You are on page 1of 12

International Journal of Fatigue 31 (2009) 16481659

Contents lists available at ScienceDirect

International Journal of Fatigue


journal homepage: www.elsevier.com/locate/ijfatigue

A two-parameter analysis of SN fatigue life using Dr and rmax


K. Sadananda a, S. Sarkar a, D. Kujawski b, A.K. Vasudevan c,*
a b c

Technical Data Analysis, VA, United States Department of Mechanical and Aeronautical Engineering, Western Michigan University, Kalamazoo, MI 49008, United States Ofce of Naval Research, 875 North Randolph Street, Arlington, VA 22203, United States

a r t i c l e

i n f o

a b s t r a c t
The effect of the load ratio, R, or the mean-stress on fatigue life has been recognized for more than a hundred years. In considering the mean-stress effects in the stress-life (SN) approach, research efforts have been mostly concentrated in establishing correlating functions in terms of the ow stress or yield stress or the ultimate tensile stress, etc., by taking, say, R = 1 test results as a reference. Very little effort has been made towards understanding the role of stress range Dr and the maximum stress rmax, (or rmean) in the fatigue crack nucleation and propagation and also how to relate this to both the stress-life and the fracture-mechanics descriptions. In this paper we rst examine crack nucleation based on the stress-life approach using a two-parameter requirement in terms of Dr and rmax, and then connect it to crack propagation using the Kitagawa diagram as the incipient crack grows to become a long crack. Since stress-life data include both nucleation and propagation, the connection of the safe-life approach to the fracture-mechanics analysis is pertinent. Comparison of the present analysis with experimental data taken from the literature demonstrates that a two-parameter approach in terms of Dr and rmax forms a basis for the SN analysis. Published by Elsevier Ltd.

Article history: Received 27 October 2008 Received in revised form 18 February 2009 Accepted 3 March 2009 Available online 16 March 2009 Keywords: Stress controlled fatigue Kitagawa diagram Crack nucleation and growth Aluminum and steel alloys Mean stress effects

1. Introduction It has been shown previously [15] that an unambiguous description of fatigue crack growth requires two loading parameters: DK and Kmax. It also has been demonstrated that the fatigue crack growth phenomena including load ratio effects, underload and overload effects, environmental effects, acceleration of short cracks, etc., can be accounted for without invoking any extraneous factors, such as crack closure. Since both DK and Kmax govern fatigue crack growth, the natural consequence of this is the existence of two limiting thresholds, namely DK and K th max;th , which must be satised simultaneously for a crack to grow. For a given material environment system, these two thresholds measure an intrinsic material resistance based on the mechanism of cracking. Similarly, for any nonzero constant da/dN, fatigue crack growth is described by two-parameter, DK and K max which vary with crack growth rate and correspond, respectively, to the DK value at asymptotically high Kmax, and the Kmax value at asymptotically high DK. It has been shown that a map of these parameters in terms of DK vs. K max for the range of da/dN provides a characteristic crack growth trajectory which is characteristic of the physical crack growth mechanisms [6]. In this paper the term crack growth trajectory or simply trajectory is understood to refer to this curve,
* Corresponding author. E-mail address: vasudea@onr.navy.mil (A.K. Vasudevan). 0142-1123/$ - see front matter Published by Elsevier Ltd. doi:10.1016/j.ijfatigue.2009.03.007

implicitly parametric in da/dN, in DK* vs. K space. Trajectorymax maps of several engineering materials indicate that mechanisms operating at the crack tip vary with the material, the environment and also with transient times during crack increment [6]. In order to apply this methodology in practice, a two-parameter crack driving force in terms of DK and Kmax has been proposed [7,8]. Recently a UNIGROW model has been developed [9] which takes into consideration this two-parameter requirement. This two-parameter driving force approach provides an effective predictive methodology without any need for adjustable parameters. It has been demonstrated that it predicts the fatigue crack growth behavior under service loading spectra [9]. If the two-parameter requirement is intrinsic to fatigue, then it should be applicable not only to fatigue crack growth but to the crack nucleation as well. Conventionally, two methodologies are used in fatigue analysis; a safe-life approach based on crack nucleation using stress-life or strain-life analysis, and a damage tolerance approach based on crack propagation using fracturemechanics analysis. In the past, the integration of the two approaches has not been very successful. Instead, these two approaches have been developed independently forcing a designer to select one or the other for practical fatigue engineering analysis. At the design stage, a designer relies heavily on the crack nucleation analysis, while at the maintenance stage one is forced to examine the damage tolerance approach, since cracks do form at critical locations during service, particularly in aging aircrafts.

K. Sadananda et al. / International Journal of Fatigue 31 (2009) 16481659

1649

Additionally, some material factors, for example grain size, ductility, and etcetera, affect crack nucleation and crack propagation differently. It is a fact that it is difcult to differentiate when nucleation ends and propagation begins. Part of this problem is associated with the limitations in crack detection methodologies. In fact, the crack propagation stage has already set in by the time a crack can be observed by currently available NDE techniques. Additional problems arise due to the limitations of the conventional fracturemechanics methods for short cracks, when only the remote applied stresses are considered in the analysis [10]. These limitations imply that short cracks can decelerate even with the increasing applied stress intensity factor, DK. Hence, short cracks are not easily amenable to analysis using conventional fracture-mechanics considering only the applied DK. As the Unied Approach [11] to fatigue ascertains, there are contributions to the crack tip driving force from internal stresses that are present at the incipient stage of short cracks, which need to be considered. The short crack problem is of paramount interest in terms of fatigue life prediction of engineering components, particularly for aircraft structures, since the time period associated with nucleation and short crack growth may occupy a signicant part of the total life. Hence, evaluation of crack nucleation and its transition to long crack via short crack growth are important stages to be considered for reliable fatigue life prediction. In this paper, we rst examine the crack nucleation based on the stress-life approach using the two-parameter requirement in terms of Dr and rmax. The effect of the load ratio, R, or the mean-stress on fatigue life [12,13] will be considered within the two-parameter framework. The Kitagawa diagram will be used as the connecting link between the crack nucleation and the growth of a crack leading to failure. Since stress-life data include both nucleation and propagation, the connection of the safe-life approach to the damage tolerance approach is pertinent. The conditions under which nucleated cracks do not propagate also will be discussed. Thus the full range of fatigue damage from nucleation to propagation using the two-parameter framework is addressed.

