1994 Elsevier Science Limited Printed in Northern Ireland. All rights reserved 0308-0161/94/$07.00 Fatigue life prediction in composites R. J. Hus t o n Division o f Aeronautical Systems Technology, CSIR, Pretoria, Republic o f South Africa Because of the relatively large number of possible failure mechanisms in fibre reinforced composite materials, the prediction of fatigue life in a component is not a simple process. Several mathematical and statistical models have been proposed, but the experimental evidence to support them is limited so far. In this paper, an attempt has been made to fit experimental results to two of the more promising models. Results obtained from repeated tension fatigue tests on unidirectional carbon fibre reinforced epoxy were used to test residual strength and residual stiffness models. Further fatigue tests were carried out under spectrum loading so that the results could be correlated with the cumulative damage predicted by the residual strength model. I NTRODUCTI ON Dur i ng t he last two decades or so, t he use of composi t e mat eri al s in t he pl ace of met al s in many industries has become wi despr ead, not least in t he aircraft i ndust ry. In gener al , a composi t e mat eri al can be defi ned as a mat eri al t hat consists of two or mor e distinct component s. The composi t e mat eri al s used t o manuf act ur e aircraft st ruct ures have usually compr i sed con- tinuous fibres embedded in a pol ymeri c matrix. The fibres provi de t he st rengt h and t he stiffness and t he mat ri x hol ds t he fibres t oget her. Consi derabl e wei ght savings can be achi eved by using composi t e mat eri al s in aircraft because of t hei r hi gher specific st rengt hs and specific stiffnesses compar ed wi t h al umi ni um alloys. Since t he days of t he Comet , fatigue has been a subject of consi derabl e i nt erest to t he aircraft industry. The i ncreased use of composi t e materials has emphasi sed t he fact t hat t hei r fatigue behavi our is mor e compl ex t han t he fatigue behavi our of metals. I n a met al , fatigue damage event ual l y appears as a cl earl y defi ned crack, t he initiation and pr opagat i on of whi ch can be predi ct ed by fract ure mechani cs analysis. In a composi t e mat eri al , fatigue damage can t ake t he form of any or all of t he following: del ami nat i on, mat ri x cracki ng, fibre fai l ure, matrix crazing, fi bre/ mat ri x debondi ng and voi d growth. It is dependent on variables associ at ed wi t h t he testing condi t i ons and t he const ruct i on and composi t i on of t he mat eri al . The S- N curve appears still to be t he most popul ar met hod of charact eri si ng t he fatigue behavi our of composi t e materials. Several em- pirical equat i ons exist for descri bi ng S- N curves. Most are based on t he classical power law t hat gives a st rai ght line in a log-log pl ot of t he fatigue dat a, but some possess refi nement s to account for mean stress or t he ratio of maxi mum stress to mi ni mum stress. One of t he si mpl er equat i ons used to descri be t he fatigue behavi our of several composi t e mat eri al s is as follows: ~ Oa = O, -- b log N (1) 131 wher e oa is t he maxi mum appl i ed stress, ou is t he static st rengt h, b is a const ant and N is t he number of cycles to failure. Ot her t heori es have been f or mul at ed for t he charact eri sat i on of t he fatigue behavi our of composi t e materials. Ther e are essentially t hr ee types: t heori es based on t he degr adat i on of residual st rengt