IntJ FatiguelO No 3 (1988) pp 171-177 The Strength-Life Equal Rank Assumption and its application to the fatigue life prediction of composite materials P. M. Barnard, R. J. Butler and P. T. Curtis The authors have re-examined the Strength-Life Equal Rank Assumption (SLERA) that is used in the fatigue life prediction of composite materials, and suggest that this assumption may be valid for a wide range of fibre-reinforced epoxy laminates subjected to tensile, compressive or shear fatigue Ioadings. The evidence presented here suggests that there may be an exact correlation between the initial static strength and fatigue life expectancy. A corollary is that scatter in the fatigue data is a consequence of the variation in the static strengths of individual specimens. Using SLERA, an efficient and usable life predictive technique has been developed for fibre-reinforced composite materials. Key words: fatigue; fatigue life prediction; composite materials; Strength-Life Equal Rank Assumption (SLERA) Introduction S c a t t e r in f a t i g u e d a t a The possibility of a correlation between the static strength and fatigue life distributions for composite materials was first suggested by Hahn and Kim, 1 though the Strength-Life Equal Rank Assumption (SLERA) was named by Chou and Croman. 2 The SLERA states that if a sample of components could be tested for both static strength and fatigue life expectancy, each individual member would occupy the same rank in both the strength and life data sets. The SLERA is normally expressed in a form that lends itself to usage in the life prediction of composite materials, that is, for a given sample, the statically strongest members would have the longest fatigue life expectancy. The SLERA, though potentially very powerful, has not found wide support in the literature for two main reasons; first, it cannot be proved; secondly, it is argued that SLERA carmot be applied to notched specimens because their s~rength increases during fatigue, see Fig. 1, and therefore the life expectancy would also increase, which is contrary to engineering intuition. The evidence presented in the subsequent sections of this paper will provide conclusive evidence for the validity of the SLERA. The argument that notched specimens invalidate the SLERA is false, because SLERA makes no assumption about residual properties. The SLERA states that there is a correlation between the static strength and the fatigue life expectancy of a sample of nominally identical components. As the fatigue life progresses, the strength of each component changes. For each component at a given cyclic life there is still a correlation between residual strength and residual life expectancy, but the correlation need not be the same as the initial one. The scatter in fatigue data has two components, a fatigue component and a static one: 4 The fatigue component is induced by variations in the propagation rates associated with the fatigue failure mechanisms. The static component is due to the variation in static strengths of the sample members under test, and may be visualized by taking a hypothetical SIN curve, Fig. 2. Take a set of specimens with static strengths that range from 500 to 1000 MPa, with a mean of 750 MPa, that are fatigue tested at 600 MPa. Specimens that have a strength less than 600 MPa will fail ~ 600 ~" [] ~ ~ % ~_50o 300 -I ×(2)* - ~ a I I I I I ~ ~. 0 I 2 3 4 5 6 , I 7 Log cycles to foilure Fig. 1 Variation of residual =rength with number of fatigue c y c l ~ for carbon f i b r e / e p o ~ ( 0 , ~ ) T ~ / 9 1 4 laminate: + is plain nonwoven; D is centre-notched n o n - w o v e n ; x is plain w o v e n and 0 is centre-notched w o v e n ~ 0142-1123/88/030171-07 $3.00 © 1988 Butterworth & Co (Publishers) Ltd Int J Fatigue July 1988 17/ Coupon strength 5oo M I000" Ml=k:l ----_ v 60(3 % Specimen strength (MPo) 500 750 I000 ~ N ( cycles ) 20%, and from Fig. 3 t h e life expectancy range would be n3 to n4. It can b e seen that as the applied stress decreases the fatigue scatter bands decrease, ie the shape-function, which is a measure of the spread of a distribution, is dependent upon the stress level. The shape function is not only a function o f the stress level, but is also a function of the slope of the S i N curve. For a fixed sample at a freed stress level, the scatter bands are larger for a shallow SIN curve than for a steep one, which is particularly important for nonlinear SIN curves. For example at the 600 MPa stress level, see