2. Two-loading parameter requirement for fatigue Fatigue-crack growth tests are customarily done at constant R or at constant Kmax. We have shown previously that all these tests are complimentary to extract the material behavior in a given environment [15]. In contrast to the crack growth tests, the stress-life tests using smooth specimens are usually performed with a constant mean-stress, rm. Most of the tests are done at rm = 0 or R = 1, using rotatingbending tests, since such tests are easy to conduct. To bring out the analogy between the fatigue crack growth behavior and the stress-life behavior, we will show a parallelism in the analysis for these two-sets of data. The data reduction procedure that has been used in fatigue crack growth (FCG) analysis will be adopted to analyze stress-life behavior, as is shown in Fig. 1. The crack growth rate data in a given environment for constant R or constant Kmax tests form the basic FCG data. The FCG rate data are normally plotted in terms of da/dN vs. DK. From these data one can plot DK vs. R and DK vs. Kmax for any given crack growth rate, da/dN. Fig. 1a illustrates the reduction scheme. Since DK vs. R generally is approximately bilinear, one can use this plot to extrapolate or interpolate the data to extract values at intermediate values of R, if available experimental data are limited to only few Rvalues. Then, a DK vs. Kmax plot provides the fundamental material curve for any given crack growth rate, da/dN. The DK vs. Kmax plot represents the interrelation between the applied values of DK and Kmax and the resistance of the material in order to sustain the selected crack growth rate. At the threshold (operationally, da/

dN $ 1010 m/cycle) we have a fundamental threshold curve (threshold is not a single value but a curve) below which a fatigue crack does not propagate. The threshold curve, thus, denes the non-propagating condition. The curve shows asymptotic limits in terms of DK and Kmax and are called the limiting thresholds, DK th and K and max;th . Similar curves and two limiting values of DK K can be also obtained at for any given FCG rate. The curve obmax tained by plotting of these two limiting values DK vs. K (paramax metrically as a function of FCG rate, da/dN) forms what we have termed a crack growth trajectory. When the DK vs. K trajecmax tory lies on a 45 line, we have the condition DK K for all max crack growth rates. The FCG rate data falling on this line implies that the fatigue damage is occurring purely by cyclic strains [6], which we refer to as pure fatigue. The crack growth process could be similar to the Lairds plastic blunting process [14]. The crack growth trajectory maps may deviate from this 45 line, when processes other than pure fatigue contribute to the crack growth. The superimposed process can be an environmentally-assisted crack growth (corrosion-fatigue) or the stress-corrosion fatigue or any other monotonic modes of crack growth, where the Kmax component contributes additionally via static load. A companion paper by the authors in this journal issue discusses various types of materials behavior that occur, based on the trajectory path [15]. We postulate that similar behavior can be expected under stress-life when rmax affects fatigue life in addition to Dr, as described below. Fig. 1b shows a data reduction scheme for the stress-life behavior parallel to that used for the crack growth analysis, Fig. 1a. Following a similar procedure to that described above, the two stress values in terms of rmax and Dr can be extracted from the data, for a given fatigue life, NF. The applied stress range, Dr, for a given fatigue life, NF, can be plotted as a function of R. The constant amplitude fatigue life data, as a function of R, in an inert environment should form the reference characterizing the material response to cyclic loads. Any deviations from that reference can be accounted in terms of additional forces that contribute to fatigue life. The additional forces could be those due to internal or residual stresses (for example, due to notch-stresses, shot peening, quenching, etc.), and environmental factors. Thus, for any given fatigue life, NF, asymptotic or limiting stress values, r and Dr can be determax mined. When NF is very large, say 107 cycles or more, the limiting values are taken as the endurance limit of the material. Note that the selection of 107 cycles to failure as the endurance limit is only for convenience. As an evolution to the conventional understand ing, we now have two critical endurance limits, r max;e and Dre , and DK for crack growth. analogous to the two thresholds, K max;th th Both limiting values, r max;e and Dre , have to be met simultaneously for fatigue damage. We cannot have Dr without rmax, while the converse is not true. Hence, a fatigue process always involves two independent loading parameters. Because of rmax, we can have superimposed monotonic modes of damage on cyclic damage, sometimes described as ratcheting or cyclic-creep under fatigue. Similarly, a trajectory path for stress-life can also be dened by plotting the relative changes in these two limiting values, with decrease in NF. Pure cyclic damage constitutes the requirement of Dr r for different NF values, which forms the 45 line on a max trajectory path. The deviations from this 45 line represent the superimposed rmax-dependent processes that include the static modes of failure and/or environmental damage. In addition, by dening a two load-parameter requirement for fatigue life, one should be able to describe the material response under variable amplitudes and changing R, similar to that for crack growth. Most importantly, we should be able to connect the safe-life approach with the damage tolerance approach using a single framework. In the stress-life approach, since NF includes both crack nucleation and crack growth, understanding the crack nucleation part is important for connecting the two stages of fatigue.

1650

K. Sadananda et al. / International Journal of Fatigue 31 (2009) 16481659

Fig. 1. Data reduction schemes for (a) fatigue crack growth and (b) stress-life fatigue.

3. Current approaches to safe-life design While it has been recognized since Goodman [12] that meanstress has an effect on fatigue life, several stress-based approaches have been proposed in the literature to quantify mean-stress effects. All of these approaches are empirically based and have the following general form:

100 90 80 70
FromWangel al. 2000

Bagci Clemson Gerber Modified Goodman Quadratic Yielding Experimental Data (Grover el al.1951)

a (ksi)

60 50 40 30 20 10 0 0
SAE 4130 Steel Su = 117 ksi Sy = 98.5 ksi Sn = 50 ksi Sn = 0.51 Sy

Dr De 1 rm =rL

where Dre is the material endurance in terms of stress range at specied number of cycles (say 106 or 107 cycles) at zero meanstress. It is a constant value for given material and environment, rm the Applied mean-stress, rL the fatigue limiting condition material yield stress, rYS, or tensile strength rUTS (i.e. design criterion shifts from fatigue to yielding or to fracture when the limiting or critical condition is reached), and n is the Exponent is 1 or 2, indicating how fast/slow the fatigue limiting condition is reached. Here, the smaller the exponent is, the faster the rate of approach, since the ratio of (rm/rL) is less than 1. Thus, Dr tends to zero when the mean-stress approaches the specied limiting value. Various models differ in terms of the denition of the fatigue limiting condition specied by rL and the value assigned to the exponent n. For example, in the Modied Goodman equation (1922) [12], rL is the ultimate tensile stress and n = 1. For Gerber [16] n = 2, and Soderberg and Sweden [17] considers rL to be yield stress with n = 1. The Soderberg model is further modied incorporating a quadratic term in terms of UTS and is called Quadratic Soderberg and Sweden [17]. Finally Bagci [18] further modies the power equation but considering the limiting condition as yield stress. Fig. 2 shows comparison of these various models [19] in relation to 4340 steel data [20]. Note that when rmean = ra or R = 0, Dr = rmax, the fatigue damage is dened as pure fatigue since corresponding Dr and rmax values fall on a 45 line. Conversely, as the mean-stress increases, monotonic modes become increasingly important. Thus, the fundamental consideration in all of the above models is the recognition that with increasing mean-stress (or increase in rmax or R) monotonic modes of failure are getting superimposed on fatigue. Subsequently, the endurance limit (cyclic damage) decreases and becomes zero when the limiting condition is reached. These models are somewhat similar to the crack growth models, where the Kmax component is introduced to account for the contribution from the limiting load to

20

40

60 m (ksi)