h, t heori es based on changes in modul us and t heori es based on act ual damage mechani sms. Most of t he life predi ct i on met hods for pol ymeri c composi t e mat eri al s current l y in favour are based on resi dual st rengt h degr adat i on (e.g. Refs 2- 5) . These have been devel oped on t he basis of 132 R. J. Hust on t hr ee assumpt i ons: (1) t he statistical variability of t he static st r engt h of t he mat er i al can be descr i bed by a t wo- par amet er Wei bul l di st r i but i on; (2) t he resi dual st r engt h aft er N cycles of fat i gue l oadi ng can be r el at ed to t he static st r engt h by a det er mi ni st i c equat i on; and (3) when t he resi dual st r engt h decr eases to t he maxi mum appl i ed stress, fat i gue fai l ure Oc c ur s . Reduct i on in stiffness caused by fat i gue l oadi ng has been r epor t ed by vari ous obser ver s, 6-11 and it has been recent l y r evi ewed by Rei f sni der . 12 Theor i es for fat i gue based on t he r educt i on in stiffness have one significant advant age over t he resi dual st r engt h t heor i es, namel y t hat t he r emai ni ng life can be assessed by non- dest r uct i ve means. A second advant age is t hat less t est i ng needs to be conduct ed. A typical r educt i on in stiffness model ~1 rel at es t he degr adat i on of modul us t o t he fract i on of life e xpe nde d at a gi ven stress ampl i t ude, assumi ng t hat t he resi dual stiffness decreases monot oni cal l y as t he numbe r of l oad cycles i ncreases. Fat i gue t heor i es based on damage mechani s ms model t he i nt ri nsi c defect s in t he mat r i x as small cracks parallel to t he fibres, and t he pr opagat i on of t hese cracks are pr edi ct ed by l i near f r act ur e mechani cs analysis. Unf or t unat el y, pr edi ct i on of fat i gue failure is difficult because sophi st i cat ed mat hemat i cal t echni ques are ne e de d t o cal cul at e t he forces r equi r ed t o pr opagat e t he cracks and t he damage accumul at i on process is si mul at ed using Mont e Carl o met hods. t 3 In this paper an at t empt is made to cor r el at e t he fat i gue t heor i es of Sendeckyj 5 and Whi t wor t h ~ wi t h t he exper i ment al resul t s ob- t ai ned f r om uni di r ect i onal car bon fibre rein- forced epoxy. EXPERI MENTAL PROCEDURE Composite material tested and its processing conditions The composi t e mat er i al syst em used to t est t he resi dual st r engt h and resi dual stiffness model s was a uni di r ect i onal car bon fibre r ei nf or ced epoxy pr epr eg syst em, Vi cot ex 913C-TS, manu- f act ur ed by Br ochi er SA, an affiliate of Ci ba-Gei gy. The resi n mat r i x was a modi f i ed epoxy wi t h good resi st ance t o t e mpe r a t ur e and humi di t y, and t he car bon fibres wer e of t he Tor ayca T300B t ype. The st andar d aut ocl ave cur i ng schedul e was appl i ed t o manuf act ur e t he sheet s f r om whi ch test speci mens coul d be pr epar ed. Thi s schedul e consi st ed of heat i ng t he uncur ed l ami nat e under vacuum to 90C at a rat e of about 3C per mi nut e and t hen hol di ng t he t e mpe r a t ur e at 90C for 30 mi n. A pr essur e of 7 bar was gradual l y appl i ed and t he vacuum vent ed when t he pr essur e r eached 1.4 bar. The t e mpe r a t ur e was i ncr eased to 120C, again at about 3C per mi nut e, and hel d t her e for 60 mi n. The l ami nat e was t hen cool ed t o 90C when t he pr essur e was r el eased, and it was r emoved f r om t he aut ocl ave when t he t emper at ur e had dr oppe d t o bel ow 