journal of composite materials-2015-song-3157-68.pdf

Upload: vahid030951

Post on 06-Jul-2018

226 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/16/2019 Journal of Composite Materials-2015-Song-3157-68.pdf

    1/12

    JOURNAL OF

    C O MP O S I TE

    MATER IALS Article

    Study on static and fatigue behaviors of carbon fiber bundle and the statisticaldistribution by experiments

     Jian Song, Weidong Wen, Haitao Cui and Sibo Zhao

    Abstract

    In this paper, the static, tension–tension fatigue, and residual strength behaviors of carbon fiber bundle have beeninvestigated using experimental methods. The corresponding mechanical models and statistical distribution regulationshave also been established based on the experimental data. The test results indicate that the test section length of fiberbundle had a marked effect on the static strength, which decreased with the increase of length. A two-parameter Weibulldistributional function was used to describe the distribution of static strength. Furthermore, the S-N curve was obtained

    by a new segmented function to reflect the changing of fatigue life against stress levels. The experimental results of residual strength show that there was an abnormal tendency, which increased first and decreased later as the increase of cyclic number, for the residual strength of carbon fiber bundle in both tested stress levels 87% and 80%, respectively.Therefore, a new model for the residual strength was put forward and the good agreement between the fitting curvesand experimental was obtained.

    Keywords

    Carbon fiber bundle, static strength test and distribution law, fatigue life test and S-N, fatigue life distribution, residualstrength test and mechanical model

    Introduction

    Carbon fiber reinforced composites have been widely

    used as reinforcement in polymer composites, owing to

    their good mechanical properties and light weight. Most

    of the early works focus on the certainty performance,1

    such as the stiffness, strength, and fatigue characteristic

    of composites, which causes that composites are often

    overdesigned, such as heavier and more costly than

    necessary. Therefore, probability design methods have

    been gradually conducted into composites research.

    Schaff and Davidson2 proposed a strength-based wear-

    out model to predict the fatigue life and residual strength

    of composite laminates and the related distributions

    were investigated by a two parameter Weibull function.

    Cheng and Hwu3 investigated the fatigue reliability of 

    composite laminates subjected to constant amplitude

    loading and a residual analysis model was proposed to

    predict the residual strength. Whitworth4–6 studied the

    statistical distribution laws on graphite/epoxy compos-

    ites based on a set of static and fatigue tests and put

    forward to p-E-N and p-S-N models. Yang et al.7–9

    proposed a three-parameter fatigue and residual

    strength degradation model to predict the statistical fati-

    gue behavior of angle-ply composites. Wu and Yao10

    derived the fatigue life distribution of composite lamin-

    ates based on their static strength distribution.

    Although carbon fiber reinforced composites has

    been studied in many aspects, the research on the per-

    formance of carbon fiber bundle as main component is

    quite limited. There are still some issues associated with

    that left to be solved in the respects of non-uniform qual-

    ityof filaments, length of limited fiber bundle and fatigue

    characteristic, all of which have an impact on the static

    and fatigue performance to some extent. Moreover, it is

    inevitable to consider the statistical distribution

    owing to the random question of self-defect and

    College of Energy and Power Engineering, Nanjing University of 

    Aeronautics and Astronautics, China

    Corresponding author:

     Jian Song, College of Energy and Power Engineering, Nanjing University of 

    Aeronautics and Astronautics, No. 29 Street, Nanjing 210016, China.

    Email: [email protected]

     Journal of Composite Materials

    2015, Vol. 49(25) 3157–3168

    ! The Author(s) 2014

    Reprints and permissions:

    sagepub.co.uk/journalsPermissions.nav

    DOI: 10.1177/0021998314560385

     jcm.sagepub.com

     at TECNISCHE UNIV BRAUNSCHWEIG on October 15, 2015 jcm.sagepub.comDownloaded from 

    http://jcm.sagepub.com/http://jcm.sagepub.com/http://jcm.sagepub.com/http://jcm.sagepub.com/

  • 8/16/2019 Journal of Composite Materials-2015-Song-3157-68.pdf

    2/12

  • 8/16/2019 Journal of Composite Materials-2015-Song-3157-68.pdf

    3/12

    corresponding results are presented in Table 1, which

    are also the residual strengths  R(0) in the condition of 

    zero cycle.

     U ¼   4Pd 2 3000   ð1Þ

    where   P   is the peak of static load;   d   is the nominaldiameter of carbon fiber bundle in which   d   is equal

    to 7   mm.

    From Table 1, we can see that the static strength of 

    fiber bundles was increased with the decrease of charac-

    teristic lengths, which indicated a size effect on the static

    strength of fiber bundles. In addition, the representative

    load–displacement curves for different characteristic

    lengths and the curve of influence of characteristic

    length on average strength fitted by the model 2 were

    shown in Figure 2, respectively. It is clearly found that

    an obvious non-linear phenomenon happened as the

    increase of load and also the average static strengths of 

    fiber bundle with different lengths were remarkable

    reduced with the increase of characteristic length.

