Physical aging of thermosetting powder coatings

Transcription

1 Progress in Organic Coatings 55 (2006) Physical aging of thermosetting powder coatings S. Montserrat, Y. Calventus, J.M. Hutchinson Departament de Màquines i Motors Térmics, ETSEIT, Universitat Politècnica de Catalunya, C/Colom 11, Terrassa, Spain Received 10 October 2005; received in revised form 24 October 2005; accepted 25 October 2005 Abstract The physical aging behaviour of several thermosetting powder coatings has been studied in order to compare the relative stabilities of alternatives to those based on triglycidyl isocyanurate (TGIC). The aging behaviour at ambient temperature is studied by differential scanning calorimetry and is analysed by the peak shift method in order to determine the kinetic parameters of the enthalpy relaxation process. Comparison between the various systems based on carboxyl functional polyester cured with bisphenol-a epoxy resins (hybrid systems), with triglycidylisocyanurate (TGIC) and -hydroxyalkylamide shows that the aging rate, expressed as the enthalpy loss per decade of aging time, is smallest for those based on TGIC, and hence that the performance of the alternative systems requires improvement as regards their stability Elsevier B.V. All rights reserved. Keywords: Physical aging; Differential scanning calorimetry; Enthalpy relaxation; Organic powder coatings 1. Introduction Some of the most widely used thermoset powder coatings during the last few decades have been the systems based on carboxyl terminal polyester (CPE) and epoxy resins. In particular, powder coatings based upon CPE cross-linked with triglycidylisocyanurate (TGIC) have dominated the market for exterior use, basically because of their outstanding outdoor durability and chemical resistance [1 4]. However, it has been known for some time now that the TGIC systems might have unacceptable health risks, and have been designated in Europe as a toxic product. There has consequently been much activity in the search for toxicologically safe cross-linking agents or for alternative systems. One such non-toxic cross-linking agent, marketed under the name Primid, is a -hydroxyalkylamide, the cross-linking mechanism being an esterification of the amide groups with the carboxyl groups from the binder. Whilst being non-toxic, though, Primid suffers from the disadvantage that it is reportedly not as stable as TGIC in severe test conditions [1,2]. Other alternative cross-linking agents are epoxies based on bisphenol- A, for which there is a need to establish their stability. Previous work with TGIC-based coatings [5] has compared the relaxation behaviour in the glass transition region determined by different techniques, namely dielectric, mechanical Corresponding author. Tel.: ; fax: address: Montserrat@mmt.upc.edu (S. Montserrat). and calorimetric analysis. Furthermore, it has been reported that the study of the physical aging of organic powder coatings is a sensitive method of establishing the stability of their mechanical properties in service [1 4,6]. Physical aging in general refers to the relaxation that occurs in the glassy state of any property that depends on the structure, as a consequence of the non-equilibrium state below the glass transition temperature, T g. Properties commonly used to follow the process of physical aging include the specific volume and the enthalpy, and dynamic mechanical and dielectric properties, among others. In the present paper we report the information that can be obtained by differential scanning calorimetry (DSC) to study physical aging by means of enthalpy relaxation in order to compare the stabilities of powder coatings based on carboxylated polyester resins cross-linked with different systems, including TGIC, Primid and bisphenol-a epoxies. 2. Experimental 2.1. Materials All the powder coatings are based upon a carboxylated polyester resin, and both cross-linked and uncross-linked systems have been studied. Three cross-linking agents have been used: TGIC, bisphenol-a epoxy and -hydroxyalkyl amide, for which the resulting powder coatings are denoted, respectively, as PTGIC, RP7510 and Primid The polyester resin used /$ see front matter 2005 Elsevier B.V. All rights reserved. doi: /j.porgcoat