60C. Preparation of specimens The same t ype of s peci men was used in t he tensile and fat i gue tests. It was a flat, paral l el -si ded speci men appr oxi mat el y 2 2 0 mm l ong, 20 mm wi de and 1 mm t hi ck. A band saw was used to cut t he speci mens out of t he cur ed l ami nat es, and t hey wer e gr ound paral l el wi t h t he aid of a surface gr i nder . To faci l i t at e l oad t ransfer in t he grips of t he t est machi ne, al umi ni um t abs about 2 mm t hi ck wer e bonde d on to each speci men. Test programme In t he first stage of t he t est pr ogr a mme , t ensi l e tests wer e conduct ed on 30 speci mens to pl ace an upper limit on t he stress levels t o be appl i ed in t he subsequent fat i gue tests and t o assess t he variability in t he pr oper t i es of t he mat er i al . The tensile tests wer e conduct ed in accor dance wi t h t he r ecommendat i ons of ASTM D3039, and in each t est t he t ensi l e st r engt h and t he t ensi l e modul us wer e meas ur ed. Fol l owi ng t he t ensi l e t est s, fat i gue t est s wer e carri ed out to der i ve S- N curves for t he mat er i al . Thes e tests wer e car r i ed out unde r l oad cont r ol in r epeat ed t ensi on wi t h t he stress rat i o (R) equal to 0.1. The f r equency was 10 Hz. Fi ve repl i cat e speci mens wer e t est ed at each of seven stress levels to gener at e t he S- N curves. Fi ve t ensi l e st r engt h resul t s, sel ect ed r andoml y f r om t he tensile t est dat a, wer e also i ncl uded. The fai l ure cri t eri on was cons i der ed t o be speci men fract ure. Fat i gue l i f e pr edi ct i on in c ompos i t e s 133 Fatigue tests were continued until approximately 2 million cycles, at which point the specimens were considered to be run-outs. Tensile tests were conducted on each run-out to measure residual strength. At two stress levels, 1700 and 1800MPa, further fatigue tests were carried out to assess the variability of fatigue life. Twenty-five specimens were tested at both stress levels to identify a probable distribution. In the next stage of the programme, a series of load spectrum tests was performed to compare cumulative damage predictions with experimental results. A load spectrum based on FALSTAFF la was devised and 10 specimens tested. The tests were continued either to failure or were stopped after 5 million cycles, and then residual tensile strength was measured. As before, the tests were carried out under load control at a frequency of 10 Hz and with a stress ratio (R) of 0.1. Finally, to assess the residual modulus model, a few more fatigue tests were performed. At various stages in each of these tests, the static tensile modulus was measured. To do this, the fatigue tests were interrupted periodically so that an extensometer could be attached to the specimen and the measurement s made. T e s t e q u i p m e n t All the fatigue tests were conducted using a Schenck Hydropuls machine of 250 kN capacity. Tensile modulus measurement s were made using an HBM DD1 strain transducer connected to an HBM DWS73 amplifier and digital monitor. T E N S I L E T E S T R E S U L T S The results of the preliminary tensile tests are summarised in Table 1. The tensile strength data The most popular met hod of characterising the fatigue behaviour of composite materials is still the generation of S- N curves. To attach statistical value to these curves, it has usually been necessary to test large numbers of replicate specimens at different stress levels to obtain the distribution of fatigue lives. Examples of