    S 0 ¼   c0

    L

    1=mð2Þ

    where  S 0  is the average strength,  L  is the characteristiclength (30, 40, and 50 mm),  c  and  m  are the constants

    determined by experimental data (m ¼ 1.632,c0 ¼ 3,625,272.57).

    Damaged photographs subjected static loading are

    given in Figure 3. From this figure, the location in

    the surface of fiber bundles appears as the amountof fluffiness phenomena, which indicates that there

    exists a certain quantity of monofilament taking

    place failure gradually so that a de-bunching failure

    mode finally happens in the middle of carbon fiber

    bundle.

    Statistical distribution of static strength.   To research the

    statistics distribution of fiber bundles, the two-para-

    meter Weibull distribution function (3) is used to fit

    the static strengths   R(0) and thus we can plot the

    related probability paper and distribution of static

    strength in Figure 4.

    F ðX Þ ¼ P   U   X f g ¼ 1 exp     X 

     

    !  ð3Þ

    Table 1.   Static strength test results of carbon fiber bundles.

    Characteristiclength

    Specimennumber

    Peak load (N)

    Strength(MPa)

    Averagestrength (MPa)

    Scaleparameter,   

    Shapeparameter,  

    50 mm L50-1 112.84 977.32 946.50 964.83 23.99L50-2 103.35 895.17

    L50-3 104.05 901.23

    L50-4 113.21 980.57

    L50-5 106.95 926.31

    L50-6 110.79 959.57

    L50-7 113.81 985.73

    40 mm L40-1 127.84 1107.29 1097.75 1111.23 37.61

    L40-2 123.43 1069.05

    L40-3 122.02 1056.91

    L40-4 129.51 1121.73

    L40-5 129.64 1122.84L40-6 128.00 1108.71

    30 mm L30-1 130.72 1132.25 1172.73 1188.73 34.75

    L30-2 131.74 1141.09

    L30-3 132.65 1148.91

    L30-4 133.48 1156.11

    L30-5 134.92 1168.58

    L30-6 137.84 1193.86

    L30-7 140.68 1218.47

    L30-8 141.15 1222.56

    Song et al.   3159

     at TECNISCHE UNIV BRAUNSCHWEIG on October 15, 2015 jcm.sagepub.comDownloaded from 

    http://jcm.sagepub.com/http://jcm.sagepub.com/http://jcm.sagepub.com/http://jcm.sagepub.com/

  • 8/16/2019 Journal of Composite Materials-2015-Song-3157-68.pdf

    4/12

    Figure 3.  Static tensile test of carbon fiber bundle: (a) pre-test; (b) during the test; (c) photograph of failure specimen.

    0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4

    0

    20

    40

    60

    80

    100

    120

    140

    160(a) (b) L30-1 : L=30mm

     L40-3 : L=40mm

     L50-4 : L=50mm

       L  o  a   d   /   N

    displacement/%

    25 30 35 40 45 50 55 60 65

    800

    900

    1000

    1100

    1200

    1300

    1400

     Fitting curve by Eq.(2)

     Experimental data

       A  v  e  r  a  g  e  s   t  r  e  n  g   t   h   /   M   P  a

    Fiber length/mm

    Figure 2.   (a) Load–displacement curve of typical fiber bundle with different characteristic lengths; (b) influence of characteristic

    length on average strength.

    3160   Journal of Composite Materials 49(25)

     at TECNISCHE UNIV BRAUNSCHWEIG on October 15, 2015 jcm.sagepub.comDownloaded from 

    http://jcm.sagepub.com/http://jcm.sagepub.com/http://jcm.sagepub.com/http://jcm.sagepub.com/

  • 8/16/2019 Journal of Composite Materials-2015-Song-3157-68.pdf

    5/12

    where   F (X ) represents the statistical distribution of 

    carbon fiber bundles;    U   means the static strength;    

    and     are scale parameter and shape parameter,

    respectively.

    From Figure 4, several important findings about the

    static strength statistical distribution of carbon fiber

    bundles were given as follows: First of all, it is logicalto consider that the statistical distribution for fiber

    bundle strength can be estimated by using a two-

    parameter Weibull distribution function (3) according

    to the correlation coefficients (R ¼ 0.903, 0.884, and0.865 for   L ¼ 50, 40, and 30mm, respectively) inFigure 4(a) to (c). Furthermore, although the strength

    of fiber bundles varies with different characteristic

    length, the value of shape parameters has a no signifi-

    cant change (refer to Figure 4(d)) and the general

    change trend is that with the increase of length, the

    curve slope of Weibll distribution function in view of 

    one type of length decreases gradually.