80

100

120

Fig. 2. The experiment data are for SAE 4130 Steel from Grover, et al 1951 [20]. Endurance limit is 106 cycles. The limiting stress, Drn is 50% of the yield stress (Drn/ry = 0.51) for this material. Ultimate stress, ru and yield stress, ry are provided. Various lines correspond to analytical approximations of the mean-stress effects from Wang et al. [19], used for design.

fatigue crack growth [21], with a correction to Kmax at high da/dN. However, this consideration of Kmax at high end of crack growth is different from Kmax as the fundamental parameter with its own threshold for crack growth, as introduced in the Unied Approach to fatigue [15]. In the case of the stress-life approach, the consideration of the superimposition of the monotonic modes of failure on fatigue (as in the above empirical models) is different from the two-parameter consideration discussed in this paper. Superimposition of monotonic modes is an extreme case where the imposed rmax induces a static mode of damage and its contribution increases as the limiting condition is approached. Sometimes this is referred to as ratcheting. In contrast, the two-parameter requirement for fatigue, in terms of Dr and rmax, is operable even when the damage is of pure cyclic nature. When R < 0, the rmax value may converge to a constant positive value. Thus we ascertain that to understand the fatigue life, one has to consider two loading parameters, rmax and Dr,

K. Sadananda et al. / International Journal of Fatigue 31 (2009) 16481659

1651

even though under a given condition one may be more controlling than the other. 4. Analysis of SN data using the two-parameter approach Using the data reduction scheme outlined in Fig. 1b, we can now begin to analyze the published SN data for various materials. The method will be illustrated by taking two examples, where extensive constant amplitude SN data are available for several Rs. Fig. 3a shows the SN data for Ti6Al4 V alloy by Peters et al. [22], plotted in terms of rmax vs. NF. The data can be reduced to Dr vs. rmax for various selected values of NF, shown in Fig. 3b. Nearly L-shaped curves are seen which dene two asymptotic or limiting stress values of Dr and r for a given NF. The deviation max from the perfect L-shape may arise at the corner, from the interaction between the two terms, Dr and rmax, due to plasticity. At high NF (108 cycles), the limiting values can be taken as the material endurance limits, Dr and r e max;e Thus, similar to the two crack growth thresholds in terms of DK and K th max;th , we have two endur, that must be satised simultaneously in ance limits, Dr and r e max;e order for the material to fail by fatigue damage. The actual values needed for a given NF follow the corresponding curve. Note that the value of r max;e is larger (420 MPa) than Dre (120 MPa). These are the limiting (minimum) endurance limits that must be met for any fatigue failure to occur, assuming the same mechanism is operating. Examination of Fig. 3b shows that it is the r that varies max signicantly (420800 MPa) with increase in NF compared to Dr (160120 MPa). The observed behavior is similar to that noted for crack growth wherein K is P DK . In addition, the results immax ply that the limiting conditions based on empirical laws depicted in Fig. 2 have not been reached under these experimental conditions. Fig. 3c depicts the variation of the two limiting values with

NF indicating that large variation occurs mainly in the r -value max than in the Dr -value. Since Dr* corresponds to cyclic strains and rmax > Drmax , we assume that the monotonic deformation helps the fatigue damage by building up the required internal stresses to set up the condition for crack nucleation and growth. How the internal stresses play the role in fatigue will be discussed later. The fatigue life trajectory map can be drawn using these two limiting values and plotting Dr* vs. Dr . This is shown in max Fig. 3d. This is similar to the trend shown in a trajectory map for cracks growing in many of the Ti-alloys [23], as shown in Fig. 4. In the case of crack growth, with increase in crack growth rate (also implies increasing stress intensity factors), the curve runs parallel to the Kmax axis indicating the crack growth is increasingly Kmax dependent. Fig. 3d shows the behavior is somewhat similar to the extent that with decrease in life, the fatigue life is increasingly determined by the maximum stress. That is, we move from the high-cycle fatigue conditions where cyclic strains are dominant to the low-cycle fatigue conditions, where the internal stresses generated by dislocation substructure becomes a dominant factor in generating the necessary conditions of crack nucleation and growth. With increasing rmax, the tensile cyclic-creep strains could increase due to an unrestricted specimen elongation under the highly non-symmetric cyclic stresses. In addition, Feltner and Laird [24] have shown that the dislocation substructure formed in lowcycle fatigue conditions is similar to that under the monotonic deformation. In the next example, we examine the fatigue behavior of a low alloy steel in air and corrosive environments [25]. Fig. 5a shows the SN curve of a low alloy steel in ambient air. In reducing the data, following the steps outlined in Fig. 1b, it was found that there may be two distinct mechanisms operating, one at low R and one at

Fig. 3. (a) SN data for Ti6Al4V alloy plotted in terms of rmax vs. number of cycles to failure for various constant R-ratios. Data are from Peters et al. [22]. (b) Typical Lshaped curves for each NF dening two limiting endurance values. The L-shape gets distorted into smooth curve-behavior due to second order interaction effects between the two-parameters. The two stresses then have to be along the curve to enforce the same NF. (c) Variation of the limiting values as a function of no. of cycles to failure. (d) Fatigue life trajectory map showing the variation of the two limiting endurance values.

1652

K. Sadananda et al. / International Journal of Fatigue 31 (2009) 16481659

Fig. 4. Trajectory map for crack growth in many Ti-alloys from Ref. [23] and references therein.

high R. Such changes in mechanisms as a function of R have been noted before, where the crack growth changes from predominately intergranular at low R to predominately transgranular at high R [26]. Fig. 5b shows how the two mechanisms can be differentiated on the basis of the Dr vs. R curves. The differences are small, nevertheless consistent. Using the interpolated and extrapolated data based on the Dr vs. R curves, it is possible to generate the Dr vs.

rmax curves for each mechanism, as shown in Fig. 5c. Finally, Fig. 5d depicts the trajectory path for fatigue life for the alloy in three different environments; air, free-corroding potential and a superimposed potential. All of the data were obtained from the same Ref. [25]. In the high-cycle fatigue regime, the data follow closely the 45 line indicating that the fatigue life, close to endurance, is determined by cyclic damage. The 45 line means that the Dr $ r e max;e , within the experimental scatter. For the same NF value, there is reduction in the Dr and r e max;e values between air and the environment. That is, the environment is reducing the cyclic stresses needed to cause initiation and failure. Since crack initiation is the major part of life under high-cycle fatigue, the environment must be inuencing the crack nucleation by a reduction in the surface energy. Similarly, as we move towards the low-cycle fatigue regime, the trajectory deviates in the direction of the rmax axis, peaks and then drops down towards the r axis, indicating max that the monotonic modes are becoming dominant at the low-cycle-fatigue end. Detailed discussion of the various types of mechanisms governing the trajectory paths are outlined for the case of crack growth in a companion paper [15]. Similar behavior is expected for the stress-life or the SN behavior. Finally, we show an example of an Al-7075-T6 alloy where the trajectory path for the SN data of a notched specimen with Kt = 5 is plotted along with the trajectory for crack growth, Fig. 6. Data are collected from [27] (MIL-HDBK-5). Because of high Kt, the S N life is more dominated by the crack growth process. Qualitatively, the trajectory paths for both crack growth and the SN fatigue life of a notched specimen follow a single line even though the scales for the two differ and no special effort was made to match

Fig. 5. (a) SN curves for a low alloy steel tested in ambient air. The points are not experimental but a digitized representation of the original data curves in Jones and Blackie [25]. (b) Data plotted for selected NF values as a function of R. Possible change in the mechanism from Mech.1 to Mech.2 with R. Additional data were extracted by interpolation. (c) L-shaped curves for mechanism I using interpolated data along with experimental data. (d) SN trajectory paths for the low alloy steel under three different environmental conditions.

