Quenching gas enters the chamber, passes through the hollow cylindrical sample and escapes through the openings. It may not be possible in conventional dilatometers to conduct experiments in which the specimens require very fast heating or cooling rates, because of the high thermal inertia of the furnaces involved. In a highspeed dilatometer, the specimen is positioned along the axis of a cylindrical heating coil, which is connected to a radiofrequency power generator. During operation, the magnetic eld around the coil induces currents in the specimen, causing it to warm up. The induction coil itself is only mildly heated through resistive eects, but is in any case water cooled, Hence, the response of the system depends on the thermal characteristics of the specimen rather than those of the furnace. Fast quenching rates can therefore be achieved by directing highpressure jets of gas through the centre of a hollow specimen; quench rates of up to 5000 C/s can be obtained in favourable circumstances. Helium is used as the quenching gas in high cooling rate experiments because its thermal conductivity is about six times that of nitrogen. Large thermal gradients can develop in the specimen during the quench, which may cause diculties in the interpretation of the observed changes in specimen dimensions. The system should ideally be arranged to stabilise at the isothermal temperature concerned before the onset of any transformation. The specimen can be enclosed in a vacuum, but it is more usual to use an inert atmosphere; in the case of steels, an inert atmosphere results in a smaller degree of decarburisation. The specimen chamber should be evacuated before triggering the quench gas; otherwise, the build up of pressure in the chamber retards the quench, and may cause the lid of the dilatometer chamber to be blown o! The pumping system must be isolated before triggering the quench. Specimen Design Specimens for use in highspeed dilatometry are usually in the form of hollow rods, with internal and external diameters of 1.5 and 3.0 mm respectively; the length is limited by the extent of the furnace to a maximum of about 3 cm. These dimensions are somewhat arbitrary, but experience suggests that they satisfy the following requirements: 1. The specimen should be suciently thick to prevent free surface eects from altering

transformation kinetics. The tests should reect what happens in equivalent bulk samples. Nickel plating the specimens (to a thickness of about 0.08 mm) helps to reduce surface nucleation. Contrary to popular belief, this is not an eective way of preventing decarburisation in steels. To reduce decarburisation, the specimen should be copper plated; carbon is only sparingly soluble in copper. The plating material should be chosen so as not to interact with the sample, for example by penetration into the grain boundaries of the substrate. 2. The specimen dimensions must be small enough to allow rapid changes in temperature. The specimen should obviously be representative of bulk material. Its ends should be ground at and parallel to give a true cylindrical shape. Otherwise, slip at the specimenquartz interface can lead to erroneous interpretation especially under the inuence of the highpressure quenchgas jets. It is normal to isolate the specimen from the RF coil with a length of quartz tubing, not only to avoid contaminating the coil, but also to guard against the potentially disastrous consequences of accidental specimen melting. Highspeed dilatometers generally do not have long term electronic or mechanical stability. Equipment like this cannot be used for tests lasting more than a few hours. Prolonged holding at high temperatures can also lead to slagforming reactions between the specimen and quartz retainingrods. A certain amount of pressure is always necessary to hold the specimen between the quartz retaining rods and to remove backlash, so care should be taken to ensure that any resulting creep eects are negligible. Calibration The temperature calibration is similar to that of dierential scanning calorimeter and is not discussed further. However, for the calculation of thermal expansion coecients, and for the purposes of absolute dilatometry, it may be necessary to calibrate the magnication of the displacement transducer. There are two ways of doing this: 1. A pure platinum specimen with known expansion characteristics is heated at a sufciently slow rate over the temperature range of interest. The magnication M is then given by l (3) M= T lP t eP t where l is the deection of the length recording pen, T is the dierence between the initial and nal temperatures T1 and T2 respectively of the test, lP t is the length of the Pt specimen at T1 , eP t the linear expansively of Pt (obtainable from standard handbooks). This method can be accurate, but does not take account of the expansion of the part of the quartz rods within the furnace assembly. 2. A micrometer attachment on the dilatometer allows the transducer to be stimulated independently of specimen movement. The magnication is then simply l/micrometer movement. Having calculated the magnication the additional expansion to be expected from the quartz rods, lQ eQ can be obtained from : l = M (lP t eP t + lQ eQ ) (4)