    In addition, the Kolmogorov–Smirnov hypothesis

    tests were applied to verify quantitatively whether the

    proposed two-parameter Weibull distribution function

    is suitable for the statistical distribution of carbon fiber

    bundles. The hypothesis test results are listed in Table 2

    and the method is introduced briefly in Appendix 1.

    According to Table 2, the values for all of the teststatistics (sqrt(n)*D) are outside the reject domain,

    therefore, it is reasonable to accept the hypothesis

    that the statistical distribution for static strength of 

    fiber bundles follows the Weibull distribution law.

    Fatigue life test

    Testing results of fatigue life.   A set of 22 specimens sub-

     jected to nine various constant amplitude stress levels

    are listed in Table 3. The S-N curve of carbon fiber

    bundles referred to equations (4) to (6) was then plotted

    in Figure 5 and the related parameters fitted by the

    -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0

    6.78

    6.80

    6.82

    6.84

    6.86

    6.88

    6.90

    (a) (b)

    (c) (d)

     L50: Experimental data

     L50: Fitting curving

       l  n   X

    ln[-ln(1-F(X))]

    Equation y = a + b

     Adj. R-Squ 0.90297

    Value Standard Er 

    B Intercep 6.871 0.00551

    B Slope 0.041 0.00553

    -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0

    6.96

    6.97

    6.98

    6.99

    7.00

    7.01

    7.02

    7.03

    7.04

    Equation y = a + b Adj. R-Squa 0.88385

    Value Standard Err 

    D Intercept 7.0132 0.00409

    D Slope 0.0265 0.00426

     L40: Experimental data

     L40: Fitting curving

       l  n   X

    ln[-ln(1-F(X))]

    -2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0

    7.01

    7.02

    7.03

    7.04

    7.05

    7.06

    7.07

    7.08

    7.09

    7.10

    7.11

    Equat ion y = a + b

     Adj. R-Squa 0.86457

    Value Standard Err 

    H Intercept 7.0806 0.00437

    H Slope 0.0287 0.00426

       l  n   X

     L30: Experimental data

     L30: Fitting curving

    ln[-ln(1-F(X))]

    800 900 1000 1100 1200 1300 1400

    0.0

    0.2

    0.4

    0.6

    0.8

    1.0

     L30:Fitting curving

     L40:Fitting curving

     L50:Fitting curving

     L30: Experimental data

     L40: Experimental data

     L50: Experimental data

       F  a   i   l  u  r  e  p  r  o   b  a   b   i   l   i   t  y

    Static strength/MPa

    Figure 4.   Weibull distribution probability paper of static strength (a)–(c) and failure probability of carbon fiber bundles with different

    characteristic lengths (d).

    Song et al.   3161

     at TECNISCHE UNIV BRAUNSCHWEIG on October 15, 2015 jcm.sagepub.comDownloaded from 

    http://jcm.sagepub.com/http://jcm.sagepub.com/http://jcm.sagepub.com/http://jcm.sagepub.com/

  • 8/16/2019 Journal of Composite Materials-2015-Song-3157-68.pdf

    6/12

    fatigue test data was directly given in equations (4)

    to (6).

    From the experimental results in Figure 5(a) (open

    triangle ‘‘4’’), it is concluded that the normalized stress

    basically underwent a linear trend between 0.85 and

    0.94, while there was an obvious decay trend in the

    range of 0.73 to 0.85. When the normalized stress was

    less than 0.73, all of the fatigue life of fiber bundles was

    more than 1e6, which was defined as the infinite life.

    Therefore, it can be used a linear function to describe

    the variation trend between 0.85 and 0.94 and an expo-

    nential model to describe the change regularity of 

    0.73–0.85, respectively. The determined relationshipsbetween fatigue life and normalized stress were

    expressed in the following equations

    1. 0:85  max U 

    0:94 max

     U ¼ 0:0506l g N  f  þ 1:059   ð4Þ

    2. 0:75  max U 

    0:85

     max

     U  ¼1

    0:310

      lg N  f  4:014

    lg N  f  1:469 0:182

    ð5

    Þ

    3.    max U 

    0:75

    lg N  f 4 6   ð6Þ

    From Figure 5(a), it is clearly found that good coin-

    cidence between experiment and theory was obtained

    for the fatigue issue of carbon fiber bundles. Moreover,

    failure photograph under the fatigue load was shown in

    Figure 5(b) and it can be seen that the failure model

    mainly manifested de-bunching failure, which was simi-lar with the mode as that in static strength.

    Nevertheless, the difference was that the failure regions

    were concentrated in the middle of specimens compared

    to that in the static tests.