this are shown in Figs 3 and 4, where Weibull distributions have been fitted to fatigue lives obtained at maximum stresses of 1700 and 1800 MPa. The probabilistic life prediction model of Sendeckyj 5 does not require a large data base. Static, fatigue and residual strengths ar e all converted to equivalent static strengths and pooled to produce a data base large enough for statistical analysis. A t wo-paramet er Weibull distribution is fitted to the residual strength data. By assuming that the shape (tr) and scale (fl) parameters of the fatigue life and static strength distributions are functionally related to each other and to the constants C and S in the classical fatigue power law equation o u / tra = C N s (2) Tensile Tensile strength, S~1 modulus, EI~ (MPa) (GPa) Mean 2 180 135 Standard deviation 197 5.89 Coefficient of variation 0.090 0-004 No. of specimens 30 30 Maximum value 2 526 148 Minimum value 1 733 122 Distribution Weibull - - B-Basis value 1 800 -- A-Basis value 1 700 - - were analysed in the manner recommended in the Military Handbook 17B, and a two- parameter Weibull distribution was found to be the most appropriate, in preference to the normal, log-normal, and non-parametric dis- tributions. As can be seen from the Weibull plot in Fig. 1, the linearity of the results is questionable, as borne out by the correlation coefficient of -0. 92. The distribution of the tensile moduli is shown in Fig. 2. The data fit the Normal distribution well. GENERATI ON OF FATI GUE CURVES Tabl e 1. Tensi l e strength and t ensi l e modul us results f or uni di recti onal Vi cot ex 913C-TS carbon fibre rei nforced e poxy O > J~ o 0 L - o :> E o Z 0.8 0.6 0.4 0.2 0.0 -0.2 -0.4 -0.6 -0.8 -1.0 -1.2 -1.4 -1.6 -1.8 -2.0 o. " 1 3 , 7 8 13 = 2 2 6 7 [] I I I I I I I I I I 3.20 3.24 3.28 3.32 3.36 Log(Tensile strength, MPa) Fig. 1, Weibull distribution of tensile strengths. [] I 3.40 the following expression for fatigue life can be obt ai ned: ere = [ a ; ' s + C ( N - 1 ) o J ' s ] s (3) wher e ae is t he equi val ent static st rengt h. Fitting t he fatigue model to t he experi ment al dat a is an iterative process. Fr om t he log o versus log N pl ot shown in Fig. 5, initial values of C and S are est i mat ed. The fatigue dat a are convert ed to equi val ent static dat a by using t hese values in eqn (3). A t wo- par amet er WeibuU distribution is t hen fitted to t he cr~ dat a, preferabl y by us i ng t he maxi mum l i kel i hood met hod, and t he values of t he shape and scale paramet ers are st ored. New values for C and S are chosen and t he pr ocedur e is r epeat ed until the maxi mum value of t he shape par amet er is obt ai ned. This val ue, and t he correspondi ng scale, C and S values, provi de t he best fit of t he fatigue data. For t he fatigue tests r epor t ed her e, t he values of C and S wer e f ound to be 0-000 75 and 0.057 6, respectively. These gave rise to a shape par amet er of 13.24 and a scale par amet er of 2235 for t he distribution of t he equi val ent static strengths. This di st ri but i on is shown in Fig. 6. From this distribution and eqn (3), S- N curves 3.0 2.0 1.0 0.0 -1.0 [ ] -2.0 -3.0 134 R . J . H u s t o n I 120 I I I I I t I I I I I 124 128 132 136 140 144 T e n s i l e mo d u l u s , E l 1 + ( GP a ) F i g . 2. Normal distribution of tensile moduli. I J [ 1 4 8 F atigue life prediction in composites 135 0.6 o ,.Q 0 0.4 0.2 0. 0 -0. 2 -0. 4 -0. 6 - 0. 8 - 1. 0 - 1. 2 -1. 4 -1. 6 - 1. 8 - 2. 0 o o D ' ~ ct = 0,597 13 = 83200 D O | | I I s | I I I 2. 