K. Sadananda et al. / International Journal of Fatigue 31 (2009) 16481659

1653

Fig. 6. The trajectory paths for crack growth and SN for notch life data with Kt = 5. Note the striking resemblance of the two even though the continuity of the line drawn is somewhat fortuitous since the scales for the two are different. Data are from MIL-HDBK-5 [27].

the scales. The striking similarity in the SN life and crack growth life is obvious from the plot. Implication is that it is possible to relate crack growth analysis to the behaviors in the notched and the smooth specimens. Approaching in reverse, it is possible to move from crack nucleation to crack growth using a proper consistent analytical tool for characterizing the damage evolution. 5. Relating crack nucleation to crack propagation In 1976 Kitagawa and Takahashi [28] provided an important link connecting the endurance limit Dre of a smooth specimen to the crack growth thresholds Drth in a fracture-mechanics specimen. No physical explanation was suggested. The rst interpretation of Kitagawa diagram came from El-Haddad et al. [29] who added an empirical crack length,

a0

 2 1 DK th p Dr e

to the actual crack, a, in order to make a smooth transition from sloped Drth line to the horizontal Dre line in the original Kitagawa diagram (not shown here). This smooth line is given by the following relationship

DK th F Drth

p pa a0

where Drth is the far-eld stress range and F is the geometry correction factor. For F = 1 and assuming that a is very small, the above equation reduces to Drth = Dre. On the other hand, when a much larger than a0, it reduces to

DK th F Drth

p pa

In the present work, it is proposed that the smooth transition from the sloped rmax,th line (designated as the Kmax,th line) to the horizontal rmax,e line in the Kitagawa diagram (Fig. 7) is due to internal stresses generated by plasticity. These fatigue generated internal stresses alter the applied threshold stress intensity factor according to the following relation depending on the tensile or compressive internal stresses.

K total K appl K internal

Throughout the analysis we assume that the small scale yielding conditions prevail at the crack tip. For a crack emanating from an

elasticplastic notch, the effect of the notch plasticity on the stress intensity factor needs to be included. That is, in the evaluation of Kinternal, plasticity corrections have to be incorporated. How this is done will be described later. Here, it is important to note that the K max;th is an intrinsic material threshold for a given material/environment system and is independent of crack size; long or small. Fig. 7 shows a modied Kitagawa diagram. Normally, log of nominal stress range, Dr, is plotted by Kitagawa and Takahashi [28] against the log of crack length. Since we have dened that there are two endurances, Dr and r e max;e , and for all materials r P Dr , and K P DK , we propose a modied Kitagawa max;th th max;e e diagram, in terms of the log of nominal maximum stress, rmax,nom, vs. the log of crack length with the trend line for rmax,th = Kmax,th/ F(pa)0.5 (instead of Drth and DKth) for a fracture-mechanics specimen. For convenience we call this line as the Kmax,th-line within the spirit of the original Kitagawa diagram. We now add two other limiting conditions, the true tensile failure stress rF, and the critical fracture line rmax,cr = KIC/F(pa)0.5, which will be referred to as the KIC-line. Under fatigue conditions, rF, represents the limiting stress (similar to rL in Fig. 2) where the fatigue failure occurs in one half cycle. The region between the KIC-line and the Kmax,th-line is the fatigue crack growth region. For the case when the applied maximum stress, rmax, falls below rmax,e and the Kmax,th line, the growing crack becomes arrested, as often happens under sufciently high overloads or spike loads, as well as during propagating of a short cracks at low loads. For example, the compressive internal stresses that form at the crack tip can bring the total K (due to applied and residual/internal stresses) at the crack tip below the Kmax,th. Similarly, tensile residual stresses can augment the applied stress which will result in an increase in the total K, say, by applying an underload. Since the crack growth threshold [15] does not vary with crack length, a growing crack can get arrested if the total stress intensity factor falls below the thresholds. Hence, the entire Kmax,th-line represents the threshold crack growth boundary for all crack lengths. This threshold condition is valid for any given R. In Fig. 7, the regime bounded by rmax,e (below the endurance stress value) and to the left of the Kmax,th-line is designated as the nonpropagation regime where crack lengths in that regime cannot grow to failure. A fundamental question that needs to be claried in understanding the Kitagawa diagram is how a smooth fatigue specimen that has no noticeable crack at rmax,e will end up with a crack size of ac, since the crack can only grow at the nominal stress rmax,e after reaching the ac value; that is, when the crack growth threshold condition is met (where ac = (Kmax,th/Frmax,e)2/p). The assumption that a short crack of length less than ac would have a lower threshold (as is often assumed in the literature) would not address this issue, since there is an increasing threshold with increasing crack length that has be satised without increasing the applied loads. Addition of an arbitrary crack length, ac, as was done by El-Haddad et al. [29], requires a physical justication. Let us now examine the region bounded by the stresses above rmax,e and to the left of the Kmax,th-line. We label the region as the internal stress build up for propagation. In principle, for any stress above the endurance limit, failure should eventually occur, at some number of cycles less than the endurance value. We have shown that for any given number of cycles to failure, NF, there are two limiting values of r , and Dr (see Fig. 3b). Conversely max for any given rmax and Dr above the endurance limits, even the smooth specimen will eventually fail at some NF value. For failure in a smooth specimen, crack formation occurs at an in-situ generated stress concentration site due to heterogeneity in the deformation. The physics of the damage process indicates that some grains at the surface region of the sample are always more favorably oriented than the interior grains to initiate slip and protrusions. These heterogeneities lead to localized internal stresses due to strain gra-

1654

K. Sadananda et al. / International Journal of Fatigue 31 (2009) 16481659

Fig. 7. Interpretation of modied Kitagawa diagram with three regions of crack behavior.