    Statistical distribution of fatigue life.  Furthermore, in order

    to investigate the distribution law for the fatigue of 

    fiber bundles, we assume that the statistical distribution

    of the fatigue life also follows a two-parameter Weibull

    distribution with reference to composite laminates3,6,18

    F ðnÞ ¼ P N  f   n ¼ 1 exp     n

    l !   ð7Þ

    where   F (n) denotes the fatigue life distribution of 

    carbon fiber bundles,   N   is the fatigue life,    and   l  are

    the scale and shape parameter, respectively both of 

    which can be calculated by the same method as the

    static strength.

    The calculated results are presented in Table 4 and

    corresponding distributional curves are plotted in

    Figure 6. As for these results, the statistical distribution

    Table 3.   Tension–tension fatigue test results of carbon fiber

    bundles.

    Stress

    level

    Specimen

    number

    Fatigue

    life,  Nf     lgNf     Average lgNf  

    94% U   1 185 2.267 2.372

    2 243 2.386

    3 291 2.464

    92% U   4 629 2.799 2.721

    5 500 2.699

    6 450 2.653

    87% U   7 6984 3.844 3.836

    8 7949 3.900

    9 5923 3.773

    10 6550 3.816

    85% U   11 10,881 4.037 4.064

    12 12,363 4.092

    80% U   13 14,544 4.163 4.248

    14 20,883 4.320

    15 15,869 4.201

    16 19,486 4.290

    75% 17 235,079 5.371 5.397

    18 239,358 5.379

    19 274,297 5.438

    73% 20   >1e6   >6   >670% 21   >1e6   >6   >6

    60% 22   >1e6   >6   >6

    Table 2.  Hypothesis test results for static strength distribution

    function.

    Characteristic lengths   L ¼ 30mm   L ¼ 40mm   L ¼ 50mmNumber of specimens 8 6 7

    Kolmogorov  D   0.2388 0.2974 0.2438

    sqrt(n)*D   0.6755 0.7285 0.6450Reject domain at

    significant level 0.05

    [1.36,1)

    D ¼ supjF n( x ) F ( x )j, where  F ( x ) and  F n( x ) are the assumed distributionalfunction and distributional function of samples, respectively; sqrt(n)*D

    represents test statistics.

    3162   Journal of Composite Materials 49(25)

     at TECNISCHE UNIV BRAUNSCHWEIG on October 15, 2015 jcm.sagepub.comDownloaded from 

    http://jcm.sagepub.com/http://jcm.sagepub.com/http://jcm.sagepub.com/http://jcm.sagepub.com/

  • 8/16/2019 Journal of Composite Materials-2015-Song-3157-68.pdf

    7/12

    of fatigue life was fitted by the two-parameter Weibull

    distribution fairly well, as shown in Figure 6(c).According to the statistical results mentioned above,

    the excellent correlation implies that adoption of 

    Weibull distributional function in carbon fiber bundle

    fatigue analysis is feasible. In addition, there was a

    more discreteness in low stress level, which was also

    reasonable because the lower stress level causes that

    the fatigue damage takes place in more local positions

    in the surface of fiber bundles and the propagation

    paths are also more complex.

    In addition, the similar approach as the static

    strength was used to investigate whether the hypothesis

    distribution regulation was reasonable for the fatigue

    life of carbon fiber bundle at various stress levels. The

    quantitative results are given in Table 5. It is clearly

    seen that all the values of   D   were less than 1.36,

    which was the lower limit of reject domain. Thus, the

    hypothesis can be accepted.

    Testing results for residual strength of carbon

    fiber bundles

    Testing results of residual strength.   Fourteen specimens

    were fatigued at two maximum stress levels 87% and

    80%, respectively, up to 1/3 and 2/3 corresponding

    average fatigue life, which were 7852 for 87% stress

    level and 17696 for 80%. Their residual strength tests

    were then performed and the results are listed in

    Table 6. The failure photograph related to residual

    strength tests is shown in Figure 7(a).

    From Table 7, it is obviously found that the residual

    strengths had a non-monotonic tendency with the

    increase of cycle, which is similar to the notched

    strength of woven composites.19,20 Figure 7(a) exhibits

    several serious failure models with de-bunching and

    fracture damages in the middle of fiber bundles. Thepossible reason for the phenomenon could be that the

    monofilaments in the fiber bundle are gradually dis-

    persed under the fatigue load, which may lead to

    more uniform bearing in each of monofilaments.

    Therefore, a higher load and further serious damage

    are shown in the following static tensile test.

    Meanwhile, a briefness ‘‘bang’’ sound could be listened

    when the fracture occurred.