0 3. 0 4. 0 5. 0 6. 0 L o g ( C y c l e s t o f a i l ur e ) Fig. 3. We i bui l di st ri but i on o f f at i gue l i ves at ma x i mu m s t res s o f 1700 MPa. can be const ruct ed for vari ous probabi l i t i es o f survival. Exampl e s o f t hes e are s hown in Fi g. 7. The Sendeckyj mode l is a c onve ni e nt me t ho d o f descri bi ng f at i gue behavi our but is de pe nde nt on the validity o f t he t hree assumpt i ons upon whi ch it is based: 5 (1) t he S - N behavi our can be descri bed by a det ermi ni st i c equat i on; (2) t he static strengths are uni quel y rel at ed to the f at i gue l i ves and resi dual strengths after a speci fi c numbe r o f cycl es; and (3) t he static strength data can be descri bed by a t wo- paramet er Wei bul l di stri buti on. For t he f at i gue tests pres ent ed here, t he first assumpt i on is probabl y val i d, si nce an empi ri cal equat i on can be fi tted to t he data. The s econd assumpt i on coul d be i nval i dat ed i f di f f erent failure mode s were pos s i bl e, but this is unl i kel y in this i nvest i gat i on as t he stress di stri buti on resul ti ng f rom t ensi l e l oads appl i ed in t he l ongi t udi nal di recti on to a uni di rect i onal l y rei nf orced s peci men is not c ompl e x. The l i neari ty D t'x3 > " t .Q 0. 6 0. 4 0. 2 0. 0 - 0. 2 -0. 4 - 0. 6 - 0 . 8 -1. 0 -1. 2 -1. 4 - 1 . 6 - 1 . 8 - 2. 0 a = 0 , 4 8 6 ~ = 6 6 4 0 0 0 ~ 0 0 ~%o 1.0 o I I I t I I I 2. 0 3. 0 4. 0 5. 0 L o g ( C y c l e s t o f ai l ur e ) Pig. 4. We i bul l di st ri but i on o f f at i gue l i ves at ma x i mu m st ress o f 1800 MPa. 136 R. J . Huston n P 1 E "-1 E ~ X (11 O} O _.1 3 , 1 3.4 3 . 3 3.2 3.1 3.0 2.9 CI [ ] 1 I I I I I 2 4 6 L o g ( C y c l e s t o f a i l u r e ) F i g . 5 . Fatigue curve for Vi cot ex 913C-TS. of the data in Fi g. 6 provi des an i ndi cat i on o f t he validity o f t he third assumpt i on. The correl at i on coeffi ci ent is - 0 . 9 6 . Thi s suggests that t he Wei bul l di stri buti on fits t he equi val ent static strength data qui t e wel l . F AT I GUE LI FE P RE DI CT I ON U N D E R S P ECTRUM L OA D I N G If one rearranges eqn (3) to or = [o~ `s- C ( N - 1)o~,' s] s (4) and i f one make s t he assumpt i on that thi s equat i on can generat e a val ue for resi dual strength after every cycl e, Sendeckyj ' s mode l can be appl i ed to spect rum l oadi ng. The f at i gue l oad spect rum us ed in this i nvesti gati on was based on F ALS TAF F , but it was si mpl i fi ed and a factor was us ed on t he l oads to ensure that s ome s peci mens f ai l ed after not t oo many cycl es. The spect rum is gi ven in Tabl e 2. An exampl e of how t he resi dual strength model can be appl i ed to predi ct t he f at i gue l i fe of a s peci men under spect rum l oadi ng is s hown in I-- > JO .$ 0.8 0.6 0.4 0.2 0.0 -0.2 - 0 . 4 -0.6 - 0 . 8 -1.0 -1.2 -1.4 -1.6 - 1 . 8 -2.0 [ 3 a = 1 3 , 2 4 13 = 2235 f - I I I I ! I I I I I 3.20 3.24 3.28 3.32 3.36 L o g ( R e s i d u a l s t r e n g t h , M P a ) Fig. 6. Wei bul l distribution of equi val ent static strengths. [3 I I 3.40 Fatigue life prediction in composites 137 2.8 2.6 2.4 2.0 1. 