epoxy composites

PII: S1359-8368(97)00020-6

ELSEVIER

Composites Part B 29B (1998) 245-250 © 1998 Elsevier Science Limited Printed in Great Britain. All rights reserved 1359-8368/98/$19.00

Monitoring autoclave cure in commercial carbon fibre/epoxy composites

George M. Maistros and Ivana K. Partridge Advanced Materials Department, School of Industrial and Manufacturing Science, Cranfield University, Bedford MK43 OAL, UK (Received 26 July 1996;accepted 13 August 1997) Embedded dielectric sensors and thermocouples have been used to monitor the state of cure, in real time, for a range of current commercial epoxy resin carbon fibre reinforced composites. This paper summarises the methodology and data analyses involved in the monitoring procedure and gives specific examples of the results obtained. The generic rules governing the relationships between dielectric properties and material transformations for epoxy-based resins are stated. The problem of reaction exotherms in thick sections and the resulting uneven temperature and property distributions are quantified for two specific resin types. The reduction in intertaminar shear strength, resulting from the presence of the embedded dielectric sensors, is shown to be 15% for a brittle resin system and 5% for a tough resin system. The results and information obtainable from the dielectric cure monitoring are set against the background of the development of a comprehensive overall model for composite cure. © 1998 Elsevier Science Limited. All rights reserved (Keywords: B. curing; monitoring; dielectric; modelling; composite)

INTRODUCTION Dielectric methods for monitoring thermoset cure have been developed recently and used to probe the different stages of resin cure in advanced composites processing 1-3. The ultimate aim is to provide intelligent closed-loop control o f composite manufacturing processes, such as autoclave cure and R e s i n - T r a n s f e r - M o u l d i n g (RTM). Changes in the dielectric behaviour of the resin under cure give real-time and in-situ information on the advancement of the process parameters and the arrival at a desirable level of degree of cure 4. Current work at Cranfield concerns the development of a complex model scheme intended to combine the dielectric monitoring responses with models of thermoset cure. A flow diagram, depicting the essential elements of the overall model, is shown in Figure 1. On the left-hand side of the figure, mathematical models are used to convert the t i m e temperature profile of the cure process environment to the resin viscosity and the glass transition temperature, Tg. On the right-hand side of the figure, real-time measurements from embedded dielectric sensors are used in combination with analytical dielectric models to predict the viscosity and the Tg of the resin in the vicinity of the sensors. It must be noted that at the present moment different parts of the overall model are in different stages of development. The heat transfer sub-model determines the time and conditions needed to establish the required thermal history

for the curing component. The fibre type, lay-up characteristics, tooling material and component size are the main parameters affecting the characteristics of heat transfer from the environment (i.e. the autoclave or the RTM mould) to the component during the cure cycle. This sub-model is most critical in thick component curing when a reaction exotherm generates a significant proportion of the heat available to cure the resin. For typical aerospace applications, the thermal properties of the tooling are usually the determining factor. Several software packages are available, designed to tackle the heat transfer problem 5'6. A kinetic model of resin cure provides the reaction rate (dcddt) input to the energy balance equations of the heat transfer model. Such kinetic models are commonly used to estimate the reaction rates and degree of cure of the resin, for a given thermal history. For the case of most epoxy resins, the model is a combination of a simple nth order reaction and of autocatalytic effects arising from the reactivity of the hydroxyl groups generated during the cure. In general, cure reactions of thermosets carried out under isothermal conditions can be expected to be incomplete. This observation is explained in terms of diffusion control effects on the cure reaction rates, which are significant after the liquid-to-rubber transformation has been reached, and result in almost negligible reaction rates after the rubber-to-glass transformation. A specific model, recently developed at Cranfield, combines the nth order kinetics, autocatalytic and diffusion control aspects to

245

Monitoring autoclave cure: G. M. Maistros and I. K. Partridge t. Te.vl,

I

I

(HeatTransferModel]

(DielectricMeasurements 1

1

1

t, TraaterlaI

Capacitance

I

Conductance

[ KineticModel ~ l

( DEASoflware ) 1

Degree of cure ~t

Permittivity U Dielectric Loss e" --7

(StructuralModel] )

,

Chemoviscosity~

'Glass Transition Temperature Tg ~

J

I

|. -']1

[Rel xation Model)

Viscosity i"1

~- ? ~

Vitrification /

by the Cole-Cole equation:

Dielectric Sensors

~lonicConductivity(t~)~ ~3, Model .) I" ~X.~/~"'-~ Maximum

Flow

/

( ,2 logo" = 0

~Gelation

ed=e'--ie"--

(e0--e~) Fe~ 1 + (iw~'0) ~

(1)

in which w is the angular frequency, e~ is the infinite frequency permittivity, e0 is the static permittivity and/3 is the parameter representing the distribution of dipole relaxation times around the average value r0. By examining the evolution of the permittivity and dielectric loss values during isothermal cure reactions of epoxy resins, it has been observed that9: (1) the infinite frequency permittivity e~ is independent of cure time and temperature and ranges between 2 and 4; (2) the static permittivity e0 is linearly dependent on the degree of cure and decreases with increasing temperature; (3) the changes in the average relaxation time r0 reflect changes in the resin Tg; the relationship is non-linear, with log r0 being proportional to the square of the Tg. For a constant cure temperature, the changes in the logarithmic relaxation time exhibit a linear relationship with cure time 1°. Ionic conductivity effects become apparent in the dielectric loss signal. The ionic component of the measured dielectric loss shows a reciprocal dependence on the measurement frequency: (7

Figure 1 Flow diagram depicting the essential elements of an overall model of thermoset cure

achieve a very good fit to the isothermal cure behaviour of a new type of epoxy-based resin for aerospace applications (RTM 6) 7. The parameter that is most sensitive to the advancement of the cure process and indicative of the thermoset network density is the glass transition temperature Tg. The fact that Tg increases non-linearly with the degree of cure makes it more sensitive in the later stages of the cure when the reaction rates are low and Tg is close to the cure temperature. A structural model relates the values of 'degree of cure' to Tg. Provided that the kinetic scheme is such that the crosslinking characteristics of the molecular network depend only on the level of chemical conversion and not on the thermal history of the cure process, a unique one-to-one relationship between Tg and conversion can be expected. During the period in cure preceding gelation and vitrification, chemorheology principles can be used to estimate viscosity as a function of the advancement of the cure s. Current work at Cranfield concentrates on this aspect of the overall model development. Turning to the right-hand side of the model expressed in Figure 1, the measured dipolar portions of the permittivity and dielectric loss can be utilised to trace the advancement of the cure process, in real time. The dipole contributions to permittivity and dielectric loss can be separated from the ionic and interfacial components 9. The frequency dependence of the dipolar dielectric properties is described

246

ec"-- 2~rfe

(2)

where ee" is the conduction part of the dielectric loss, a is the conductivity in S/re,(is the test frequency in Hz and e is the permitfivity of free space (8.85 pF/m). The ionic conductivity is proportional to the ion concentration and mobility. As resin viscosity reflects the mobility of the chain segments in the growing network, an approximately inversely proportional relationship exists between viscosity ~ and ionic conductivity 09:

*l = A ' a - B

(3)

where A and B are resin specific and temperature dependent constants. The relationships expressed in eqns (2) and (3) are of great importance in cure monitoring, since they make it possible to follow, in-situ, the changes in the resin viscosity via direct measurements of the dielectric loss and thus the conductivity. Senturia and Sheppard ll observed that the conductivity does not fall to zero as viscosity tends to infinity after gelation. Therefore, it is expected that eqn (3) will begin to break down as gelation is approached and the time during the resin cure when d2(log a)/dt 2 = 0 is the first evidence of a gelled system. EXPERIMENTAL

Demonstrator structures for temperature distribution determinations Many of the resin systems commonly used as matrices in

Monitoring autoclave cure: G. M. Maistros and I. K. Partridge aerospace applications exhibit very considerable cure exotherms, which can have an overriding effect on the temperature distribution of a curing component. An experimental composite component used to quantify this problem for a simple case was a step structure, with thicknesses of the individual step section varying from 2 to 25 mm. It was made by hand lay up from unidirectional prepreg tape. The materials used were Fibredux GF/913 and Fibredux CF/920. The essential difference between these two materials was in the different thermal conductivities of the reinforcements. The resins have very similar cure cycles, involving imposed thermal dwells at 90 and 120°C. Thermocouples were embedded in the central sections of each step thickness and used to monitor the local temperature during the course of autoclave cure 12.

absolute values of the Tg reduction cannot be related to the different temperature overshoots in the two systems, because the chemical compositions of the two resins are different.