    Furthermore, in order to investigate the difference

    between static and residual strength, Figure 7(b) pre-

    sents the load–displacement curves for three represen-

    tative specimens used in static strength and residual

    strength tests, respectively. It is found that the value

    of residual strength experienced an increase firstly,

    and then showed a decreased trend against cycle

    number. The reasons of this abnormal phenomenon

    may be that on the one hand, the stress state for each

    of monofilaments is more uniform as mentioned above,

    which could lead to the more bearing capacity for single

    fiber bundle. On the other hand, the extent of accumu-

    lative damage in monofilaments may increase subjected

    cyclic loading, which could cause the decrease of the

    mechanical performance to some degree. Therefore,

    2 3 4 5 6

    0.6

    0.7

    0.8

    0.9

    1.0

     

      m  a  x 

      u   l   t

    lgNf 

    (a) (b)

    Figure 5.  (a) The curve of fiber bundles normalized stress against lgNf  ; (b) fatigue test specimen of carbon fiber bundle.

    Table 4.  Parameters in Weibull distributional function of fatiguelife.

    Weibull distribution

    parameters

    87%

    stress level

    80%

    stress level

    G    7263.01 19,100.89

    l   6.96 5.11

    Song et al.   3163

     at TECNISCHE UNIV BRAUNSCHWEIG on October 15, 2015 jcm.sagepub.comDownloaded from 

    http://jcm.sagepub.com/http://jcm.sagepub.com/http://jcm.sagepub.com/http://jcm.sagepub.com/

  • 8/16/2019 Journal of Composite Materials-2015-Song-3157-68.pdf

    8/12

    the variation in residual strength versus cyclic number

    may be the result of interaction of both factors

    mentioned.

    In order to describe the phenomenon quantitatively,

    a new model was proposed. It is assumed that

    dRðnÞdn

      ¼  gðS , rÞ f ðnÞ ð8Þ

    -1.5 -1.0 -0.5 0.0 0.5

    8.65

    8.70

    8.75

    8.80

    8.85

    8.90

    8.95

    9.00

    (a) (b)

    (c)

     Stress leve : 87%  Ult Fitting curve

    Equation y = a + b*

     Adj. R-Squar 0.95498

    Value Standard Erro

    B Intercept 8.89055 0.01531

    B Slope 0.14367 0.01787

    ln[-ln(1-F(X))]

       l  n   X

    -1.5 -1.0 -0.5 0.0 0.5

    9.55

    9.60

    9.65

    9.70

    9.75

    9.80

    9.85

    9.90

    9.95

    10.00

     Stress leve : 80%  Ult Fitting curve

    Equation y = a + b*

     Adj. R-Squar 0.91796

    Value Standard Err 

    H Intercept 9.8574 0.02849

    H Slope 0.1955 0.03326

    ln[-ln(1-F(X))]

       l  n   X

    4000 8000 12000 16000 20000 24000

    0.0

    0.2

    0.4

    0.6

    0.8

    1.0

     Distribution function withmax

    =80% U

     Experimental data withmax

    =80% U

     Distribution function withmax

    =87% U

     Experimental data withmax

    =87% U

    Fatigue cycle Nf 

       F  a   i   l  u  r  e  p  r  o   b  a   b   l   i   t  y

    Figure 6.   Weibull distribution probability paper of fatigue life (a), (b) and failure probability (c).

    Table 6.  Residual strength tests results of carbon fiber bundles.

    Stress

    levels

    Terminal

    number

    Peak 

    load (N)

    Strength

    (MPa)

    Average

    strength (MPa)

    87% 1/3Nf  * 169.41 1467.30 1391.34

    150.69 1305.19

    158.58 1373.58

    163.86 1419.28

    2/3Nf  * 155.26 1344.81 1337.41

    156.59 1356.34153.66 1330.93

    152.12 1317.57

    80% 1/3Nf  ** 153.36 1328.33 1333.86

    152.45 1320.43

    156.19 1352.82

    2/3Nf  ** 110.35 955.81 1085.37

    136.94 1186.06

    128.64 1114.23

    Nf  * and   Nf  ** means the fatigue life of 87% and 80% stress levels,

    respectively.

    Table 5.   Hypothesis test results for fatigue life distribution

    function.