2 36 S. Montserrat et al. / Progress in Organic Coatings 55 (2006) Table 1 Composition of the powder coatings Sample Cross-linking agent (c.a.) Polyester (%) c.a. (%) Pigments + additives (%) Ratio polyester/c.a. PTGIC TGIC /7 RP7510 Bisphenol-A epoxy /30 Primid Hydroxyalkyl amide / T 100 in the Primid coating is denoted as 8400-T. This nomenclature was assigned by Cray Valley Ibérica, who supplied all the materials. The compositions of the powder coatings in respect of the content of polyester and cross-linking agent, and of pigments and additives, where appropriate, are listed in Table Thermal analysis equipment The DSC that was used in this study for all thermal analysis procedures was a Mettler Toledo DSC 821e with intracooler. Dry nitrogen with a flow rate of 50 ml/min was used as the purge gas for all experiments. Temperature and heat flow calibrations were performed with indium, and the furnace was cleaned periodically by heating to 600 C with an air gas flow. Data analysis was performed by means of the Mettler Toledo STAR software in order to determine glass transition temperatures, heats of reaction during cure and relaxation enthalpies Cure schedules In order to cross-link the coatings, samples of 5 6 mg were encapsulated in standard aluminium pans and submitted to a cure schedule in the DSC, as follows. First, the sample of powder coating was heated at 10 K/min to 260 C to cure the sample nonisothermally. During this scan, there is an endothermic peak in the glass transition region of the uncured resin, approximately in the range C depending on the cross-linking agent, which in TGIC and hybrid systems is followed by a broad exothermic peak due to the cross-linking reaction, which occurs at much higher temperature, roughly between 120 and 240 C. In the Primid sample a broad endotherm is observed due to the liberation of water during the reaction, and for this reason it is not possible to follow the cross-linking reaction kinetics at atmospheric pressure by calorimetry for these systems. After this first scan, the sample is cooled to 10 Cat 20 K/min, stabilised at 10 C for 5 min and then reheated at 10 K/min to 150 C. During this second scan the glass transition of the cross-linked coating is observed Physical aging schedules Samples were submitted to physical aging at two aging temperatures: (i) 25 C, representative of ambient temperature, for RP7510, Primid 6594 and 8400-T and (ii) T g 20 K, i.e. 20 C below the glass transition temperature, for PTGIC. The glass transition temperature of the cured samples is known previously from the second scan following the non-isothermal cure, as described immediately above. For the uncross-linked samples, the glass transition temperature was determined by first heating each sample at 10 K/min to about 90 C, sufficiently far above T g such that the effects of the unknown previous thermal history are eliminated, then cooling it at 20 K/min to 25 C before finally reheating it at 10 K/min to about 90 C again. The T g is determined from this last heating scan. Samples for physical aging were heated to above their T g for a few minutes, and then cooled at 20 K/min to the aging temperature, T a, in the DSC. For aging times less than 12 h, the samples were aged in situ in the DSC. For longer aging times, the samples were removed from the DSC and placed in glass tubes, these tubes then being filled with nitrogen, sealed and placed in a thermostatically controlled silicone oil bath (Techne Tempunit TU-16D) set at the required aging temperature, T a. In this way, aging times, t a, spaced approximately equally on a logarithmic time scale, of up to more than 2000 h were achieved DSC experiments After aging a sample at the required temperature for the required time, it was then placed in (or, for t a < 12 h, already was in) the DSC, previously stabilised at the same aging temperature, T a, and was allowed to rest there for 2 min. It was then cooled at 20 K/min to approximately 20 K below T a, and was then submitted to a first scan at 10 K/min to about 90 C, well above T g. The sample was then again cooled at 20 K/min to the same lower temperature of T a 20 K before submitting it to a second scan at 10 K/min over the same temperature range as for the first scan. These two scans represent the response of the aged and unaged (or reference) samples, respectively, to the heating rate in the DSC, from which the enthalpy lost during the aging process can be determined, in a way to be described below Intrinsic cycles In order to determine the apparent activation energy of the enthalpy relaxation process, the procedure proposed by Moynihan et al. [7] was used. This procedure makes use of so-called intrinsic cycles, in which the sample is cooled at a certain rate from equilibrium above T g to a temperature in the asymptotic glassy region, and then immediately reheated at 10 K/min through the glass transition region. The glass transition temperature for the prior cooling rate is obtained from this heating scan. By using a range of cooling rates, in the present case from 0.5 to 20 K/min, the dependence of T g on the cooling rate can be determined.