Interpretation of Transformation Curves Dilatometric data are plotted as graphs of relative length change versus time. These plots are usually sigmoidal in shape, with the length change being dependent on the extent of transformation. If the length of the specimen before the beginning of transformation is l1 , and that at the termination of transformation at a time t = t2 is l2 , then at time t, Vt l l1 =f t (5) V2 l2 l1 where t = 0 at the beginning of transformation and lt is the specimen at any time t. Vt is the volume fraction of transformation corresponding to t, and V2 is that when reaction has stopped. The latter quantity may either be deduced by using an independent technique, or, if sucient data are available, it can be calculated from the magnitude of l2 , where l2 = l2 l1 For an austenite () to ferrite () transformation in a plain carbon steel, in which the formation of ferrite enriches the residual austenite with carbon, 2V a3 + (1 V )a30 a3 l = l 3a3 where l/l is the length change per unit length; V is the volume fraction of ferrite; a is the lattice parameter of ferrite at the transformation temperature; a0 is the lattice parameter of austenite at the transformation temperature when the austenite has the mean composition of the steel (x); a is the corresponding lattice parameter of carbonenriched austenite. The extent of enrichment can be estimated from mass balance: x = x + V (x x )/(1 V ) where x is the carbon concentration in the ferrite. Texture and Anisotropy of Thermal Expansion Most polycrystalline materials are crystallographically textured. This becomes important when testing materials with low crystallinesymmetry and anisotropic thermal expansion characteristics. The assumption that the measured length change corresponds is about a third of the volume change is no longer valid and it becomes necessary to specify the crystallographic directions along which measurements are made. It turns out that such eects may be advantageously exploited. Under the inuence of neutron irradiation Uranium has a pronounced tendency to swell in the < 0 1 0 > direction and contact in the < 1 0 0 > direction. In polycrystalline specimens, the swelling tendency must depend on the texture, which can be determined using tedious Xray techniques. Alternatively, an indirect measure of preferred orientation can be obtained by measuring thermal expansion and electrical resistivity along a variety of directions of the specimen shape. The expansion coecient and thermal resistivity characteristics of uranium are both anisotropic. High resistivity in a particular direction indicates an excess of < 1 0 0 > directions over random, while a low expansion coecient indicates an excess of < 0 1 0 > directions. Hence these two measurements made in the (6)

same direction on the specimen give an estimate of the tendency to irradiation growth. (Journal of Nuclear Materials, 4, 109, 1969). Miscellaneous It is known that cycling a specimen through a phase transformation can lead to permanent length change, so that the apparent length change observed may alter with the number of cycles, even though there may be no real changes in transformation behaviour. When dealing with specimens of low thermal conductivity, care should be taken to ensure that the imposed heattreatments allow the specimen to attain thermal equilibrium within the time period of the test.

University of Cambridge, Materials Science & Metallurgy

H. K. D. H. Bhadeshia

Other Techniques
Thermogravimetry Thermogravimetry (TG) is a technique by which the weight of a substance, in an environment heated or a cooled controlled rate is recorded as a function of time or temperature. Derivative thermogravimetry exploits the rst derivative of the TG curve. Thermogravimetry seems to have been used rst in 1912 in the study of the eorescence of hydrated salts and has since been an aid in the quantitative investigation of decomposition reactions; it is at its best when used in conjunction with calorimetric analysis. The equipment consists of a precision balance, a furnace with a programming facility, a reaction chamber and a suitable recording system. The assembly has to be capable of continuously registering any weight changes in the test sample whilst the latter is being heattreated. There are essentially two types of apparatus the nullpoint balance and the deection balance. In the former, any movement of the balance beam caused by weight changes is counteracted by a restoring force to bring the beam back to its original position; this force is then taken to be proportional to the weight change concerned. Such instruments are readily adaptable to operation in a vacuum. The deection instrument can be more robust and reliable because it is based a conventional analytical balance. The thermocouple is usually placed in direct contact with the sample. However, any connections to the support must aord an extremely small torque, or in the case of null balance instruments, a small or highly reproducible torque. TG is a quantitative and dynamic technique, and a number of factors can eect the shape of the TG curve. Temperature gradients, air buoyancy, convection currents within the furnace tube and other factors contribute to the so called buoyancy eect in which the weight of an inert, empty crucible changes with temperature. A correction curve can be determined to compensate for this eect, making certain that the conditions of such a calibration correspond to those of the actual experiment. TG gives absolute changes in sample weight so that the calculated extent of reaction is not aected by the heating rate used, although the start and nish temperatures are a function of heating rate because of kinetic barriers. The main applications of TG include the measurement of thermal stability, ageing characteristics, decomposition, reactivity and the structures of compounds. It has obvious uses in the determination of the moisture content of powders, water of hydration and of carbon monoxide and carbon dioxide evolution from carbonates etc. Decomposition reactions can be studied in a variety of imposed environments to yield information on the reduction of metal ores. In organic chemistry the technique has been widely used to study the degradation of polymers and to investigate the pyrolysis of coals, peats and bitumens. TG can also be used to record isothermal and isobaric weight changes. When a specimen is heated in ambient air or any gas of comparable density, the apparent weight changes with temperature due to the change in weight of the displaced gas; the specimen then appears to gain weight on heating.