Fig. 3, as the SIN curve gets steeper, so the life expectancy range changes f r o m n a "-~ tt2 tO n 5 - - ~ n 6. Estimation of the static c o m p o n e n t of f a t i g u e scatter Fig. 2 HypotheticalS/Ncurve on loading, .that is they are effectively tested at 100% of their strength. Progressively stronger specimens are-effectively tested at lower percentages of their ultimate strength, thus specimens with a strength of 1000 MPa are tested at 60% of their strength, and consequendy have longer fatigue life expectancies than weaker specimens. The SLERA will only apply if the fatigue component of the fatigue scatter is negligible, or else is some function of the static component. The SLERA states that the stronger the specimen, the longer the life expectancy. As a result an exact correlation exists between strength and life expectancy. In addition the SLERA~ implies that the life expectancy is a function of the appli-~d load, therefore the lower the applied stress-the longer the fatigue life. This effectively defines the SIgN (stress-life) curve, therefore the SLER_A and the SIN curve are essentially the same. If the applied stress is plotted as a function of specimen strength, the SIN curve may be envisaged as the applied stress-life expectancy relationship. Then .at a f~xed applied stress~ each specimen in a sample will be tested at different percentages of its static strength, and its fatigue life expectancy can be estimated from the X/N curve. For example, let the range of strengths in th~ sample be 800-1000 MPa and the applied stress level be 600 MPa; the applied stress as a function of specimen strength thus ranges from 75 to 65%. From Fig. 3 this would result in a corresponding life expectancy range of n 1 to nz. If the applied stress level had been 200 MPa, the applied stress as a function of specimen strength would be 25 to The technique The SLERA states that there is a correlation between the static distribution f(6) and the life distribution fin), where. f ( ) is the probability density function of stress or life. The cumulative functions P(o) and P(n) are similarly correlated. Most authors (eg References 1 and 2) have assumed a direct correlation between the cumulative functions, ie P(cr) = P(n). In the present study such a correlation has not been assumed, and it is suggested that f(o) = g[f(n)] and P(o) = b[P(n)], whereg and b are some function of the bracketed terms. The present authors suggest that the functional relationship between the probability density functions is the gIN curve, where S is the applied stress and N is the mean cycles to failure. If the applied stress is expressed as a percentage of the individual specimen strength, and a generalized SIN curve is used, the correlation is given by IOOS _ a (log (n) + c)b + d where a, b, c and d are constants, S is the applied stress, o is the specimen strength and n is the specimen life. (Note ff e or d = 0 and b = 1 this reduces to a log-linear relationship). Since this relationship is for an individual specimen, any quantile of the static distribution may be used to obtain the corresponding life quantile, and the relationship may be rewritten as 100S _ a(log (np) + c) b + d % I00 where p denotes the percentile (/e p = 1,2,...99). It should be noted thatp/100 = P(xp), where x = ~ or n. If it is assumed that the fatigue distribution is a twoparameter We]bull distribution then P(n,) ~ (1) =1-expE - ',"'/(n'Y~'l 50 -~ m 25 0 n I /75 n6 nz n3 r/4 Cycles Fig. 3 Dependency of fatigue scatter upon the applied stress level and slope of the SIN curve 172 where I]~is the shape parameter and ~1~is the scale parameter. For the present, it is assumed that the SIN constants a, b, c and d are known, and that the static distribution is also known. Then for any stress level, by choosing any two values of ~0, the corresponding n~0values can be found by using Equation (1), and consequently the values of lie and ~le Let the subscripts y and z denote the first and second n~ values, then P(n~,) = 1 - exp [ - \ fir/ d Int J Eatigue July 1988 and P(n~) = 1 - exp[ - ooo .. (\ ~n r' Y) r l ] Therefore I-logO- P(.0) l , = '°gbo 1 . . . . . m))J ~aoo~-~.