3 S. Montserrat et al. / Progress in Organic Coatings 55 (2006) and considers the structure of the glass to be defined by the fictive temperature. The non-linearity parameter x can be determined from the dependence of T p on the enthalpy loss during aging, δ H,bythe peak shift method [12]. The procedure is first to evaluate relaxation rates per decade for both T p and δ H as dt p /dlog t a and dδ H /dlog t a, respectively, ensuring that the cooling and heating rates in the DSC remain constant. From the ratio of these rates, the dimensionless quantity F(x) is obtained as: F(x) = C p dt p dδ H (2) Fig. 1. Upper curves show DSC scans at 10 K/min for uncross-linked PTGIC sample aged for t a = 28 days at T a =51 C (full line) and for unaged sample (second scan, dashed line). Lower curve shows difference between these two scans, which is integrated between the limits shown to give the enthalpy loss during aging at 51 C. The peak endotherm temperature during the first scan is identified as T p. 3. Data analysis 3.1. Enthalpy loss during aging In the DSC experiments described in Section 2.5 above, a sample aged at temperature T a for time t a is submitted to two consecutive scans in the DSC: first, the scan of the aged sample, and then the second scan for the unaged or reference sample. The STAR software is used to subtract the second scan from the first, to leave a single curve, of the difference in heat flow versus temperature, which is integrated between temperature limits in the glassy and equilibrium rubbery states to give the enthalpy lost during the aging process. Fig. 1 illustrates this analysis for an uncured PTGIC sample aged for 28 days at 51 C Peak shift analysis The DSC scan of the aged sample gives rise to the familiar endothermic peak, which occurs at a temperature T p, shown in Fig. 1. This temperature depends on the whole previous thermal history of the sample since it was last in equilibrium above T g, in a way that can be predicted from the KAHR model analysis [8] of the relaxation kinetics in the glass transition region, which separates the temperature and structure contributions to the relaxation time τ, according to an equation which can be written in the form [9]: [ x h ] (1 x) h τ(t, T f ) = τ g exp + h (1) RT RT f RT g In Eq. (1), τ g is the relaxation time in equilibrium at temperature T g, x the non-linearity parameter (0 x l), h * the apparent activation energy for enthalpy relaxation, R the universal gas constant and T f is the fictive temperature of the glass. This equation is frequently referred to as the Tool Narayanaswamy Moynihan (TNM) equation [7,10,11], where C p is the increment in specific heat capacity at the glass transition. The quantity F(x) is a function of x which has been determined theoretically [13], known as the master curve, and which displays a sensitive dependence on x. Hence, the experimental determination of F(x) according to Eq. (2) allows the evaluation of x directly from the master curve Determination of apparent activation energy The glass transition temperature T g depends on the cooling rate q 1 in a way that is determined by the apparent activation energy h *. This dependence may be written [7]: dln q 1 d(1/t g ) = h R where h * /R is referred to as the reduced apparent activation energy. In practice, it is the fictive temperature T f that is measured during the heating scan in the DSC, but, for a glass reheated immediately after cooling at rate q 1 the fictive temperature of the glass is identical to T g (q 1 ). For this reason, the experiments for the determination of h * make use of intrinsic cycles at various cooling rates. The fictive temperature is evaluated by the equal area method [14], which is available as part of the STAR software Estimation of non-exponentiality parameter It is well known that the relaxation kinetics in the glass transition region involve a distribution of relaxation times. A convenient way of introducing a distribution into the analysis is by means of the Kohlrausch Williams Watts (KWW) stretched exponential response function [15]: [ ( t ) β ] φ(t) = exp (4) τ where β (0 β 1) is the non-exponentiality parameter. The smaller is β, the wider is the distribution of relaxation times. It has been shown [16] that it is possible to estimate the value of β, within some limits, from an analysis of the so-called upper peak heights in the DSC heating scan of the same intrinsic cycles as are used for the determination of the apparent activation energy, in Section 3.3 above. The procedure is to compare the dependence of these upper peak heights on the prior cooling rate (3)