dients. For the purpose of illustration, let us consider rmax, a stress above the endurance stress rmax,e. Without loss of generality, let us assume that a smooth specimen has an arbitrary incipient crack or defect of size <ac (shown as point A in Fig. 7) and is fatigued at this maximum stress. Such a specimen should fail at a number of cycles smaller than the endurance life, say less than 107 cycles. A pre-existing incipient crack at point A cannot grow to reach point C in Fig. 7, where only the crack propagation conditions are met for the specimen to fail by crack growth. However, experiments show that the specimen does fail at some number of cycles by localized crack nucleation and growth processes. In fact, many cracks may form, but most of them remain non-propagating since they do not meet the propagation-threshold requirement. Usually one main crack that meets the threshold Kmax,th condition can grow to meet the KIC-line. This is a general experimental observation. Essentially, in our example, the crack has to move from A to C (Fig. 7) to sustain continuous crack growth at the nominal stress rmax. However, it cannot grow, since for all crack lengths less than C, the applied stress intensity factor is less than the threshold Kmax,th. The only way it can grow from A to C is to go from A to B (increase in stress at constant crack length) where it can meet the threshold criterion for crack growth, and then move down to C (as in a constant Kmax test), and then proceed along CD (constant nominal stress) to failure. Looking from the nominal stress level, one way to accomplish this is to increase applied stress from A to B and then decrease along the line BC to rmax, and keep it constant along CD. Experiments, however, show that without increasing the remote applied stress, the specimen eventually breaks, since rmax is greater than the endurance value. To understand the physics of the localized crack growth process, we bring in the internal stress concept to account the growth of the crack without altering the nominal stresses. Internal stresses are physically realistic and arise due to deformation gradients from dislocation pileups, slip bands, intrusions and extrusions, etc. These internal stresses are still of long range of the order of crack size or plastic zones sizes that are formed. These provide an additional local crack tip driving force and act in a similar way as nominal stresses do. Hence the Kitagawa diagram, viewed from the point of a crack, involves build up internal stresses via localized plasticity to augment the applied nominal stresses to initiate and grow the crack. These

stresses are localized and extend to the range of the order of crack lengths. The stress intensity factors for the uctuating internal stresses can be computed using, for example, the weight function methods [30] or numerical techniques [31]. Thus, from the physical considerations, internal stresses are generated by cyclic plastic strains to meet both the magnitude (equivalent to B from the point of nominal stress) and the minimum gradient corresponding to BC to insure that the crack grows continuously along the Kmax,th-line without arrest until it becomes self-sustaining by the nominal stress rmax. A similar explanation is valid for any pre-existent crack of size less than ac for applied stresses greater than or equal to the endurance limit, rmax,e. The criteria for crack growth will be that the magnitude and gradient of internal stresses should be equivalent to applying a nominal stress B and reducing to C as the crack grows. Analysis by Argon and Chang [32] indicates that such highly localized internal stresses can be generated at dislocations pile-ups to facilitate the formation of cracks and voids. Without these high local internal stresses cracks cannot physically initiate or grow at stresses less than rF. Thus, in switching from nominal to local stresses from the point of crack growth, we are trying to be physically realistic without violating any constitutive laws. Here, the Kitagawa diagram provides a clue of how much localized deformation gradients should be generated that can contribute to failures at all stresses above the endurance limit, expressed from the point of nominal stresses. In the same way, for any stress below the endurance limit, the incipient crack smaller than ac cannot grow, since the applied stresses are not sufcient to generate the requisite cyclic plastic strains and hence the corresponding internal stresses to drive the crack along the line BC. Thus, all cracks that are below the endurance limit rmax,e and the Kmax,th-line remain as non-propagating cracks. Perhaps one can draw another curve below the endurance limit rmax,e where the remote and internal stresses are high enough to initiate a crack but not high enough to grow, resulting in an initiated but non-propagating cracks. We will show in the next section, that this crack initiation line corresponds to (rmax,e/Kt), where Kt is the stress concentration factor. These cracks that are initiated do not grow to failure until the Kmax,th condition is met by the remote stress. Unless this condition is met, the internal stresses and their gradients generated from plasticity are not suf-

K. Sadananda et al. / International Journal of Fatigue 31 (2009) 16481659

1655

cient enough to sustain the continuous growth of the initiated crack. The fatigue endurance limits are therefore the stresses in terms of Dre and rmax,e that are needed to build up required internal stresses via localized plasticity to initiate and grow the cracks under the loading conditions. Below these limits, local plasticity and thus the internal stresses can be generated and they may be sufcient to initiate but not to propagate. The stresses in some cases may not be sufcient even to initiate the cracks. Shot peening, for example, can suppress the generation of these required internal stresses [33]. Likewise, periodic electro-polishing the fatigue specimen can remove any internal stress build up due to dislocations thereby extending or rejuvenating the fatigue life [34]. Hence understanding the role of internal stresses is very important in the fatigue life prediction. Here we are providing the physical meaning for the Kitagawa diagram using the Internal Stress Concept. This will eventually help us to develop criteria for crack initiation and its growth to incorporate in the UNIGROW fatigue life prediction model [9]. 6. Analysis of the stress gradient required for crack growth Experimentally, it is well known that the acceleration and deceleration of crack growth can occur by underloads and overloads. This has been accounted for by the excess internal stresses generated by localized plasticity that can cause increase or decrease of the stress intensity at the crack tip. The same concept should be applicable for acceleration and deceleration of short cracks. Hence, the above analysis shows that total stress intensity factor from applied and internal stresses will determine if an incipient crack will grow or not to cause failure. For short cracks, which are nucleated at some stress concentrations, plasticity at the localized stress concentration provides the necessary internal stresses that augment the applied stresses to meet the crack growth condition. In addition, the total stresses (applied and internal) must satisfy the requirement of a minimum stress gradient condition in order to sustain continuous crack growth. For example, in Fig. 7, the internal stresses not only have to move the point A to the point B but their gradient also should be sufcient to move the crack from point B to the point C. Thus the Kitagawa diagram provides the minimum requirements for both magnitude and gradient in internal stresses to sustain continuous crack growth. Observed crack arrests and non-propagating cracks during short crack growth or spike overloads are the result of not meeting simultaneously the above minima criteria. Since smooth specimens and fracture-mechanics specimens form two extremities that are connected via the Kitagawa diagram, to understand the complete physical signicance of this diagram, we will consider a