~ ~ and 1 ~l~ = ( _ log[1 - P(ny)])tmf = ( - log[1 - P(n3]) amf The values of ~ and ~le are for the static component of the fatigue scatter. Thus from a known SIN curve, the amount of scatter as a direct result of the variations in static strength can be calculated using these ~f and llf values. The probability of surviving n: cycles is the combined probability of having a strength greater than the applied load plus the probability of having a life expectancy of greater than n:. Assuming that the static distribution is also a twoparameter Weibull distribution, the probability of failure at or before n: is given by P(n,) =I - exp[ - (nP~ ~f - \nd , 102 103 , 104 105 106 107 108 N (cycles) Fig. 5 Experimental data and the 1 and 99 percentile curves for the static component of fatigue scatter for a 58% fibre volume fraction [(0,90 + 45)a], carbon/epoxy laminate under compressive loading 25 ('Y'l \~] J (3) where ~, and ~l, are the shape and scale parameters of the static distribution and the values for fl~ and ~]e have been calculated as above. The life percentiles are therefore given by (.s~,]~l~ ~, ~ ~{-logO - ~ , ~ - ~ ] ~ i'1 ¢ X × X XX X ~ z( _~ 15 IO Experimental verification The 1 and 99 percentile curves of the static component of fatigue scatter have been determined for a number of unidirectional, crossply and quasi-isotropic 'E'-glass and XAS carbon-fibre-reinforced epoxy laminates subjected to constant amplitude tension-tension, compression-compression and shear-shear fadgue loading. A representative sample of these curves are given in Figs 4 to 6. A full description of material and experimental details and test results for tension and compression loadings may be found in Reference 4, and for the shear in Reference 5. For all the laminates tested, there is a good correlation between the experimental fatigue dam and the I and 99 percentile scatter bands estimated from the static data and the XIN curve as described above. 10z ~o~ =04 io5 ~o6 =ov ~o8 N (cycles) Fig. 6 Experimental data and the 1 and 99 percentile curves for the static component of fatigue scatter for a 58% fibre volume fraction (0), carbon/epoxy laminate under shear loading It would appear that for fibre-reinforced epoxy laminates, the scatter in the fatigue data may be a direct consequence of variations in the static strength, ie there is no fatigue component. If there were a fatigue component, the experimental fatigue scatter bands should be greater than those for the static component alone. These findings strongly support the validity of the SLERA for the materials under investigation. I000 900 800 ~,,,~ Fatigue life prediction using the SLERA x~ XX ~ Case1 : $ / N c u r v e is known (lea, b, cand d are known) ~ 700 ~ 600 ~ 50C 400 3o0 I 103 104 I I 105 106 107 N (cycles) Fig. 4 Experimental data and the 1 and 99 percentile curves for the static component of fatigue scatter for a 57% fibre volume fraction (0ha 'E'-glass/epoxy laminate under tensile loading Int J F a t i g u e J u l y 1988 Assume that the SIN fatigue curve has been developed previously together with the static strength distribution, for a batch of a certain material. If static tests on a new batch of the same material reveal a different strength to that measured on the initial batch, the fatigue life expectancy of the new batch may be determined as follows. It will be assumed that the static distribution of the new batch has been characterized by the static tests, ie ]3, and ~1, are known, therefore % and P(%) can be calculated. From Equation 1 if % = vh then IOOS qs = a (log 01r) + c) + d 173 Therefore /fF100s n, : @1- c) (4) Case 2: From Equation (2) • [log0 ~r = tOgLlog(1 predicted 1 and 99 percentile curves, are shown in Fig. 7. It can be seen that the prediction shows a good correlation with the experimental data. $/N curve u n k n o w n This is often the case for a new material or component for which no previous data are available. The question often asked is what safe fatigue Life expectancy can we expect for this material/component before any service experience is available? To answer this question the and the distributional form of the life expectancy need to be known. The main object of Life prediction is to extrapolate data generated in laboratory environments to service environments. Without any further information, the safest way to do this is to assume either a linear or log-linear extrapolation. Therefore, if the curve is assumed to be log-Linear, k b = 1 and d = 0, conservative estimates of the life expectancy will be made. If we assume that the SLERA applies, and that the fatigue distribution is a two-parameter Weibull distribution, it follows from Equations (3) and (4), as shown in the previous example, that l P(n,))J l°g (n~) SIN curve but l [log0 - P(nz))]= A °gL og(1 where _/t is a constant. Therefore