Vous êtes sur la page 1sur 12

Acta mater. Vol. 46, No. 14, pp.

4861±4872, 1998
# 1998 Acta Metallurgica Inc.
Published by Elsevier Science Ltd. All rights reserved
Printed in Great Britain
PII: S1359-6454(98)00194-3 1359-6454/98 $19.00 + 0.00

VOID NUCLEATION AND CRACKING AT GRAIN


BOUNDARIES
P. SHEWMON{ and P. ANDERSON
Department of Material Science and Engineering, Ohio State University, Columbus, OH 43210, USA

(Received 26 November 1997; accepted 27 May 1998)

AbstractÐDimpled grain boundary fracture occurs in creep tests, stress relief cracking and hydrogen attack
of steels. While the growth of these voids by grain boundary di€usion is well established, the mode of void
nucleation is uncertain. It is shown that the reduction of surface energy by solute adsorption plays an
essential role in giving easy void nucleation. A calculation is given for the stress needed for this mode of
nucleation using data for phosphorous in steel. Numerous examples exist of strongly adsorbing solute
inducing elevated temperature grain boundary cracking. A row of voids growing by stress driven boundary
di€usion is shown to develop a tensile stress maximum which aids void nucleation, giving rise to dimpled
grain boundary fracture. Our cracking model involves repeated void nucleation and is thus fundamentally
di€erent from the steady-state Hull±Rimmer model. At times and temperatures too low for void formation,
adsorption can lead to smooth grain boundary cracking at a rate controlled by solute di€usion (adsorp-
tion). # 1998 Acta Metallurgica Inc. Published by Elsevier Science Ltd. All rights reserved.

Intergranular creep fracture can occur through the hardly sucient to reach the theoretical nucleation
development of closely spaced voids on grain stresses. Thus the problem of cavity nucleation can-
boundaries oriented normal to the applied stress. not be regarded as being quantitatively under-
This mode of fracture is common in alloys at stres- stood.''
ses and temperatures low enough to allow only lim- The basic problem concerning nucleation is that
ited dislocation motion through the lattice. Thus, it if one takes the value of surface energy found in
covers a large part of the fracture maps of Ashby et high temperature zero-creep-rate experiments, the
al. [1] for alloys below about 0.75Tm (Tm is melting stress required for nucleation is much higher than
temperature in 8K). This fracture mode, with a the stresses that the metal will support at these elev-
®nely dimpled grain boundary fracture, is also ated temperatures. Two hypotheses have been put
found in some forms of atmospherically assisted forward to explain the nucleation process. Argon et
fracture such as the hydrogen attack of Cr±Mo al. [8, 14] argue that the only place where adequate
pressure vessel steels [2], as well as in stress relief stresses can develop in a metal undergoing creep is
cracking in welds in Cr±Mo steels [3, 4]. Numerous around grain boundary inclusions that lie on a slid-
authors have analyzed the growth of a regular ing grain boundary that is diagonal to the applied
array of voids on boundaries normal to the applied stress. If boundary sliding persists for some time
stress [5±14]. These models allow one to calculate a the stress can be high enough to give nucleation
time to fracture that depends primarily on the void near the inclusion on the diagonal boundary. The
spacing and the boundary di€usion coecient, Db. theory gives no satisfactory explanation for the
It satisfactorily describes the formation of the development of a uniform array of voids on bound-
dimpled fracture surface, but e€orts to ®t the the- aries normal to the applied stress. The second the-
ory to fracture times are less satisfactory, perhaps ory has been developed by Raj [16]. He concludes
due to the fact that the theory cannot predict when that the voids must nucleate on inclusions which
or how the voids form on the various grain bound- tend to be wet by the void, i.e. in which the void±
precipitate and void±matrix interfacial energies are
ary facets. That is, there is an unknown incubation
much lower than that of the precipitate±matrix
time before the assumed array of grain boundary
interface. To get a regular array of voids in this
voids develops on any given boundary. The ques-
model requires a density of grain boundary particles
tion of how the uniform array of voids originates is
at least equal to the density of voids, a high par-
not well understood. In a recent review article on
ticle±matrix interfacial energy and a particle radius
fracture mechanisms Riedel [15] concludes that: ``As
at least several times that of the critical radius of
a result of stress analyses it appears that the stress
the nucleating void. Such an array of particles is
concentrations cannot be exceedingly large, and are
possible but uncommon, and not as common as the
occurrence of intergranular (dimpled grain bound-
{Author to whom correspondence should be addressed. ary) creep fracture.
4861
4862 SHEWMON and ANDERSON: VOID NUCLEATION AND CRACKING AT GRAIN BOUNDARIES