notched specimen with different values of stress concentrations, Kt. In the limit of Kt = 1, we arrive at a smooth specimen and with increase in Kt we converge to the behavior of a cracked specimen. Fig. 8 is redrawn from a 1957 classical work [35,36] on the fatigue of notched specimens with varying Kt Fig. 8a shows the constant stress range Dr required to cause a crack nucleation and failure as a function of Kt of a notch. The triangles denote the minimum stress required to initiate a crack while the circles denote the minimum stress required to cause failure. For Kt = 1, that is for smooth specimen, the endurance limit is around 260 MPa. With increase in Kt, the endurance drops rapidly. However, with increase in Kt > 3, the curve bifurcates; the minimum stress for crack nucleation decreases gradually and levels off while the notch endurance limit based on fracture becomes independent of Kt. It is believed that the bifurcation point corresponds to the critical gradient of the stresses ahead of the notch. The lower curve can be estimated [37] as Dr = Dre/Kt which means that the stress range at the notch root, Drnotch, is equal to the smooth specimen endurance limit, Dre. Below this curve, i.e. Drnotch = Kt Dr < Dre, no crack initiation occurs. Thus, the crack initiation is governed by the local notch root stress equal to that of a smooth specimen endurance limit. Above the curve the cracks are formed but they do not grow continuously, giving rise to non-propagating cracks. Only above the stress denoted by the endurance line, cracks that were nucleated at very low loads can grow until failure. These data are familiar and many interpretations have been provided starting from El-Haddad, Topper and Smith [29]. In Fig. 8b, we examine the maximum stress at the notch tip by taking the product (Kt Dr). Interestingly, the minimum stresses that are needed for crack initiation now fall on a constant line along with the endurance limit of a smooth specimen. On the other hand, the constant line representing endurance for a notched specimen in Fig. 8a forms a steep line. Failure occurs only when the local stress at the notch tip exceeds this value, for any Kt. Some key points are to be noted. Comparison of Fig. 8a and b indicates that the crack initiation occurs at a notch tip, at the same maximum stress as the nominal endurance stress of a smooth specimen. The implications are that the local maximum stress controls nucleation and it occurs at the same local stress level irrespective of the notch severity. Thus an initial growth of an incipient crack to short crack, traditionally called crack initiation, is independent of any stress concentration, and is governed perhaps by the dislocation density and distribution, which are functions of the local stresses and material ow properties. As Kt increases, the notch essentially mimics a crack (notch tip radius ( notch depth), yet the crack initiation is essentially similar to that of smooth specimen and occurs at the same value of the lo-

Fig. 8. Effect of stress concentration: (a) nominal stress amplitude vs. Kt. No failure below an endurance limit; (b) magnied stress (Kt times Dr) and propagating conditions at the notch tip. Note local magnied stress (internal stress) has to be signicantly high for propagation, in relation to nominal stress amplitude for nucleation.

1656

K. Sadananda et al. / International Journal of Fatigue 31 (2009) 16481659

cal stress range. However, the initiated crack may or may not grow. For it to grow, higher local stresses are needed. In fact Fig. 8a shows the endurance limit decreases by an amount Kt until Kt % 3 and then remains constant with further increase in Kt. Thus, the endurance limit for a smooth specimen is around 260 MPa, while the constant value in Fig. 8a is around 90 MPa, about 1/3 of 260 MPa. For the initiated crack to propagate, higher local stresses than Dre (which must also satisfy the minimum gradient requirement) are needed, as can be seen in Fig. 8b. The requirements for propagation are governed by the crack growth thresholds in terms of DKth and Kmax,th, which must be met simultaneously. Note that the estimated elastic local stress at the notch tip can be very high of the order of 1400 MPa even when the nominal stress is only 90 MPa. As a result, local yielding will occur, which in turn, will reduce the local peak stress and stress gradient but increase the local peak strain and strain gradient. From Fig. 8a and b, it is clear that the incipient crack nucleation energetics may be different from the kinetics of growth. Nucleation is fully governed by the local stress level alone, whereas propagation is by both the stress level and its gradient. Whether nucleation or propagation controls the fatigue life of a specimen depends on which of the two is an easier process for a given condition. If there are already pre-existent stress concentrations, as in the case of notches, the local stress can reach its required value for crack nucleation early in life; hence for those cases propagation will be the life-limiting factor. Conversely, where the local stresses have to reach their maximum by localized plasticity as in smooth specimen, then crack nucleation can be a large part of the fatigue life. From the above analysis of crack propagation of an incipient crack that is nucleated at a stress concentration, it is clear that a simple elastic analysis of the notch tip stress elds is inadequate, and we need to resort to elasticplastic analysis. For the purpose of illustration, we consider below two cases; (a) specimen with root radius of q = 3 mm, but increasing depth starting from Kt = 3, and (b) specimen with Kt = 3, but with changing root radius, q. We use a simplied elasticplastic analysis to determine K of a crack growing in the plastic strain eld of a notch. The simplied expressions are deduced recently by Kujawski [38] using Neubers rule and considering the RombergOsgood stressstrain relation for the material [13] Eq. (6) was used to estimate the stress intensity factor for cracks emanating from an elasticplastic notch

r K lim k s
q!0

pq
4

In the above equation q = qnotch + a, where q is the notch tip radius, a is the crack length, ke corresponds to an elasticplastic strain concentration factor at the distance x = a from the notch tip. The values of ke can be calculated numerically using FEA software or can be estimated utilizing the well-known Neubers rule. The Kplastic for a crack growing in the plastic eld of a notch is calculated using the above expression. Fig. 9a and b show the

two cases. In Fig. 9a, the stress intensity factor, K, for an incipient crack growing in the plastic strain eld of a notch are plotted as a function of crack length. The K values are normalized with respect to remote stress and plotted as a function of normalized crack length, for increasing Kt, but with a constant root radius, q = 3 mm. The initial sharp increases of K in Fig. 9a and b are in the process zones involving the incipient crack formation, and the analyses are beyond the scope of the continuum mechanics. After the formation of an incipient crack, the K for that crack decreases initially and then gradually increases with the crack length. Thus there is a minimum in the curve at some intermediate crack lengths. If the material has a long crack threshold (Kmax,th), then the situation can arise with K at the minimum falling below the Kmax,th threshold for crack growth. Then, the growing crack arrests when K < Kmax,th. At higher Kt, due to increased plasticity, the minima can be higher than the threshold allowing the crack to grow continuously. The crack growth rate can decrease as K decreases, reach a minimum, then increase as K increases with crack length. Such a deceleration and acceleration has been observed [10,11] during both short crack growth as well as crack growth after overloads. Hence the elasticplastic conditions at the notch tip contribute to the crack arrest phenomenon, if the K for the incipient crack falls below the threshold for crack propagation. In Fig. 9b, a case is illustrated to account for the experimental data where cracks formed at smaller holes do not grow while those formed at larger holes can grow to cause failure. In this plot, the Kt is xed at 3 and root radius q is increased. For the same Kmax,th used in Fig. 9a, we nd that non-propagating conditions are set for q = 3.5 or less but for larger hole sizes the minima are above the threshold insuring continuous crack growth without any arrest. Fig. 10 shows the experimental data of Murakami and Endo [39] showing that for holes less than some critical size there are no changes in the endurance limit, while holes larger than some critical size only contribute to lower fatigue limit. Based on Fig. 8a, it appears that the mechanics of crack growth is not changed signicantly by the presence of a notch for all Kt > 3; the nominal stress for crack propagation remains the same. In all cases the stress gradient is lower than the critical amount needed for crack growth. That is, the number of cycles required to generate the necessary internal stresses and their gradients will decrease with increase in Kt. In addition to nominal stress, local internal/ residual stresses can also affect the stress intensity factor, K. The crack growth requirement, therefore, is that the total K as a crack tip driving force has to exceed the threshold K max;th . When there is no crack to start with, an incipient crack has to form by a nucleation process. Once crack has formed the subsequent growth is determined by the threshold condition for crack growth. Nucleation of incipient crack followed by its growth is the determining factors for the total fatigue life. Criteria for the two are different, as shown by Fig. 8a and b. In particular, initiation is independent of stress gradient whereas propagation is. This is true for a smooth

Fig. 9. (a) Elasticplastic calculations for conditions: (a) at constant q = 3 mm and Kt varying, and (b) Kt = 3 constant with q varying.