A ~f ~ SIN A m log(n - log(n,, From Equation (1) Hz xpF '['°°s d)}lID- c] S = mlog ('qr) + c and S = Pf Fatigue testing at two stress levels thus allows the constants and A to be evaluated, and consequently vlf and ~3f values at all stress levels allowing the generation of (probability-stress-life) curves. By using Equation (1) the static distribution parameters vh and 13s can also be estimated. m, ¢ Therefore I r= .4 - A (IOOS {)'/o_ (IOOS ~)'/~ (5) \ It can be seen that the only unknowns in Equations (4) and (5) are rlf and ~f. Therefore, the rlf and ~f values as a function of S can be determined. Example of process Fatigue data on a [(0,90, 4- 45)2], laminate of Ciba Giegy's XAS/914 have been generated by Chung. 6 Subsequent work by the present authors indicated that the static strength of a supposedly identical laminate was 25% greater than that of Chung's. This increase was attributed to increases in fibre strengths that occurred during the time that elapsed between the two sets of work. 7 Using Chung's data and the static distribution o f the new material, the fatigue distribution of the new material could be estimated. The constants for the relationship from Chung's work were a --4.75694541, b = 1, c = -27.0209935 and d -- 0. This is a log-linear curve given by P-S-N Example of process Nineteen specimens manufactured from a material, which was later identified as the ICI APC2 material (carbon-fibrereinforced poly ether ether ketone), were supplied by one of the authors (PTC). The object was to identify the full tensile static and fatigue distributions of this material, including the 1 and 99 percentiles, from the nineteen specimens. One of the specimens was statically tested to give art indication of the mean tensile strength, so that appropriate fatigue stress levels could be chosen. Two samples of 8 specimens were fatigue tested at 60% and 40% of the strength of the statically tested specimen. The final two specimens 70( o ChungIs] x Presentwork ~ x SIN × SIN ~ 100S % - 128.537391 - 4.75694541 log(np) For the new material ~ls = 700.89 and ~, = 34.33, and the corresponding 1 and 99 percentiles are ~l = 613 and o99 = 732.78. From Equation (4) x x x ~ g_ ~ 50C ~ o . . . . . ~ × × ×7 ~ qf = exp( 27.0209935 - 0.02999330 from Equation (5) 1093.07291 ~f- S The ~lf and ~f values as a function of S can therefore be calculated, as can the corresponding n~ values from Equation (3). The Chung and new data, including the 174 . . . . . $0( I0 ~ I 104 I 105 I 106 %~ ~ 107 ~ I0 8 N (cycles) Fig. 7 Chung's s and the pre~nt authors" experimental data for a ~ % fibre volume fraction [0,~, • 45)=], carbon/epo~ laminate under tensile loading and the 1 and 99 percentile c u ~ for the static component of fatigue sca~er for the authors" material predict~ from Chung's data Int J F a t i g u e J u l y 1 9 8 8 were used to verify the prediction. A fuU description o f material and experimental details and test results may be found in Reference 4. The tensile strength of the single specimen was 2054 MPa. The two fatigue stress levels were therefore 1232.4 and 821.6 MPa. The resulting fatigue data were rl~0) = 11 390, ~f(60) = 0.25, ~lf(~0) = 6061 510 and ~f(~0) = 0.67. To evaluate the constant A , the ~f value at the lowe~ stress was chosen. This was because it is advisable to use data that have been generated at stress levels below the influence of the static distribution. 80 70 ~ 6O .1::: so N .~ i1~ 4o .~ ~ ~0 20 I0 A = 821.6 x 0.67 = 550.472 i03 Therefore 550.472 S- (6) I I I I I I04 i05 ]06 i07 i08 ,I I09 gf Fig. 9 Predicted Weibull scale parameter vs applied stress for a 60% fibre volume fraction (0)1= carbon/PEEK laminate under tensile loading Also S = m l o g 01f) + c ~ 75 1232.4 = m l o g (11 390) + c and - 25 'l 821.6 = m l o g (6061 510) + c x x Therefore 4o m = - 6 5 . 4 4 5 5 a n d c = 1843.6931 ~ 5o Thus (7) ~ ~o The values o f ~e and ]If as a function o f applied stress are shown in Figs 8 and 9. The resulting P - S - N curves and experimental data are shown in Fig. 10. In this case, the P - S - N curves do not include the conditional probability that the strength is greater than the applied stress. I0 S = - 6 5 . 