4 38 S. Montserrat et al. / Progress in Organic Coatings 55 (2006) Table 2 Values of T g, T a, C p and h * /R for samples studied Sample a T g ( C) T a ( C) T g T a ( C) C p b (J/g K) h * /R (kk) RP7510 nc ± ± 6 RP7510 c ± ± 7 Primid 6594 nc ± ± 6 Primid 6594 c ± ± 8 PTGIC nc ± ± 12 PTGIC c ± ± T ± ± 6 a c and nc refer to cross-linked and not cross-linked, respectively. b Values of C p are calculated per unit mass of reactive material (resin plus cross-linking agent). with a theoretical dependence derived using the KAHR model and the KWW response function of Eq. (4). 4. Results and discussion First, the glass transition temperatures and the specific heat capacities of the various samples were determined by DSC during a heating scan at 10 K/min immediately after cooling at 20 K/min. The glass transition temperature T g is determined as the mid-point between the glassy and rubbery asymptotes, at which temperature the increment in specific heat capacity C p is also measured. The values of T g and C p are given in Table 2 for all the samples studied, including both cross-linked and uncross-linked materials. It should be noted that C p is calculated with respect to the mass of reactive material (resin plus cross-linking agent). For the materials that are cross-linked, the degree of crosslinking is not very high; there is very little difference in C p between the cross-linked and uncross-linked samples, and only a small but systematic increase in T g that results from crosslinking. On the other hand, the values of T g for the TGIC-based coatings, both cross-linked and uncross-linked, are significantly higher than those for the alternative coating systems, which will be seen later to correlate with the stability of the coating systems in terms of their physical aging rates. Since the T g values for the various systems differ significantly, and since many aging experiments were conducted at a fixed aging temperature T a of 25 C, this means that aging at 25 C involves different temperature intervals below T g for the different systems, which has an influence on the rate of physical aging. For this reason, Table 2 gives also the values of T g T a for each system. A typical example of a set of heating scans from intrinsic cycles involving cooling rates in the range 0.5 to 20 K/min is shown in Fig. 2 for the uncross-linked sample RP7510. The STAR software is used to determine the fictive temperature T f for each cooling rate, q 1, following the method of Richardson and Savill [14], and ln( q 1 ) is plotted against the reciprocal of T f in Fig. 3. The least-squares best-fit line through the data is shown, and corresponds to a reduced apparent activation energy of 121 kk. For each sample, at least two sets of intrinsic cycle experiments were performed, and the average values are given in Table 2, where it can be seen that the uncertainty in the value of the apparent activation energy is of the order of 5 10%. Fig. 2. Heating scans at 10 K/min in the DSC immediately after cooling through the glass transition region at the rates indicated against each curve, for uncrosslinked sample RP7510. These results indicate that taking into account their uncertainty there is no significant difference between the activation energies of the cross-linked and uncross-linked samples. In fact, all systems have a high activation energy, implying a strong temperature dependence, the lowest value being displayed by Primid and its base resin, 8400-T. High values of activation energy can be advantageous, in that they imply greatly increased relaxation times in equilibrium at temperatures below T g, which implies greater stability under these conditions. However, the usual state below T g is one of non-equilibrium, and hence it is not the activation energy alone which determines the relaxation Fig. 3. Activation energy plot of log(cooling rate) vs. reciprocal fictive temperature for uncross-linked sample RP7510.

5 S. Montserrat et al. / Progress in Organic Coatings 55 (2006) Fig. 4. DSC heating scans at 10 K/min for uncross-linked sample PTGIC aged at 51 C, i.e. at T g 20 K, for the times indicated in hours. time at any temperature, but it is the way in which the relaxation time depends on both temperature and structure which is critical. This last is controlled by the non-linearity parameter, x, defined in Eq. (1), and which can be evaluated by applying the peak shift method to the DSC aging experiments, as follows. Fig. 4 shows a set of DSC scans at 10 K/min for a sample aged for the different times indicated at 20 K below T g. These results show the typical effect of aging time on the response of any glassy system to heating through the glass transition region: the endothermic peak grows in magnitude, with the area of the aged peak relative to that for the unaged sample giving the enthalpy loss during aging, and the peak temperature shifts to higher values, representative of the increase in relaxation time during aging and hence the need to go to higher temperatures in order that the relaxation time should decrease to a value commensurate with the experimental timescale imposed by the heating rate. The enthalpy loss and the increase in peak temperature both display approximately linear dependences on the logarithm of aging time, which implies the need to extend the aging experiments to rather long aging times (in the present work, as well as in earlier studies (e.g. Refs. [17 19]), aging times of more than 2000 h have been used) in order to determine with greater precision the slopes of these linear dependences. Sets of experimental data similar to those shown in Fig. 4 have been obtained for all the samples, both cross-linked and uncrosslinked, at the aging temperatures listed in Table 2. From these sets of data, the values of enthalpy loss and peak temperature have been determined and are plotted versus log(aging time) for RP7510, Primid and PTGIC in