T

/

,

200+/

"*' 15o-/' 100-~//

~

50-

~-

0

/

/

~] /

/ -/ 200

100

-

~

300

/ 10mm section / 2rnm section //'Air temp.

Time (min)

(a) /

J

/

/ 200 // /

O" 150 / ' / 0...... 100

o.

-

•

//

I

l- • - i_

"

_ /

~

/

50

"/~

/'

/

25ram section

/15mm section ~

100

2mm section

~ / A i r Temp.

198

294

388

Time (rain)

(b)

RESULTS OF MONITORING OF COMPOSITE CURE

Figure 2 shows the imposed temperature profiles and the development of the resultant temperature distributions within the experimental stepped composite structures, in real time. As might have been expected, the effect of the reaction exotherm is much more pronounced in the glass fibre reinforced structure (Figure 2a), with the maximum temperature recorded in the thickest part of the structure exceeding the nominal cure temperature by 120°C. The corresponding difference in the case of the carbon fibre reinforced structure was less that 60°C (Figure 2b). Although the exotherm is short lived, it does appear to affect the final state of the resin, with a detrimental effect on the final Tg of the cured material. The figures on the top of each step thickness section, depicted in Figure 3a and Figure 3b, are the maximum temperatures recorded in the centre of the corresponding section. The figures in the centres of the step sections are the Tg values determined by DMTA on samples of the cured material cut from the appropriate parts of the structures. The drop in Tg from the thinnest to the thickest step sections is quite evident. The

~

250 / O" o_.

0

Temperature distribution

i-!__

/

Dielectric cure monitoring technique Microelectrodes were embedded in the required positions in the composite in the lay-up stage. They are thin, flat capacitative interdigitated sensors (Dek-Dyne), producing a 30/zm deep fringing electric field over their active surface. The conductive carbon fibres were kept away from the active sensor surface by using a thin porous PTFE film. The connection between the sensors and the data acquisition equipment (Solartron SI 1260A) was made via ultrathin PTFE sleeved low resistivity wires. Capacitance and conductance measurements, from multiple sensors, were made at fixed time intervals, throughout cure, over a frequency range from 1 Hz to 10 MHz. Data reduction was performed via an in-house produced software (DEA) package. Detailed description of the dielectric monitoring procedure can be found in previous publications 4"]3.

!

Figure 2 Temperature mapping, during autoclave cure, in identical stepped structures made from (a) Fibredux GF/913 and (b) Fibredux CF/ 920

229

---1

11

144 174

1108 110 "----~ 147 ---

116 119

3 126

121 124~ I

2mm

Figure 3 Maximum temperatures recorded during cure in the centres of corresponding step sections in (a) GF/913 and (b) CF/920 composite structures. The figures in italics are the Tg values determined by DMTA measurements (1 Hz) of final cured samples from the central parts of each step

247

Monitoring autoclave cure: G. M. Maistros and I. K. Partridge Dielectric monitoring

el point flection point)

Another epoxy-based resin, in common use in the aerospace industry, is used here to illustrate the information obtainable from the real-time dielectric cure monitoring. The particular system is Fibredux CF/924, requiring cure at 180°C. The applied autoclave cure cycle is indicated in Figure 4. The point of pressure application was selected by reference to changes in the dielectrical signal, as illustrated in Figure 5. The dielectric permittivity rises steeply from its initial value as the flowing resin penetrates the protective PTFE cloth and comes in contact with the electrode. The very high values of e ' measured at the low frequency end are an indication of electrode polarisation 9. The values of the permittivity begin to drop once the resin has undergone a significant amount of reaction and begins to approach gelation. The accompanying development in the c~-relaxation (Tg) is most evident at the higher frequencies where the electrode polarisation has a smaller effect on the measured properties. The fiat portion of the graph in the later stages of the cure is an indication that vitrification has been achieved in the resin. A fuller analysis of the dielectric data permits a more accurate determination of the individual events in the course of resin cure. Previous work in the laboratory has established that the inflection point in the graph of log e"

200

/

5O

I--

1.8 1.4 t~

~1.0 _o 0.6 -5.0

-,;.5

-,:.o

-a.5

-a.o

log o (S/m)

Figure 7 Relationship between unreinforced 924 resin viscosity, 7, measured on a cup-and-bob Contraves Rheomat system at shear rate of 100s-l}, and its ionic conductivity, o, under different isothermal conditions

0

trr=.now

-4 ~ ~

v

tO tO

.= 2 n

,8o

60

tgel

\

Figure 4 Imposed conditions of temperature and pressure during autoclave cure of a CF/924 composite component

pressure application: 7 bar

sensor wetting s-relaxation Dielectric permittivity output recorded during autoclave cure of a CF/924 composite component, with indications of sensor wetting, the point of external pressure application (viz. Figure 4) and the obscured c~relaxation

248

Figure 6 Dielectric loss output corresponding to data in Figure5, with identification of the region of maximum resin flow and the gelation region

6

Cure time (min)

Figure 5

minimum viscosity (maximum conductivity)

?