    Stress levels 87% 80%

    Number of specimens 4 4

    Kolmogorov  D   0.2465 0.2696

    sqrt(n)*D   0.4931 0.5392

    Reject domain at

    significant level 0.05

    [1.36,1)

    3164   Journal of Composite Materials 49(25)

     at TECNISCHE UNIV BRAUNSCHWEIG on October 15, 2015 jcm.sagepub.comDownloaded from 

    http://jcm.sagepub.com/http://jcm.sagepub.com/http://jcm.sagepub.com/http://jcm.sagepub.com/

  • 8/16/2019 Journal of Composite Materials-2015-Song-3157-68.pdf

    9/12

    Indicating that  R(n) (the rate of residual strength at

    the   nth stress cycles) is equal to the product of the

    effect,  g(S , r), of the stress range,  S , of the stress ratio,

    r and the effect,  f (n), of the number of cycle. The stress

    range   S   is defined as the difference between the max-

    imum and minimum cycle stress. So, the simplest

    approximation to the functional form of   g(S , r) and

     f (n) is expressed as

     gðS Þ ¼ k1S b,   S ¼ ð1 rÞ max,   k14 0 f 

    ðn

    Þ ¼ Qn

    þv,   Q4 0,   v4 0

    ð9Þ

    where k1, b, Q, and v  are constants. In equation (9), it is

    assumed for simplicity that the stress ratio is fixed so

    that g(S, r) is a function of  S  alone and  f (n) is the linear

    function to ensure the variation of  R(n) is in accordance

    with the experimental data.

    Substituting equation (9) into equation (8) and car-

    rying out the integration from  n1   to  n2 cycles, we have

    Rðn2Þ Rðn1Þ ¼ k1S bQ

    2  n22 n21 þ k1vS bðn2 n1Þ

    ð10Þ

    For  n2¼ n,  n1¼ 0, equation (10) reduces to

    RðnÞ Rð0Þ ¼ k1S bQ

    2  n2 þ k1vS bn   ð11Þ

    Let  a ¼ k1Q/2,  c ¼ k1v, then

    RðnÞ ¼ Rð0Þ S bðan2 cnÞ ð12Þ

    Equation (12) is a theoretical residual model of 

    carbon fiber bundle involving three parameters   a,   b,

    and  c, which will be determined by amount of residual

    strength tests in the following section.Based on the data of residual strength tests, the par-

    ameters involved in equation (12) can be obtained. The

    results for residual strength tests at high and low stress

    amplitudes are shown in Table 7 and the related curves

    are plotted in Figure 8. Additionally, the corresponding

    residual strength under 87% and 80% stress levels for

    1/3N  f  and 2/3N  f  cycles calculated by the equation (12)

    are listed in Table 8. According to the results, the max-

    imum error was only 8.14%, which took place under

    80% stress levels for 2/3N  f , therefore, it is reasonable to

    accept the theoretical formula (12) to predict the resi-

    dual strength of fiber bundle.

    Furthermore, as equation (12) reflects the relation-

    ship between the static strength and residual strength, it

    is necessary to further validate whether the converted

    static strengths   R(0) calculated by equation (13) are

    suitable to the statistical distribution of static strength.

    Therefore, the residual strength data listed in Table 6

    can be converted into the equivalent static strength

    using the model given by equation (13). And the cor-

    responding results are plotted in Figure 9 where the set

    of converted equivalent static strength,  R(0), using the

    values of   a,   b, and   c   given in Table 7 was plotted as

    0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4

    0

    30

    60

    90

    120

    150

    180(b)(a)

     Residual strength at 1/3Nf 

     Residual strength at 2/3Nf 

     Ultimate strength

       L  o  a   d   /   N

    displacement/%

    Figure 7.   Residual strength failure photograph of carbon fiber bundle.

    Table 7.   Parameters in residual strength model.

    Parameters

    Low stress

    amplitudes (80% U)

    High stress

    amplitudes (87% U)

    a   1.55e-3 3.86e-3

    b   0.532   1.009c    10.016 59.454

    Song et al.   3165

     at TECNISCHE UNIV BRAUNSCHWEIG on October 15, 2015 jcm.sagepub.comDownloaded from 

    http://jcm.sagepub.com/http://jcm.sagepub.com/http://jcm.sagepub.com/http://jcm.sagepub.com/

  • 8/16/2019 Journal of Composite Materials-2015-Song-3157-68.pdf

    10/12

    solid squares and the solid curve was the same Weibull

    distribution of the static strength presented in

    Figure 4(d) with the 50 mm characteristic length.

    Rð0Þ ¼ RðnÞ þ S bðan2 cnÞ ð13Þ

    where   R(0) represents the converted equivalent

    strength.

    It can be observed from Figure 9 that the correlation

    between the Weibull distribution obtained directly by

    static strength and converted data from residual

    strength was comparatively well despite of stress

    levels. Thus, it is reasonable to describe the residual

    strength at various stress levels by equation (12).

    Conclusions

    The performances of static strength, fatigue life, and

    residual strength for carbon fiber bundles of T300

    fibers have been tested and the corresponding mechan-

    ical models have also been built based on the testing

    results. The following conclusions can be made:

    1. The inner original defects derived from manufactur-

    ing process existing in each of monofilaments caused

    that the static strength of fiber bundle was correlated

    with test section length. Therefore, the specimen

    with a longer characteristic length has generally

    lower strength under static tests.