Figs. 5 7, respectively. Also included in Fig. 6 are the data for the polyester resin 8400-T. From these plots, the values of the slopes of the best-fit least- Fig. 5. (a) Dependence of enthalpy loss, δ H, on logarithmic aging time (in hours), log(t a ), for uncross-linked RP7510. (b) Dependence of peak temperature, T p,on logarithmic aging time (in hours), log(t a ), for uncross-linked RP7510. The open symbols relate to upper peaks (see text). The lines indicate the least-squares fits to the data indicated by closed symbols, which relate to main peaks (see text). squares lines have been obtained for each sample, and are listed in Table 3. In fact, the data for the cross-linked RP7510 are not shown, as they do not display the appropriate peaks necessary for the peak shift analysis. For these samples, the endothermic peaks appear similar to so-called sub-t g peaks [20], with the peak temperatures for the shortest aging times occurring below T g, and hence they do not represent well-stabilised glasses, as is required of the peak shift method. The reason for this is that the aging temperature for these samples is 42 K below T g,a significantly greater difference than for any of the other samples studied here. A similar phenomenon occurs for the dependence Table 3 Values of R H, R T, F(x), x and β for samples studied Sample R H a (J/g) R T a (K) F(x) x β RP7510 nc 1.85 ± ± ± ± RP7510 c Primid 6594 nc 1.32 ± ± ± ± Primid 6594 c 1.52 ± ± ± ± PTGIC nc 0.96 ± ± ± ± PTGIC c 0.97 ± ± ± ± T 1.13 ± ± ± ± a Rates are per decade; R H is calculated per unit mass of reactive material.

6 40 S. Montserrat et al. / Progress in Organic Coatings 55 (2006) Fig. 6. (a) Dependence of enthalpy loss, δ H, on logarithmic aging time (in hours), log(t a ), for Primid, both cross-linked ( ) and uncross-linked ( ), and for the base resin 8400-T ( ). (b) Dependence of peak temperature, T p, on logarithmic aging time (in hours), log(t a ), for Primid, both cross-linked ( ) and uncross-linked ( ), and for the base resin 8400-T ( ). The open circles relate to upper peaks (see text) for the base resin 8400-T. The lines indicate the least-squares fits to the data related to main peaks (see text). Fig. 7. (a) Dependence of enthalpy loss, δ H, on logarithmic aging time (in hours), log(t a ), for PTGIC, both cross-linked ( ) and uncross-linked ( ). (b) Dependence of peak temperature, T p, on logarithmic aging time (in hours), log(t a ), for PTGIC, both cross-linked ( ) and uncross-linked ( ). The full line for the cross-linked system and the dashed line for the uncross-linked system indicate the least-squares fits. of the peak temperature on aging time for the uncross-linked RP7510 and 8400-T samples at short aging times, as can be seen in Figs. 5b and 6b, respectively. Here, instead of appearing as sub-t g peaks, the endotherms for these short aging times display peaks at temperatures which do not fit the trend of those for longer aging times. Again, the reason for this is that these samples are not well-stabilised glasses before heating, and such peaks are referred to as upper peaks [16] in contrast to main peaks which appear for well-stabilised glasses. Consequently, when evaluating the dependence of T p on log(aging time), data points corresponding to these upper peaks are ignored (and are shown in Figs. 5b and 6b as open symbols), and only the longer aging time data is used. Whether sub-t g peaks or upper peaks occur in any given situation depends largely on the distribution of relaxation times, and hence on β, which is discussed shortly. The uncertainties in the values of the slopes R H and R T quoted in Table 3 are determined at the 75% confidence level using the Student s t distribution [21]. It can be seen that in some cases the uncertainties are rather large. This arises from the use of often much larger differences (always at least 20 K, occasionally much greater) between T g and T a than would normally be appropriate for the evaluation of x by the peak shift method [12]. This selection of the aging temperatures T a was made because one purpose of this investigation was to study the aging affects at ambient temperature, but it limits the range of logarithmic aging time over which the evaluation of the slope can be determined, since short aging times do not yield the required well-stabilised glasses, as has been observed above. Nevertheless, it is still possible to draw some conclusions from these results. First, and most significantly, the aging rate as expressed by the quantity R H, which represents the enthalpy loss per decade, is markedly smaller for PTGIC than for the alternative paint systems RP7510 and Primid. Furthermore, this aging rate for PTGIC is evaluated only 20 C below T g, and hence at a higher temperature relative to its T g than for either of the other two systems. As a consequence, it can be concluded with some certainty that the original PTGIC system is significantly more stable to aging, as determined by enthalpy relaxation, than are the alternatives investigated here. If, as is often assumed to be the case, the aging of mechanical properties such as creep compliance and strength mirrors the enthalpy relaxation, then the mechanical durability of RP7510 and Primid does not match that of PTGIC.