/

_~ lo0

.t.-,

0

\

8

150

E

\

E o

[

-5

0 -1

-6

-2

-7

-3

-8

0

50

1O0

150

o

-4 200

Cure time (min)

Figure 8 Different stages of resin cure, within the CF/924 composite, determined from the dielectric output during autoclave cure

against cure time (Figure 6) correlates well with arrival at a viscosity level of the order of 104 Pa s in many epoxy based resins 4'9. This high level of viscosity can, for practical purposes, be identified with gelation having taken place. The relationship between resin viscosity ~ and the frequency independent ionic conductivity a in the pre-gel region was verified experimentally for the unreinforced Fibredux 924 resin 4 and is shown in Figure 7. The linear relationship expected from eqn (3) is seen to hold and the temperature dependence of the parameters of the equation for this resin system is: log A = - 3 0 . 3 1 + 12500/Tcure and B = - 7 + 0,0182 × Tcu~e-The in-situ monitoring of changes in the resin conductivity with cure time therefore offers a means of following the initial viscosity changes in the resin

Monitoring autoclave cure: G. M. Maistros and I. K. Partridge Table 1

Effect of embedded 'sensor' on the interlaminar shear strengths of two unidirectional composites

ILSS (MPa)

CF/934 Control

With 'sensor'

CF/920 Control

With 'sensor'

62 -+ 5

57 -+ 5

56 ± 1

53 -+ 1

matrix within a curing composite. This is true for both isothermal and non-isothermal imposed conditions. The different stages of cure can then be depicted on a single graph of log a against cure time, as shown in Figure 8. The figure also shows the linear increase in the logarithmic relaxation time of the resin with increasing cure time, determined within the isothermal 180°C dwell step of the cure, obtained from analysis of the permittivity changes 4. The vitrification time tvit is obtained by extrapolating the relaxation time r to a quasi-static state (1 s).

2.5

P

2.0

-o

6i E

1.5

0~

1.0

Mechanical integrity considerations The use of embedded electrodes raises the obvious question of how the presence of such a 70/~m thick "defect" might influence the mechanical integrity of the composite structure. Table 1 shows a comparison of static short beam bending strength tests carried out on control and with "sensor" unidirectional beam samples of identical dimensions, for two different composite types. The Fiberite CF/934 material can be considered representative of the first-generation brittle composites whilst the Fibredux CF/ 920 material is exceptionally damage resistant. The samples with "sensor" were made by embedding a piece of Kapton material between the central plies of the sample during layup. In order to achieve a realistic assessment of the effect of the embedded electrode these Kapton pieces were of the same thickness and dimensions as the substrate of the DekDyne electrodes. The presence of the "sensor" has a detrimental effect on the interlaminar shear strength (ILSS) values in both types of composite. However, the percentage strength reduction is smaller in the tougher materialt2.

DISCUSSION The overall model as described in Figure 1 is designed to validate the models on the left half of the figure. These models require time and temperature as inputs and are derived from thermal analysis tests on the unreinforced resin system and on the composite. The achievable level of accuracy in the predictions of resin states (i.e. the Tg, viscosity, and the arrival at maximum flow, gelation and vitrification) from the models is expected to be determined by the nature of the resin system and the type of cure process(es) involved. It is the monitoring side of the overall model which has been the primary subject of our studies in recent years. The relationships of the dielectric properties to material

20 t,O Time at 160"C {rain)