8 n r~ n P r o b a b i l i t y o f survival 0 ~ 1.6 ~ 1.4 i 1.2 0.1 0.8 L- - - 0 2 4 6 L o g ( C y c l e s t o f a i l u r e ) Fig. 7. S- N curves for 0.9, 0.5 and 0.1 probabilities of survival. Tabl e 2. Fati gue spect rum Maximum stress (MPa) Number of cycles 1 480 8 100 1 200 50 000 990 225 000 1 310 23 000 1 800 500 1 630 1 900 1 100 150 000 850 1 000 000 Tabl e 3. For a probabi l i t y of survival of 0-9, t he equi val ent tensile st rengt h is 1945MPa. The residual st rengt h decreases progressi vel y t hr ough each stage of each block. The val ue aft er each stage is given in Tabl e 3. When t he resi dual st rengt h has decr eased to a val ue bel ow t he applied stress, failure occurs. This is pr edi ct ed t o happen after 4 576 600 cycles, when t he maxi- mum applied stress is 1800MPa in t he fourt h block. Some experi ment al results are pr esent ed in Tabl e 4 for compari son with t he resi dual st rengt h model predi ct i on. As can be seen, t wo speci mens had fatigue lives shor t er t han t hat predi ct ed, one consi derabl y so, whi l e ei ght speci mens survived 5 million cycles. Speci men 4 failed duri ng t he 1800 MPa stage of t he t hi rd bl ock, and Speci men 7 failed duri ng t he 1630 MPa stage of t he first block. The resi dual tensile st rengt hs wer e measur ed for t he ot her speci mens and qui t e a wi de vari at i on was observed. Nat ural l y, all wer e hi gher t han t he highest maxi mum appl i ed stress of t he fatigue spect rum. The predi ct ed fatigue life was for a probabi l i t y of survival of 0.9, so one woul d expect one Tabl e 3. Fati gue l i fe prediction: residual strength mode l Maximum stress, Oa (MPa) Number of cycles, N Residual strength after N cycles, or (MPa) 1st block 2nd block 3rd block 4th block 1 480 8 100 1 200 50 000 990 225 000 1 310 23 000 1 800 500 1 630 1 900 1 100 115 000 850 1 000 000 .1 939 1 905 1 857 1 772 1 938 1 904 1 855 1 768 1 938 1 903 1 855 1 769 1 936 1 901 1 850 1 757 1 923 1 883 1 822 Failure 1 914 1 870 1 798 - - 1 913 1 869 1 797 - - 1 913 1 869 1 797 - - Equivalent tensile strength = 1 945 MPa. Number of cycles to failure = 4 576 600. 138 R. J. Hust on Table 4. Fatigue spectrum test results Specimen Fatigue life Residual strength (cycles) (MPa) 1 - - 2 322 2 - - 2098 3 - - 2 094 4 3 153 350 - - 5 - - 2 450 6 - - 1908 7 308 390 - - 8 - - 1834 9 - - 2166 10 - - 1971 speci men in t en t o have a s hor t e r t han pr edi ct ed life. Wh e n t he t est s wer e car r i ed out , t wo speci mens out of t he t en had lives s hor t er t han pr edi ct ed. Cons i der i ng t he smal l sampl e, t he cor r el at i on is adequat e. R E S I D U A L S T I F F N E S S M O D E L The r esi dual stiffness mode l of Whi t wor t h is a phenomenol ogi cal mode l and is l i mi t ed t o speci mens s ubj ect ed t o cons t ant ampl i t ude fat i gue l oadi ng. The mode l r equi r es t he cal cul a- t i on of t hr ee pa r a me t e r s f r om e xpe r i me nt a l dat a, and it is as s umed t hat r esi dual stiffness decr eas es monot oni cal l y as t he n u mb e r of cycl es i ncr eases. The basis of t he mode l is an equat i on t hat assumes t hat t he r at e of t he r esi dual modul us r educt i on is i nver sel y pr opor t i onal t o a powe r of t he r esi dual modul us itself: d E( N' ) - Df f ( E ( 0 ) , o, ) - - - ( 5 ) d N' x E X - I ( N ' ) wher e E ( N' ) is t he r esi dual modul us af t er N fat i gue ~ycles, N' is t he r at i o of t he appl i ed cycl es t o t he fat i gue life, E( 0) is t he initial modul us , x and D ar e const ant s and f ( E( 0) , Oa) is a f unct i on of t he initial modul us and t he st ress r ange, Oa. I nt egr at i on of eqn (5) f r om 0 t o N' : E~ ( N ' ) = E~(O) - DU( E( O) , o , ) N (6) and this r epr es ent s t he r esi dual modul us af t er a cer t ai n por t i on of t he f at i gue life has been compl et ed. If it is as s umed t hat f i E( 0) , oa) is a l i near f unct i on