7 S. Montserrat et al. / Progress in Organic Coatings 55 (2006) Second, there appears to be no significant difference between the rates as expressed by R H for the cross-linked and uncrosslinked systems. This seems surprising, since it might be anticipated that the restrictions on molecular mobility imposed by the cross-linking would reduce the enthalpy relaxation rate, particularly as the glass transition temperature is increased by cross-linking, albeit by only about 10 C (see Table 2), so that the cross-linked samples at 25 C are relatively further from T g than are the uncross-linked samples. The implication of this is that the increase in T g resulting from cross-linking cannot be assumed to be indicative of a reduction in all aspects of molecular mobility. Indeed, if the values of R T are compared for the cross-linked and uncross-linked systems, then in one case (PTGIC) there is a significant increase as a result of cross-linking, whereas in another case (Primid) there is a significant decrease. Again, the implication is that it is not advisable to draw conclusions about structural stability simply from a comparison of glass transition temperatures. In fact, R T depends in a rather complex way on the structural relaxation kinetics, which are controlled to a significant degree by the non-linearity parameter x. From the values of R H and R T listed in Table 3, and using the values of C p for each sample, listed in Table 2, it is possible to calculate F(x) from Eq. (2). These values of F(x) are given in Table 3, together with their corresponding uncertainties, and from these the corresponding values of x are then determined from the master curve. It can be seen that the values of x all fall roughly in the range , which is typical of polymers in general. However, the uncertainties in the values of x for the different powder coatings studied here indicate no significant differences between most of these systems at the 75% confidence level, and in particular that there is no significant effect of cross-linking. The smaller average values of x for PTGIC, both cross-linked and uncross-linked, suggest that the enthalpy relaxation process in PTGIC may be more dependent on the structure than the other systems, which may be related to the greater stability of PTGIC, but this interpretation must be regarded as tentative. The values of the non-exponentiality parameter β are determined from the upper peak heights of the intrinsic cycles, such as those shown in Fig. 2 for the uncross-linked sample RP7510. A plot of the normalised upper peak heights, Cp,u N, as a function of the logarithm of the ratio of cooling rate to heating rate is shown in Fig. 8 for the system 8400-T. These data are compared with theoretical curves using the value of x calculated by the peak shift method and given in Table 3, and selecting the value of β which gives the best fit. It should be stressed that this is only an approximate method for estimating β, and only selected values of β (0.2, 0.3, 0.456, 0.6) are used. The most interesting result is the significant reduction in β when the PTGIC sample is cross-linked. This implies that the effect of cross-linking is to broaden the distribution of relaxation times, which may help to explain why this system is the most stable. If the broadening of the distribution occurs by an extension to longer relaxation times, then the enthalpy relaxation process will occur more slowly in the cross-linked sample than in the uncross-linked sample at the same temperature interval T g T a, and this can be seen to be Fig. 8. Normalised upper peak height C N p,u as a function of log( q 1 /q 2 ) for 8400-T. Experimental data for two different samples shown as open symbols; theoretical dependence for x = 0.52 and β = 0.6 shown as full line. the case in Fig. 7a where the best-fit line for the cross-linked system lies below that for the uncross-linked system, implying a reduced enthalpy loss for the former for any given aging time. Likewise, Fig. 7b shows that the cross-linked sample displays endothermic peaks at considerably higher temperatures than for the uncross-linked sample, which may again be explained by the need for the former to reach higher temperatures before the enthalpy loss resulting from the longer relaxation times can be recovered. 