60

Figure 9 Determination of vitrification of RTM6 resin by monitoring changes in the heat capacity by MDSC and from in-situ real time cure monitoring

transformations have been shown to obey some generic rules: (1) the maximum in the resin conductivity coincides with the point of minimum viscosity under isothermal or dynamic (thermal) conditions; (2) the inflection point in the drop of the logarithmic ionic conductivity after the point of maximum flow provides the first evidence of a gelled material; (3) the extrapolated maximum in the dipolar loss, at the frequency of 1 Hz, correlates well with the arrival at the vitrification stage of the cure, under isothermal conditions. Other workers 14 have based their analysis of the dielectric phenomena solely on the changes in conductivity o with time (see Figure 8). They also specify the point of inflection in the log cr vs cure time curve as being associated with gelation. Vitrification is identified as the reaction endpoint by the "flattening of the logarithmic reciprocal conductivity vs cure time curve". The negligible rate of change of ionic conductivity indicates the lack of any additional crosslinking which would further restrict the mobility of the ions. However, in practice this endpoint is difficult to assign with any precision and the method provides no predictive capability. We believe that our analysis method, based on the treatment of the relaxation time data, is superior and easier to implement. At first sight, the identification of resin relaxation time • of 1 s with arrival at vitrification might seem arbitrary. Very recent measurements made on a Modulated DSC apparatus provide direct experimental evidence that the coincidence of the two phenomena is in fact extremely good, at least for the resin system studied: the observed position in time of the change in the heat capacity of the resin, ACp, coincides with the extrapolated

249

Monitoring autoclave cure: G. M. Maistros and I. K. Partridge ACKNOWLEDGEMENTS

1o0 Ixm

This work is supported by EPSRC/DTI LINK (GR/J96222) and IMI (GR/K68523) programmes and by contributions from Short Brothers PLC, British Aerospace (Airbus and Sowerby Research Centre) and the MoD (DRA Farnborough). Supply of pre-preg from Hexcel Composites (UK), the MDSC data from Short Brothers and assistance with composite structure manufacture from the Cranfield Composites Centre are gratefully acknowledged.

REFERENCES (a)

(b)

(c)

Figure 10 Polished cross-sections of cured UD samples of CF/934 composite. Autoclave pressure applied (a) too early, (b) at optimum time as indicated by the dielectric signal and (c) too late

1. 2. 3. 4.

time to vitrification, tvit, from our dielectric measurements

(Figure 9). Finally, the potential usefulness of the real-time monitoring capability is illustrated in Figure 10. These micrographs are taken from polished cross-sections of unidirectional CF/934 samples autoclave cured under identical imposed thermal schedules but with the autoclave pressure deliberately applied (a) at the very beginning of cure, (b) at optimum time as identified by the dielectric signals and (c) after gelation of the resin. The microstructure of the samples is clearly influenced by the state of resin flow at which the pressure is applied )2'13. As some composite properties are known to be highly dependent on this level of microstructural detail, there now exists a direct means of influencing those properties in the curing stage.

250

5. 6. 7. 8. 9. 10. 11. 12. 13. 14.

Mijovic, J., Kenny, J.M., Maffezzoli, A., Trivisano, A., Bellucci, F. and Nicolais, L., Compos. Sci. Tech., 1993, 49, 277. Mangion, M.B.M. and Johari, G.P., J. Polym. Sci.: Part B: Polym. Phys., 1990, 28, 1621. Choi, J.H. and Lee, D.G., J. Compos. Mat., 1995, 29, 1181. Maistros, G.M. and Partridge, I.K., Compos. Sci. and Tech., 1995, 53, 355. Ciriscioli, P, and Springer, G.S., Smart Autoclave Cure of Composites. Technomic Publishing, Lancaster, PA, 1990. Bruschke, M.V. and Advani, S.G., SAMPE Quarterly, 1991, 23, 2. Karkanas, P.I., Partridge, I.K. and Attwood, D., Polymer International (in press). Mijovic, J. and Ott, J.D., J. Compos. Mat., 1989, 23, 163. Maistros, G.M. and Bucknall, C.B., Polym. Eng. Sci., 1994, 34, 1517. Mangion, M.B.M. and Johari, G.P., J. Polym. Sci.: Part B: Polym. Phys., 1991, 29, 1127. Senturia, S.D. and Sheppard, N.F., Adv. Polym. Sci., 1986, 80, 3. Antonsen, G.A., Dielectric cure monitoring of carbon fibre/epoxy composite structures. MSc Thesis, Cranfield University, 1995. Maistros, G.M., Antonsen, G.A., Pa~idge, I.K. and Cracknell, J.G., in Proc. 4th Int. Conf. on Automated Composites (IC1C'95), Nottingham (UK), September 1995. Day, D.R., Shepard, D.D. and Craven, K.J., In-process endpoint determination of epoxy resin cure. Technical paper RP043, Micromet Instruments Inc., Newton Centre, MA, USA.