of t he stiffness change over t he cyclic life f ( E( 0) , o~) = k A E ( N' ) (7) and if t he s t r es s - s t r ai n r es pons e is l i near t o fai l ure, t hen at f ai l ur e E ( N' ) = Ef and a a ---~ Ef ef (8) wher e Ef is t he r e duc e d stiffness at f ai l ur e and ef is a cons t ant r el at ed t o t he st r ai n at f ai l ur e. Equat i on (6) be c ome s E x ( N ' ) = Ex(O) - OkX[ E( O) - Oa/ef]XN ' = Ex ( o) - HI E( 0 ) - B] x N ' (9) wher e H = Dk x and B = aa/ef. As can be s een in eqn (9), t hr e e pa r a me t e r s (x, H and B) char act er i s e t he stiffness r educt i on and ar e de t e r mi ne d exper i ment al l y, x and H ar e i nde pe nde nt of t he t est var i abl es and B is de pe nde nt on t he ma xi mum appl i ed st ress. Mor e det ai l s of t he mode l can be f ound in Ref s 11 and 15. To de t e r mi ne t he const ant s x, H and B, f at i gue t est s wer e car r i ed out on five speci mens. Dur i ng each t est t he st at i c modul us was me a s ur e d on several occasi ons. The r educt i on in modul us f or all five speci mens is pl ot t ed agai nst cycl es in Fig. 8. The modul us af t er N cycl es is gi ven as a f r act i on of t he initial modul us and t he n u mb e r of cycl es is gi ven as a f r act i on of t he f at i gue life. The ma xi mum st ress was 1875MPa and t he fat i gue lives var i ed f r om 2520 cycl es t o 344 310 cycles. Fi gur e 8 shows t hat t he r educt i on in modul us, t hough slight, is al most l i near up t o a fract i onal life of about 0-75, af t er whi ch t he decr eas e is a little mor e rapi d. Howe ve r , t he ma xi mum obs er ved dr op was onl y 6%, f r om 135 t o about 127 GPa. The val ues of t he const ant s x, H and B wer e f ound t o be 1-15, 1.74 and 127, r espect i vel y. The cur ve ge ne r a t e d by eqn (9) appear s t o fit t he exper i ment al resul t s qui t e well. The meas- ur ed r educt i on in stiffness was r a t he r smal l and per haps uni di r ect i onal car bon r e i nf or c e d epoxy was not t he cor r ect mat er i al t o t est t he model . However , one can cl ear l y see t he necessi t y f or accur at e modul us me a s u r e me n t if t he mode l is t o be appl i ed successful l y. CONCLUSI ONS The r esi dual s t r engt h mode l of Sendeckyj and t he r esi dual stiffness model of Whi t wor t h have been used t o descr i be t he r e pe a t e d t ens i on fat i gue pr oper t i es of uni di r ect i onal car bon fi bre r ei nf or ced epoxy. Bot h model s fi t t ed t he exper i ment al resul t s qui t e well. Fatigue life prediction in composites 139 1.0 0.9 0.8 L L I 0.7 Z v LJJ 0.6 . . . . . . . . A ~ ~, " 0.5 0.4 w Ak~ I I I I I I I 1 I 0 0.1 0.2 0.3 0.4 0.S 0.6 0.7 0.8 0.9 1.0 F r a c t i o n a l l i f e , N ' F i g . 8. Normalised static modulus as a function of the fraction of fatigue life at a maximum stress of 1875 MPa. The resi dual st rengt h model was also used t o predi ct fatigue life under spect r um l oadi ng. The correl at i on bet ween t he predi ct i on and t he experi ment al observat i ons was satisfactory. If t he assumpt i ons on whi ch t he resi dual st rengt h model is based are shown to be valid, t hen t he model can be used with some confi dence. The par amet er s C and S are pr obabl y dependent on factors like l ami nat e or i ent at i on, envi r onment al condi t i ons and R- r at i o, so t he model woul d need t o be modi fi ed to accommod- ate these. However , wor k carri ed out at Nor t hr op has i ndi cat ed t hat , if strain is used to define fatigue l oad i nst ead of