4 4 5 5 l o g 0if) + 1843.6931 Estimation of static distribution From Figs 10 and 11 it can be seen that the Weibull mean SIN curve is nonlinear at such low values of the slope function, and is not representative of the fatigue population. It is suggested that the median S/N curve should be used as the strength-life correlation, and will be used in this section. The procedure is outlined as follows: I I0 3 ~0 2 I ~0 4 I {O 5 I tO 6 I ~0 7 10 8 N (cycles) Fig, 10 Experimental data and predicted P - S - N curves for a 60% fibre volume fraction (0)1= carbon/PEEK laminate under tensile loading IO0 90 70 Estimate the median SIN curve from the life parameters estimated from Equations (6) and (7). Since the median • o~ 60 -~ E 50 90 ~ 8O ~0 ~ 70 2O "~ 60 I0 f 0 "~ ,50 Asymptotic approach of mean to = 43% I I I I I I I I I I I I I 2 3 4 5 6 7 8 9 I0 ~1 I?- P Fig. 11 Mean and median of the Weibull distribution as a function of percentage of the population vs shape parameter ~ 4o .g ~ Median Mean $0 20 1C I 0.2 I I 0.6 I I 1.0 I I 1.4 ~ 1.8 2.2 I~f Fig. 8 Predicted Weibull shape parameter vs applied stress for a 6 ~ fibre volume fraction (0)1= carbon/PEEK laminate under tensile loading Int J Fatigue July 1988 SIN curve is logslinear, only two stress levels are required. At 1500 MPa, 13f = 0.367 and rlf = 191. Therefore nso = 70. A t 500 MPa, I~f = 1.101 and vie = 825 485 225. Therefore nso = 591 748 000. 1500 = mlog (70) + c 500 = mlog (591 748 000) + c 175 2.25 900' - 002.00 - . . . . . . -~,o~,~ • • ~ ~t~ ~ ~ , , ~ 800 90- i I .75 - ~ o~ 1.S0- "~70 , tO • ~ • 22pts II °~ e e ~~1 ~ • 700 80- o~ = ~. <' 600 I~ ._.E ~:} & • ~ . m 1.25 - ~= I t~. I 2 p t s ~ QI 2 ~ 5oo ~ 5o 71~'l"T • 224 2 2 p t ~ , , ~ " 40O 1.00 - ° t - - ' ° %~. ~s 40 0.75 - I ~00 30.- 102 I 103 I 104 I 105 /V (cycles) I 106 I io 7 io 8 Fig. 12 Experimental data for a 45% fibre volume fraction (0)~= "E'-glass/epoxy laminate under tensile loading . - Thus m = -62.6955 and c = 1766.362 The median SIN curve is therefore given by S = 1766.362 - 62.6955 log (n) * • • Calculate the 1 and 99 life percentiles at any stress level. Let S = 1027 MPa, then ~ - 0.536 and ~lf = 265 756. Therefore n~ = 49 and n99 = 4 538 900. F r o m the median curve calculate the stress level, applied to an average specimen, that would result in these lives: S~1 = 1766.362 - 62.6955 log (49) = 1522.36 MPa S~9 = 1766.362 - 62.6955 log (4 538 900) = 805.35 MPa Calculate the percentage ratio o f S, to the single static strength: %S~ = 1522.36 x 100/2054 = 74.17% %S~ 9 = 805.35 x 100/2054 = 39.21% • Using the cumulative distribution, the strength distributional parameters can be estimated, and are listed below with some experimental data in brackets. The experimental data were provided at a subsequent date by one o f the authors (PTC). p, = 9.63 (14.17) vl~ = 2234.01 (2049.07) p. = 2124.32 (1975.48) ~ = 259.70 (169.49) c.v. = 12.23% (8.58) estimated statistics are estithe experimental statistics size o f 20. It can be seen between the estimated and Some limitations of the SLERA The main limitation of the S L E R A is that it cannot predict a change in fatigue failure mode as a function o f the stress level. Dhar,an9 suggested that for 'E'-glass/epoxy under tensile loadiag the dominant failure mechanism is a function o f applied strain. Bamard et all° found that such a change in failure mechanism may lead to a discontinuity in the SIN curve, see Fig. 12. The S L E R A appears to be valid above and below this discontinuity though the correlation between strerigth and life is different, s Unless a change in failure mode increases the slope o f the SIN curve the S L E R A will provide conservative estimates, see Fig. 13. A specimen tested at a fatigue stress level of 74.17% of its strength would thus have a life expectancy of 49 cycles, and at 39.21% an expectancy o f 4 538 900. The corresponding strength percentiles may be calculated from the applied stress level: ~t = 1027 x 100/74.17 = 1384.66 M P a ~9~ = 1027 x 100/39.21 = 2619.23 MPa 176 It should be noted that the mates of the population, while were calculated using a sample that there is a good correlation experimental values. IOOC 90C 80~ ~ ~ 2 ~ 70C n° 2pts 22p~s I 1 ~ °'" *~ ~, ~ ~ "'~... ~ ~OC ~ ~ ~ ~ ~ "-. ~, ~o~ ××x "~pts~ V~-~ ~ ~~ ~ ~ " ~ #s x ~x ~ x.~...x=~ ~ ~...~..~. .~= 2p~ ... ~ 40¢ ~... x ~.2 ~ Ors , , x~,~x 30C ?-pts 20C t ]02 10`3 I I I IO4 105 N (cycles) 106 I 107 i08 Fig. 13 1 and 99 percentile curves for the static component of fatigue scatter for the data in Fig. 12 predicted from the data in Fig. 4 Int J F a t i g u e J u l y 1 9 8 8 The SLERA cannot predict the residual strength, though some researchers have attempted to do so, eg Chou and Croman. 