    2. Under T-T cyclic loading, the de-bunching phenom-

    ena are more obviously and mainly concentrated in

    the middle of fiber bundle, and less in the root part.

    A segmented function was proposed to describe the

    S-N curve of carbon fiber bundles and the related

    parameters were then obtained based on the fatigue

    0 1000 2000 3000 4000 5000 6000 7000

    0

    200

    400

    600

    800

    1000

    1200

    1400

    (a) (b)

     Residual strength curve obtained by Eq (12)

      of 87% stress level

     Average value of 87% stress level

     Experimental data of 87% stress level

       R  e  s   i   d  u  a   l  s   t  r  e  n  g   t   h   R   (  n   )

    Cycle number n

    0 4000 8000 12000 16000 20000

    0

    200

    400

    600

    800

    1000

    1200

    1400

     Residual strength curve obtained by Eq (12)

      of 80% stress level

     Average value of 80% stress level

     Experimental data of 80% stress level

    Cycle number n

       R  e  s   i   d  u  a   l  s   t  r  e  n

      g   t   h   R   (  n   )

    Figure 8.  Residual strength curves of carbon fiber bundle at high stress level (a) and low stress level (b).

    840 860 880 900 920 940 960 980 1000 1020 1040

    0.0

    0.1

    0.2

    0.3

    0.4

    0.5

    0.6

    0.7

    0.8

    0.9

    1.0

     Convered ultimate strength using equation (13)

      Weibull distribution obtained form equation (6)

       D   i  s   t  r   i   b  u   t   i  o  n   f  u  n  c   t   i  o  n

    Convered ultimate strength

    Figure 9.  Theoretical prediction of static strength and residual

    strength data converted to static strength for carbon fiber bundle

    under both stress levels.

    Table 8.  Comparison between the predicted values of residual

    strength based on equation (12) and test data.

    Stress

    levels

    Terminal

    number

    Test

    results

    Predicted

    results Error

    87% 1/3Nf  * 1391.34 1386.26 0.37%

    2/3Nf  * 1337.41 1345.20 0.58%

    80% 1/3Nf  ** 1333.86 1245.92 6.59%

    2/3Nf  ** 1085.37 1173.69 8.14%

    Nf  * and Nf  ** means the fatigue life of 87% and 80% stress levels, respect-

    ively. All of the test data are from Table 6.

    3166   Journal of Composite Materials 49(25)

     at TECNISCHE UNIV BRAUNSCHWEIG on October 15, 2015 jcm.sagepub.comDownloaded from 

    http://jcm.sagepub.com/http://jcm.sagepub.com/http://jcm.sagepub.com/http://jcm.sagepub.com/

  • 8/16/2019 Journal of Composite Materials-2015-Song-3157-68.pdf

    11/12

    life data. In order to investigate the statistical distri-

    bution of fatigue life, a two-parameter Weibull dis-

    tribution function was well used to describe the

    distribution.

    3. In the end, the residual strength tests were conducted

    and a theoretical model was derived based on the

    assumption that the residual strength of fiber bun-dles is non-monotonic. Additionally, the failure

    degree for residual strength tests was the most ser-

    ious compared to others, which had not only obvi-

    ous de-bunching damages, but also fracture

    damages.

    Funding

    This work was supported by Jiangsu Innovation Program for

    Graduate Education [grant number KYLX_0237].

    Conflict of interest

    None declared.

    References

    1. Degrieek J and Paepegem WV. Fatigue damage modeling

    of fibre-reinforced composite materials: Review.   Appl 

    Mech Rev  2001; 54(4): 279–300.

    2. Schaff JR and Davidson BD. Life prediction method-

    ology for composite structures, part I: Constant ampli-

    tude and two-stress level fatigue.  J Compos Mater  1997;

    31(2): 128–157.

    3. Cheng HC and Hwu FS. Fatigue reliability analysis of composites based on residual strength.   Adv Compos

    Mater   2006; 15(4): 385–402.

    4. Whitworth HA. Modeling stiffness reduction of graphite/

    epoxy composite laminates.  J Compos Mater  1987; 21(6):

    362–371.

    5. Whitworth HA. A stiffness degradation model for com-

    posite laminates under fatigue loading.   Compos Struct

    1998; 40(2): 95–101.

    6. Whitworth HA. Evaluation of the residual strength deg-

    radation in composite laminates under fatigue loading.

    Compos Struct  2000; 48(5): 261–264.

    7. Yang JN and Miller PK. Effect of high load on statistical

    fatigue of unnotched graphite/epoxy laminates. J Compos

    Mater  1980; 14(4): 82–94.

    8. Yang JN and Liu MD. Residual strength degradation

    model and theory of periodic proof tests for graphite/

    epoxy laminates.  J Compos Mater  1977; 11(2): 176–202.