5. Conclusions The effect of cross-linking is to increase the glass transition temperature in all the paint systems studied here, but this does not correlate with its effect on the aging rate as expressed by the quantity R H, which represents the enthalpy loss per decade. For all the samples for which this rate could sensibly be measured, there was no significant difference between cross-linked and uncross-linked samples. On the other hand, the absolute value of the enthalpy loss for any given aging time was always smaller for the cross-linked samples. Comparison between the different powder coatings, though, shows that the PTGIC system is the most stable if R H is used as a measure of stability, as has been suggested, which means that the environmentally more acceptable alternative systems to PTGIC are less attractive in respect of the retention of their properties during aging. An analysis of the aging behaviour of all these systems yields values of the non-linearity and nonexponentiality parameters, x and β, respectively, which do not, on the whole, show large variations between the different systems. This may be attributed to the fact that the degree of cross-linking is limited, as well as to the difficulty of determining their values precisely in view of the generally rather large temperature interval between T g and T a which necessarily results from studying the aging behaviour at ambient temperature. One noticeable effect, however, is that the distribution of relaxation times is significantly broader for the cross-linked PTGIC system, to which the greater level of stability in this system may be attributed.

8 42 S. Montserrat et al. / Progress in Organic Coatings 55 (2006) Acknowledgements Financial support has been provided by CICYT Project MAT C The authors are grateful to Cray Valley Ibérica for supplying polyester resin and the powder coatings. This work has been facilitated by the award of a Ramon y Cajal grant to J.M.H. References [1] P.G. de Lange, Powder Coatings, Chemistry and Technology, second ed., Vincentz, Hannover, 2004 (Chapter 3; updated edition of the book by T.A. Misev). [2] T.A. Misev, R. van der Linde, Prog. Org. Coat. 34 (1998) 160. [3] R. van der Linde, E.G. Belder, D.Y. Perera, Prog. Org. Coat. 40 (2000) 215. [4] K.H.M. Cansen, Saint Coat. Ind. 10 (1994) 58. [5] X. Ramis, Y. Calventus, A. Cadenato, F. Roman, J.M. Morancho, P. Colomer, J.M. Salla, S. Montserrat, Prog. Org. Coat. 51 (2004) 139. [6] D.Y. Perera, Prog. Org. Coat. 47 (2003) 61. [7] C.T. Moynihan, A.J. Easteal, M.A. DeBolt, J. Tucker, J. Am. Ceram. Soc. 59 (1976) 12. [8] A.J. Kovacs, J.J. Aklonis, J.M. Hutchinson, A.R. Ramos, J. Polym. Sci. Phys. 17 (1979) [9] J.M. Hutchinson, J. Therm. Anal. Cal. 72 (2003) 619. [10] A.Q. Tool, J. Am. Ceram. Soc. 29 (1946) 240. [11] O.S. Narayanaswamy, J. Am. Ceram. Soc. 54 (1971) 491. [12] J.M. Hutchinson, M. Ruddy, J. Polym. Sci. Phys. 26 (1988) [13] A.R. Ramos, J.M. Hutchinson, A.J. Kovacs, J. Polym. Sci. Phys. 22 (1984) [14] M.J. Richardson, N.G. Savill, Polymer 16 (1975) 753. [15] G. Williams, D.C. Watts, Trans. Faraday Soc. 66 (1970) 80. [16] J.M. Hutchinson, M. Ruddy, J. Polym. Sci. Phys. 28 (1990) [17] S. Montserrat, P. Cortés, A.J. Pappin, K.H. Quah, J.M. Hutchinson, J. Non-Cryst. Sol (1994) [18] J.M. Hutchinson, D. McCarthy, S. Montserrat, P. Cortés, J. Polym. Sci. Phys. 34 (1996) 229. [19] P. Cortés, S. Montserrat, J.M. Hutchinson, J. Appl. Polym. Sci. 63 (1997) 17. [20] M. Ruddy, J.M. Hutchinson, Polym. Commun. 29 (1988) 132. [21] M. Mulholland, C.R. Jones, Fundamentals of Statistics, Butterworths, London, 1968.