stress, less modification is r equi r ed. 1 6 The resi dual stiffness model also pr ovi ded a reasonabl e fit t o t he exper i ment al results obt ai ned. It shoul d pr ove t o be a useful t ool in moni t ori ng fatigue life, as long as t he modul us is measur ed accurat el y. Its mai n di sadvant age is t hat it can be appl i ed onl y to const ant ampl i t ude fatigue loadings. REFERENCES 1. Mandell, J. F., Huang, D. D. & McGarry, F. J., Tensile fatigue performance of glass fibre dominated composites. Comp. Technol. Rev., 3 (1981) 96-102. 2. Halpin, J. C., Jerina, K. L. & Johnson, T. A. , Characterization of composites for the purpose of reliability evaluation. In Analysis of Test Methods for High Modulus Fibres and Composites (ASTM STP 521). ASTM, Philadelphia, PA, 1973, 5-64. 3. Yang, J. N., Reliability prediction for composites under periodic proof tests in service. In Proceedings of 4th Conference on Composite Materials: Testing and Design (ASTM STP 617). ASTM, Philadelphia, PA, 1977, 272-95. 4. Whitney, J. M., Residual strength degradation model for competing failure modes. In Long Term Behaviour of Composites (ASTM STP 813). ASTM, Philadelphia, PA, 1983, 483-97. 5. Sendeckyj, G. P., Fitting models to composite materials fatigue data. In Test Methods and Design Allowables for Fibrous Composites (ASTM STP 734). ASTM, Phila- delphia, PA, 1981, 245-60. 6. Componeschi, E. T. & Stinchcomb, W. W., Stiffness reduction as an indicator of damage in graphite/epoxy laminates. In Composite Materials: Testing and Design (Sixth Conference) (ASTM STP 787). ASTM, Philadel- phia, PA, 225-46. 7. Schulte, K., Stiffness reduction and development of longitudinal cracks during fatigue loading of composite laminates. In Mechanical Characterization of Load Bearing Fibre Composite Laminates, ed. A. H. Cardon & G. Verchery. Elsevier Applied Science Publishers, London, 1985, pp. 36-54. 8. Yang, J. N., Jones, D. L. , Yang, S. H. & Meskini, A., A stiffness degradation model for graphite/epoxy laminates. In Proceedings of AI AA/ ASME/ ASCE/ AHS 29th Structures, Structural Dynamics and Materials Conference, Williamsburg, Virginia, 1988, 958-66. 9. Rotem, A., Stiffness change of a graphite epoxy liminate under reverse fatigue loading. J. Comp. Technol. Res., U (1989) 59-64. 10. Naeem, M., Fatigue damage-compliance relationship for GRP. Composites, 20 (1989) 589-92. 11. Whitworth, H. A., Modelling stiffness reduction of graphite/epoxy composite laminates. J. Comp. Mater., 21 (1987) 362-72. 140 R. J. Huston 12. Reinfsnider, K. L., Damage and damage mechanics. In Fatigue of Composite Materials, ed. K. L. Reifsnider. Elsevier Science Publishers, Amsterdam, The Nether- lands, 1990, pp. 11-77. 13. Wang, A. S. D. , Chou, P. C., Lei, C. S. & Bucinell, R. B., Cumulative damage model for advanced composite materials, AFWAL-TR-85-4104. Wright-Patterson Air Force Aeronautical Laboratory, Dayton, OH, 1985. 14. Van Dijk, G. M. & De Jonge, J. B., Introduction to FALSTAFF. In Proceedings of 8th ICAF Symposium, International Committee on Aeronautical Fatigue, Lausanne, 1975, 3.61/1-39. 15. Whitworth, H. A., Static and dynamic stiffness reduction of graphite/epoxy composite laminates under fatigue loading. DSc Thesis, George Washington University, Washington DC, 1983. 16. Jeans, L. L., Deo, R. , Grimes, G. C. & Whitehead, R. S., Durability certification of fighter aircraft primary composite structure. In Proceedings of 11th ICAF Symposium, International Committee on Aeronautical Fatigue, Noordwijkerhout, 1981, 3.1/1-54.