    9. Yang JN and Jones DL. Statistical fatigue of graphite/

    epoxy angle-ply laminates in shear.   J Compos Mater

    1978; 12(4): 371–389.

    10. Wu FQ and Yao WX. A model of the fatigue life distri-

    bution of composite laminates based on their static

    strength distribution.   Chin J Aeronaut   2007; 21(6):

    241–246.

    11. Phani KK. Evaluation of single-fibre strength distribu-

    tion from fibre bundle strength.   J Mater Sci   1988;

    23(3): 941–945.

    12. Yu WD, Postle R and Gyan HJ. Evaluating single fiber

    and fiber bundle tensile curves.   Text Res J  2003; 73(8):

    875–882.

    13. Joffe R, Andersons J and Sparniņs ˇ   E. Applicability of 

    Weibull strength distribution for cellulose fibers withhighly non-linear behaviour. In:   Proceedings of the 17th

    international conference on composite materials (ICCM-

    17), Edinburg, UK, 27–31 July 2009.

    14. Yuan H, Wen WD, Cui HT, et al. The random crack core

    model for predicting the longitudinal tensile strengths of 

    unidirectional composites.   J Mater Sci   2009; 44(12):

    3026–3034.

    15. Zhu YL, Cui HT, Wen WD, et al. Experiments on fatigue

    damage failure test of carbon fiber yarn.   Acta Mater

    Compos Sin  2012; 29(5): 179–183.

    16. ASTM.   Standard test method for tensile strength and 

    Young’s modulus for high-modulus single filament mater-

    ials.  ASTM D3379-75. West Conshohocken, PA: ASTM,1989.

    17. ASTM.   Standard recommended practice for constant-

    amplitude low-cycle fatigue testing.   ASTM E606. West

    Conshohocken, PA: ASTM, 1980.

    18. Yang JN. Fatigue and residual strength degradation

    for graphite/epoxy composites under tension-

    compression cyclic loadings.   J Compos Mater   1978;

    12(19): 19–39.

    19. Li JL, Yang HN and Kou CH. Fatigue properties of 

    three dimensional braiding composites.   Acta Mater

    Compos Sin  2005; 22(4): 172–176.

    20. Ken G, Yu KF, Hiroshi H, et al. Fatigue behavior of 2D

    laminate C/C composites at room temperature.   Compos

    Sci Technol   2005; 65(5): 1044–1051.21. Teng SZ and Feng JH.  Mathematical statistics. Dalian:

    Dalian University of Technology Press, 2005.

    Appendix

    The Kolmogorov–Smirnov test.

    This method21 proposed by Kolmogorov was

    used to test quantitatively the difference degree

    between the assumed distributional function   F (x)

    and the distributional function of samples   F n(x)

    (plotted in Figure 10). From the following

    Figure 10.   Schematic diagram of  Dn.

    Song et al.   3167

     at TECNISCHE UNIV BRAUNSCHWEIG on October 15, 2015 jcm.sagepub.comDownloaded from 

    http://jcm.sagepub.com/http://jcm.sagepub.com/http://jcm.sagepub.com/http://jcm.sagepub.com/

  • 8/16/2019 Journal of Composite Materials-2015-Song-3157-68.pdf

    12/12

    equation, it is defined as a distance   Dn   to describe

    the difference

    Dn ¼ sup  F nðxÞ F ðxÞ   ð14Þ

    A reasonable test under a significant level is that if the hypothesis distribution is true, the value   Dn   men-

    tioned above has a reduced trend, else the value

    would be increased. Therefore, there exists a thresh-

    old value   k, i.e. when sqrt(n)*Dn< k, the hypothesis

    could be accepted, else be rejected, where sqrt(n) is

    used to reflect the effect of the number of test

    specimens.

    The limiting distribution of sqrt(n)*Dn   was derived

    by Kolmogorov as follows

    limn!1

    P ffiffiffi

    np 

      Dn  t ¼ F ðtÞ

    ðt

    Þ ¼1

    2X

    1

    i ¼1 ð1

    Þi 1e2i 

    2t2  ð15Þ

    where F (t) was calculated and listed in the limiting dis-

    tributional table of  Dn.

    Finally, the rejection domain can be obtained by

    checking out the limiting distributional table under a

    given significant levels  , i.e. when  ¼ 0.05, the rejectdomain  U ¼ [1.36, 1).

    3168   Journal of Composite Materials 49(25)

     at TECNISCHE UNIV BRAUNSCHWEIG on October 15, 2015 jcm.sagepub.comDownloaded from 

    http://jcm.sagepub.com/http://jcm.sagepub.com/http://jcm.sagepub.com/http://jcm.sagepub.com/