K. Sadananda et al. / International Journal of Fatigue 31 (2009) 16481659

1657

Fig. 10. Relation between fatigue strength and the variation in hole diameter for 0.13% carbon steel [39].

specimen, a notched specimen or a cracked specimen. Only difference is the nucleation is easier at a notch since the critical local stress for nucleation can be easily met at low nominal stress amplitudes, while it is difcult for a smooth specimen since conditions of concentrated local stresses have to be established by local plasticity. That is, the number of cycles required to generate the same internal stresses for nucleation will be less with increasing Kt. We have to bear in mind that the Kitagawa diagram connects the two extreme cases involving nucleation in a smooth specimen and a crack growth from a pre-existing crack/defect. For all notches, the local notch root stresses have to be enhanced to meet the initiation criterion and the stress gradients have to meet the propagation criterion. Since the nucleation stress is the same as the endurance stress of a smooth specimen we can look closely the initial stages of crack formation and its kinetics of growth. Fig. 11 shows the simplied Kitagawa diagram (with rmax,e and Dre) and the process that can lead to crack formation. Since Kmax,th has to be met for all crack lengths, line AB denotes the decrease in stress with increase in crack length that satises the critical gradient requirements along

the path. From the point of stress, it denotes internal stresses that provide the same K as the remote nominal stresses. Hence, for any stress above the endurance limit, crack should form and grow to failure by generating sufcient internal stresses by the cyclic plasticity to move the crack along the path AB. For engineering materials which are generally polycrystalline, the deformation is inhomogeneous due to favorably oriented surface grains, causing gradients in the internal stresses. As intrusions and extrusions [40] form due to irreversibility of the slip process (in Fig. 11, see inserts), they behave like notches with stress concentrations. Additional localized slip can lead to further build up of internal stresses and modify their gradient. A stage is reached when a crack could form and begin to grow. Thus the localized plasticity sets up the total equivalent stress and its gradient that meet the minimum criteria in terms of nominal stresses given by the Kitagawa diagram, Fig. 11. Both nucleation and growth occur by building up the internal stresses via the formation of a suitable dislocation structure. The details of the micro mechanics of the process, particularly the incremental dislocation density in each cycle and how they build up the necessary and sufcient internal stresses and their gradients are unknown. Fig. 12 shows the labeled Region of Internal Stress Build up and the possible variations in the internal stress gradients that can be set up by an appropriate dislocation density and its distribution. In Case 1, the internal stresses and their gradients are more than those needed for the formation of a crack and its growth. In this case, failure is insured. For Case 2, the internal stresses are initially higher than the minimum but drop rapidly below the minimum required leading to crack arrest. The situations are similar to Fig. 9a and b where a growing crack can get arrested when K due to stresses falls below Kmax,th. For Case 3, the internal stresses are well below the minimum requirement for crack growth, but above the endurance limit of a smooth specimen. Hence initiation is insured but not propagation. Propagation ultimately can occur only after building additional internal stresses by cyclic plasticity, if the nominal stresses are at or above the endurance limits. Thus internal stresses and their gradient can play an important role both during initiation and growth stages. Thus, the crack nucleation is governed by the local maximum stress at the surface of a smooth sample or at the notch tip. The crack propagation is governed by stress intensity parameter, where

Fig. 11. Intrusions and extrusion formation provide internal stresses and gradients to nucleate cracks (Illustrations from Witmer et al. [40]).

1658

K. Sadananda et al. / International Journal of Fatigue 31 (2009) 16481659

stresses via the dislocation processes. Both slip and its degree of reversibility determines the rate of accumulation. Grain size and local microstructure will have a strong bearing on the process of internal stress build up via dislocations. All these aspects are embedded in the determining the endurance limit as well as the criteria for the growth of the incipient crack formed, as described in the modied Kitagawa diagram (Fig. 7). Fatigue life prediction in terms of crack nucleation and growth therefore depends closely in the rate of accumulation of internal stresses and their gradients. These are related at micro level to dislocation density and their gradients and at continuum level to localized strains and their gradients.

8. Summary In the above analysis, we have analyzed several aspects that are involved in the crack nucleation and growth and thus the total fatigue life. First, we have shown that there are several common factors between crack nucleation and crack propagation as well as some divergent factors. Common factors include the two-parameter requirement of fatigue damage which manifests as Kmax and DK for crack growth and rmax and Dr for SN life. Just as there are two thresholds, K max;th and DK th , for crack growth, we have two endur ance limits for the SN fatigue, r max;e and Dre . In all materials, P DK , and similarly r P Dr . Most life prediction K max;th th max;e e methodologies, including the simple Miners rule, ignore the twoparameter nature of fatigue, hence are empirical. We believe that consideration of these two-parameters in each cycle would help in better prediction of fatigue life, as has been done in the UNIGROW model [9] where the role of both Kmax and DK are included. Finally, the trajectory maps for both crack growth as well as for fatigue life can be developed that give details of the changing mechanisms as a function of crack growth or the SN life. The pure fatigue behavior where the damage is governed by only cyclic strains can be seen under the high-cycle fatigue conditions where r Dr , somewhat similar to the crack growth condition where max under pure fatigue crack growth occurs when K DK . The max deviations from pure cyclic strain controlled process occur under both crack propagation and the SN life. In the SN life, monotonic modes become dominant at the low-cycle fatigue regime, while the cyclic strain controlled process dominates in the high-cycle fatigue regime. The SN life includes both crack nucleation and crack growth. In the high-cycle fatigue region, the nucleation life may be major part of the life, while in the low-cycle fatigue region the crack propagation is dominant. The analysis also shows that in some respects the mechanics of crack nucleation is different from crack propagation. The uncertainties and ambiguities in dening when the nucleation ends and propagation begins are not related to the mechanics of the process but to the limitations in the detection of a nucleated crack. The differences between the two get magnied when we have conditions where the crack nucleation is possible without their growth, resulting in non-propagating cracks. Nucleation is governed by the maximum local stress and not maximum nominal stress. This maximum local stress is the same as the stress required to nucleate a crack in a smooth specimen. On the other hand, the propagation is governed by stress intensity factor which has to meet the propagation threshold, K max;th . In the UNIGROW model [9] the total crack tip driving force Dj was expressed in terms of the two-parameters DK and Kmax in a form that could collapse all the crack growth data into a single curve. Similarly it would be convenient to express the SN life in terms of a single parameter that incorporates both rmax and Dr. Then, one can develop a consistent fatigue model that can be used for spectrum loads. Efforts in that direction are currently being

Fig. 12. Schematic illustration showing three possible internal stress proles indicating the required minimum for steady growth of an incipient fatigue crack. This internal stress triangle is above the endurance line in Fig. 7.

we switch from the local maximum stress to the stress intensity factor K of the nucleated crack. The value of K depends not only on the local internal stress but also on its distribution along the length of the crack. Thus there appears to be change in the mechanics of damage from a local peak stress during crack nucleation to stress intensity factor that is somewhat global during crack propagation. Fig. 8a and b depict this change. However, this has to be understood in terms of the local plasticity affecting the local strains and strain gradients which in turn affecting the K of a crack through strain-energy function. Thus we do have local strain strain behavior at the notch or crack tip affecting the Kplastic, and thus the kinetics of crack growth. This switch from the elastic to the elasticplastic constitutive behavior is unavoidable in fatigue, since fatigue is a plasticity-induced damage, while we are characterizing it on the basis of an extended linear elastic fracturemechanics, assuming that all the plasticity is localized or it corresponds to small scale yielding conditions.