2 Chou and Croman assumed that the locus of the residual strength quantile was equal to the combined relevant static and fatigue quantiles, ie n,/ =(57'\n4 \nO \7 j where R is the residual strength and b is an experimentally derived parameter that indicates the rate of degradation with respect to fatigue cracks. Chou and Croman found that for values of 13 greater than 1 the degradation is weak, and for values less than 1 the degradation is strong. The present authors have found that if a value of b = 1 is used in conjunction with the rl~ and ~f values, as calculated above, then for all the laminates investigated the residual strength predictions have a good correlation with experimental data. Full details of the shear residual strength predictions can be found in Reference 5, and the tensile and compressive residual strength predictions can be found in Reference 4. If the laminate is. notched the residual strengths predicted using the model of Chou and Croman will be conservative. Therefore the authors recommend that if residual strength predictions are required a value of b = 1 should be used. Therefore " = - 1¢0 J S-N 2) The SLERA appears to be valid for a wide range of fibre-reinforced plastic laminates, provided there is no change in failure mode with fatigue life. The SLERA is a definition of the SIN curve, therefore the correlation between the strength and life is the SIN Curve. 3) 4) 5) Scatter in fatigue data is primarily due to static strength variations, though additional scatter may occur in the region of a failure mode change. The scatter is a function of the applied stress, thus the lower the stress the lower the scatter; and is an inverse function of the slope of the S/N curve, thus the greater the slope the lower the scatter. A technique for predicting the static, fatigue and residual strength distributions from a small sample size has been suggested. This small sample size may allow testing of components without the prior requirement of laboratory specimen and scale model testing. This technique: can be easily computerized so that from input data comprising one static result and perhaps two sets of eight: fatigue results, the output data would be the static digtribution parameters, P-S-N curves and P(R)- Int J Fatigue July 1 9 8 8 of residual strength-stress-life) Acknowledgements The authors would like to thank the technical staff at the Cranfield Institute of Technology, and in particular Messrs M. Crook, S. Gowans, M. Jones and M. Williams, for the excellent technical services provided. This work has been carried out with the support of the Procurement Executive of the UK Ministry of Defence. References 1. 2. 3. 4. Hahn, H. "r. and Kim, R. Y. 'Proof testing of composite materials' J Comp Mater9 (1975) pp 297-311 Chou, P. C. and Croman, R. 'Residual strength in fatigue based on the Strength Life Equal Rank Assumption" ibid 12 (1978) pp 177-194 Curtis, P. T. and Moore, B. B. "A comparison of the fatigue performance of woven and non-woven CFRP laminate' RAETR 85059 (1985) Barnard, P. M. and Young, J. B. 'Cumulative fatigue and life prediction of fibre composites: Final report" Report on RAE Famborough Contract No. 11494/2028-0134 XR/MAT (1983) 5. Barnard, P. M. and Young, J. B. 'Shear fatigue life prediction of fibre composites: Final report' Report on RAE Farnborough Contract No. 11494/2028-0138 XR/MA T ( 1987 ) 6. Chung, K. M. "Cumulative fatigue and life prediction of carbon fibre reinforced laminates" MSc thesis (Cranfield Institute of Technology, England, 1982) 7. Anon. Private communication (Ciba-Geigy Plastics and Additives Company, Duxford, England) 8. Barnard, P. M , "Cumulative fatigue and life prediction of unidirectiona!ly reinforced 'E'-Glass/Epoxy laminates' PhD thesis (Cranfield Institute of Technology, England, 1986) 9. Dharan, C. K. 'Fatigue failure mechanisms in a unidirectionally reinforced composite material' Fatigue of Composite Materials, ASTM STP 569 (American Society for Testing and Materials, 1975) 10. Barnard, P. M., Butler, R. J. and Curtis, P. T. 'Fatigue scatter of unidirectional 'E'-Glass/Epoxy a fact or fiction?" in Composite Structure 3, I. H. Marshall (ed) (Elsevier Applied Science Publishers, 1985) Conclusions 1) (probability curves. Authors i Dr Barnard was formerly with the Cranfield Institute of Technology and is now with the New Products Division of Ruston Gas Turbines Ltd, Lincoln, UK. Mr Butler is with Analysis & Design Consultants Ltd, Newport Pagnell, UK and Dr Curtis is with the Materials & Structures Department of the Royal Aircraft Estabfishment, Famborough, UK. 177