Evolved Gas Analysis Useful information can often be gained by studying the gases which arise through the thermal decomposition of compounds. Evolved Gas Analysis (EGA) apparatus can be of two kinds: a) That in which the gas is evolved inside the gas detector or analysing apparatus, enabling the detection of transient species evolved during the course of the decomposition reaction. Diculties can arise in the control of the temperature, pressure and degree of dilution of the sample environment. b) That in which the evolved gas is led into a separate detection system, enabling more exible apparatus design, the possibility of pretreating the gas before analysis, and generally better control of the specimen environment. The possibility of secondary reactions occurring during the passage of the gases from the source to the detector cannot be discounted. Many types of detectors have been used for evolved gas analysis; those which sense the thermal conductivity of gasses are popular, but not very discriminating; gasdensity and ionisation detectors are also used. Gases can be absorbed in suitable solvents for subsequent chemical analysis, and the use of mass spectrometers for highresolution analysis is common. References to Thermal Analysis Techniques Books and Reviews 1 M. I. Pope and M.J. Judd: Dierential Thermal Analysis, London, Heyden, 1977. Analysis, 1,2, London, Academic Press. 2 J. Sestak, V. Satava and W.W. Wendlandt: Thermochimica Acta, 7, 333 (1973). 3 A. L. Greer: Ph.D. Thesis, University of Cambridge (1979). 4 V. Konryushin and L.N. Larikov: J. Mat. Sci., 13, 1 (1978). Specic References 5 A. L. Greer: Thermochimica Acta, 42, 193 (1980). Curie point measurement in the DuPont DSC system. 6 R. A. Baxter:Thermal Analysis, 1, 65 (1969) - eds. R.F. Sswenker and P.D. Garn, New York, Academic Press. This is the original paper on the Dupont DSC system. 7 H. K. Yuen, C.J. Yosel: Thermochimica Acta, 33, 281 (1979). Specic heat measurement in the DuPont DSC system. 8 L. M. Clareborough, M. E. Hargreaves, D. Mitchell and G. West: Proc. Roy. Soc., A215, 507 (1952). Description of the original DSC concept, and measurement of the stored energy of deformation. 9 M. G. Scott and P. Ramachandrarao: Mat. Sci. & Eng. 29, 137 (1977). Deals with the establishment of the baseline and some kinetic investigations. 10 M. G. Scott: J. Mat. Sci., 13, 291 (1978). Kinetics of glass crystallisation. Anal. Chem., 29, 1702 (1957). Deals with the Kinetics of non-isothermal, homogeneous transformations.

11 D. W. Henderson: J. Non-Cryst. Sol., 30 301 (1979). Kinetics of non-isothermal, nucleation and growth transformations. 12 E. A. Marseglia: J. Non-Cryst. Sol., 4, 31 (1980). Kinetics of non-isothermal nucleation and growth transformations. 13 J. W. Christian: Theory of Transformations in Metals and Alloys. 2nd ed., (1975), Part 1, Pergamon Press, Oxford. Elegant and comprehensive review on the formal theory of transformation kinetics. 14 K. Matusita and S. Sakha: Non-Cryst. Sol., 38-39, 741 (1980). Kinetics of glass crystallisation. 15 A. P. Gray: Thermochimica Acta, 1, 563 (1970). Polymer crystallisation. 16 R. O. Simmons and R.W. Ballu: Phys. Rev., 117, 52 (1960) & 125, 862 (1962). Dilatometric measurement of vacancy concentrations. 17 M. Perakh: Surface Technology, 4, 527 (1976) & 4, 538 (1976). Dilatometric determination of stress. 18 T. Inoue and B. Raniecki: J. Mech. Phys. Sol., 26, 187 (1978). Use of dilatometry in the prediction of thermal hardening and transformation stresses. 19 J. J. Stobo and B. Pawelski: J. Nuc. Mat., 4, 109 (1961). The inuence of texture on the anisotropy of thermal expansion. 20 J. Valentich: J. Mat. Sci., 14, 371 (1979). General dilatometry. 21 W. G. Hall and T.N. Baker: Phase Transformations, York Conference, (1979), No. 11, Series 3, 2, 11-90. Resistively changes during the ageing of steels. 22 J. Bass: Advances in Physics, 21, 431 (1972). Resistance measurements and the deviations from Mathiessons rule.