7. Crack nucleation in a smooth fatigue specimen: key issues It is well known that for a polycrystalline annealed material, the crack nucleation occurs at a surface of a grain that is favorably oriented to the deformation slip. Localized deformation leads to a formation of intrusions and extrusions (or protrusions), which help to build up the local internal stresses. Incremental deformation in each cycle along the slip planes sets up the dislocation dipole arrays inside the grain forming protrusions at the surface, thereby augmenting the local stresses or more specically the internal energy of the system. Crack formation is ensured if there is sufcient energy to nucleate a crack; similar to Grifths fracture condition. Since the dislocations arrays are formed on the slip planes, crack can form along the slip plane if there is a reduction in the total energy when the dislocated material is replaced by a crack. The nucleation kinetics can be somewhat similar to that proposed by Mura [41]. The cleavage planes could become favorable if environment can lower the surface energy and thus reduce the energy of the crack formation. Excluding those special cases, the crack nucleation occurs generally along the slip plane forming Stage I. As the crack grows, it changes to the Stage II crack growth when the growth condition is met, i.e. the K at the crack tip exceeds the Kmax,th. The number of cycles required to nucleate a crack depends on the number of cycles needed to build up the necessary local

K. Sadananda et al. / International Journal of Fatigue 31 (2009) 16481659

1659

pursued. By separating the nucleation life from the propagation life in the SN fatigue, one can go from the nucleation model to the propagation model incorporating rmax and Dr for nucleation and Kmax and DK for propagation. These aspects are brought together using a modied Kitagawa diagram for which a physical interpretation has been provided. References
[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] Vasudevan AK, Sadananda K, Louat N. Mater Sci Eng 1994;A188:122. Sadananda K, Vasudevan AK. Int J Fatigue 2003;25:899914. Vasudevan AK, Sadananda K. Metall Trans 1995;26A:122134. Sadananda K, Vasudevan AK. Int J Fatigue 1997;19:S99S109. Vasudevan AK, Sadananda K. Int J Fatigue 1999;21:S26374. Sadananda K, Vasudevan AK, Holtz RL. Int J Fatigue 2001;23:S27786. Kujawski D. Int J Fatigue 2003;25:793800. Dinda S, Kujawski D. Eng Fract Mech 2004;71(12):177990. Mikheevskiy SM, Glinka G. Int J Fatigue, in press, 2009. Suresh S, Ritchie R. Propagation of short fatigue cracks. Int Met Review 1984;29:44576. Sadananda K, Vasudevan AK. Short crack growth behavior. In: Piascik RS, editor. Fatigue and fracture mechanics, vol. 27. ASTM-STP 1296, West Conshohocken (PA): American Society for Materials and Testing; 1997. p. 30116. Goodman J. Mechanics applied to engineering. London: Longmans Green; 1899. Dowling NE. Mechanical behavior of materials: engineering methods for deformation fracture and fatigue. 2nd ed. Upper Saddle River (NJ): Prentice Hall; 1999. Laird C, Smith GC. Philos Manage 1962;8:84757. Vasudevan AK, Sadananda K. Int J Fatigue, in press, 2009. Gerber HZ. Bayerischen Architeckten Ingenieur-Verins 1874;6:10110. Soderberg CR, Sweden V. ASME Transactions AER-1S 1930;52(1):115. Bagci C. Mech Mach Theory 1981;16:33959.

[12] [13]

[14] [15] [16] [17] [18]

[19] Wang SJ, Dixon MW, Huey CO, Chen SC. J Mech Des 2000;122:1436. [20] Grover HJ, Bishop SM, Jackson LR. Axial load fatigue tests of unnotched sheet specimens of 24S-T3 and 75S-T6 aluminum alloys and SAE 4130 steels. National Advisory Committee for Aeronautics Technical Note 2324; 1951. [21] Forman RG, Kearney VE, Engle RM. J Basic Eng 1987;89:45964. [22] Peters JO, Boyce BL, Chen X, McNaney JM, Hutchinson JW, Ritchie RO. Eng Fract Mech 2002;69:142546. [23] Sadananda K, Vasudevan AK. Int J Fatigue 2005;27:125566. [24] Feltner CE, Laird C. Acta Metall 1967;15:162133. [25] Jones WJD, Blackie AP. Int J Fatigue 1989;11:41722. [26] Sadananda K, Vasudevan AK. Fatigue Fract Mater Struct 2003;26:83545. [27] Metallic Materials and Elements for Aerospace Vehicle Structures (MIL-HDBK5J), US Department of Defense, Section 3.7.6; 2003. [28] Kitagawa H, Takahashi S. Applicability of fracture mechanics to very small cracks or the cracks in the early stage. In: Proceedings of the second international conference on mechanical behavior of materials. Metals Park (OH): ASM; 1976. p. 62731. [29] El-Haddad MH, Topper TH, Smith KN. Eng Fract Mech 1979;11:57384. [30] Shen G, Glinka G. Determination of weight functions from reference stress intensity factors. Theor Appl Fract Mech 1991;15:23745. [31] Wu Z, Glinka G, Jakubczak H, Nilsson L. Calculation of stress intensity factors for cracks in structural and mechanical components subjected to complex stress elds. J ASTM 2004;1(9). [32] Chen I-W, Argon AS. Acta Metall 1981;29:1321. [33] Dieter GE. Mechanical metallurgy. New York: McGraw Hill; 1986. p. 409. [34] Kramer IR. Mechanism of fatigue failure. Met Trans 1974;5:173542. [35] Frost NE, Dugdale DS. J Mech Phys Solids 1957;5:182. [36] Frost NE, Marsh KJ, Pook LP. Metal fatigue. Oxford: Clarendon Press; 1974 [Chap. 4]. [37] Peterson RE. In: Sines G, Waisman JL, editors. Metal fatigue. NY: Mc Graw Hill; 1959. p. 293306. [38] Kujawski D. Eng Fract Mech 2008;75(18):52467. [39] Murakami Y, Endo M. Eng Fract Mech 1983;17:1. [40] Witmer DE, Farrington GC, Laird C. Acta Metall 1987;35:1895909. [41] Mura T. Mater Sci Eng 1994;A176:6170.

You might also like