Others have reported studies of grain boundary grain boundary, and P is an internal gas in the void
cracking controlled by surface and grain boundary which aids the expansion of the void/bubble and
di€usion. Chuang has treated the isolated crack, does work similar to the stress normal to the
whose growth produces a smooth grain boundary boundary. The pressure P was omitted from the
fracture [17]. Vitek [18] treated the growth of an other treatments since they have had no interest in
isolated void in a uniform stress and in a KI type treating this aid to nucleation. This leads to a
stress ®eld. He did not treat the nucleation of the nucleation rate of p (m2/s):
void, and assumed that the isolated void front
p ˆ z‰4pgs =…Sr ‡ P †O4=3 ŠDb dr
advances at a constant rate (a process we ®nd to be
unphysical). Raj and Baik [19] considered growth  exp‰ÿ4g3s f …a†=…Sr ‡ P †2 =kT Š …1†
driven by a KI type stress ®eld which drops rapidly
in front of the advancing crack. They assumed that where z is the Zeldovich factor, which is about
a new void nucleates in front of the crack when a 0.01 [21], gs=surface energy, O = atomic volume,
critical stress is exceeded, and that the voids all Db=boundary di€usion coecient, d = boundary
grew at a rate equal to that found in an array of thickness, a = contact angle between surface and
closely packed voids on the boundary. the plane of the grain boundary, and f(a) is a geo-
In a recent study, Chen and Shewmon [20] metric factor which when multiplied by the cube of
reported on the rate of dimpled grain boundary the radius of curvature of the critical nucleus, r*,
crack growth in quenched and tempered Cr±Mo gives the volume of a critical nucleus. (The radius
steels at elevated temperatures (440±5008C) in high of curvature of the spherical segment making up
pressure hydrogen. This is stress assisted hydrogen the void is related to the radius of the void on
attack in which an array of methane ®lled bubbles the grain boundary, a, by the equation a = r sina.)
develop on a few isolated boundary facets well in For a plane grain boundary f(a) 0 fv1 = (2p/
front of the main crack, e.g. 10±100 grain diam- 3)(2 ÿ 3cosa + cos3a) [22]. This will be of primary
eters. These isolated cracks raise the load on adjoin- interest to us, but we also wish to ®nd how much
ing boundaries leading to void formation on other easier nucleation is at the intersection of three
boundaries and ®nally to link-up and crack boundaries. In this case f(a) equals fv3, which is a
advance. There is a threshold stress intensity below much more complicated function. The equation can
which crack growth is negligible, and above which be found in Ref [22]. Values of fv1 and fv3 over a
crack growth increases with KI raised to a power of range of values of a are given in Table 1. As a
6 to 7. Fractography and growth kinetics indicate approaches 308, fv3 approaches zero so there is no
that the bubbles grow by stress assisted boundary barrier to nucleation of voids at triple lines for
di€usion just as in creep fracture. However, how a R 308. The void volume in terms of the radius a is
they nucleate has not been explained. a3fv1/sin3a. The term fv1/sin3a is also tabulated in
There is a well established tradition in this ®eld Table 1.
of analysis, starting with Hull and Rimmer [4], to In Cr±Mo steels, cracks grow at velocities of 0.1
assume all grain boundary voids are present at time to 10 micron/hour, and the bubbles are 2±4 microns
equal to zero. The analysis given here suggests that apart [20]. If voids form sequentially that are 2 mm
voids do not form uniformly across a grain facet apart the frequency needed for a crack growth rate
and then grow to failure as most others have of 1 mm/hour is about 10ÿ4/s. We argue below that
assumed. Instead voids/bubbles start forming at one the stress in front of a new void only builds up
edge of a boundary facet and move across it by suc- slowly to that required for nucleation of the next
cessive nucleation of voids. Once voids cover this void. We thus assume a nucleation rate of 10ÿ3/s,
boundary the load is shifted to the nearby parallel
boundaries, and the process is repeated there. Table 1. fv1 (bubble on grain boundaries) and fv3 (bubble at triple
line)

1. GRAIN BOUNDARY NUCLEATION 8 fv1 fv3 fv1/sin3a fv3/sin3a fv1/fv3

A nucleation rate is obtained by multiplying the 20 0.0224 0.0 0.5597 0.0


number of sites per unit area where a nucleus could 25 0.0534 0.0 0.7079 0.0
30 0.1077 0.0 0.8619 0.0
form, r, by the probability a critical nucleus will 35 0.1931 0.0051 1.0234 0.0269 38.0
develop on that site, exp(ÿG*/kT), times the fre- 40 0.3171 0.0321 1.1939 0.1210 9.9
quency with which a vacancy enters (atom leaves) a 45 0.4864 0.0974 1.3757 0.2754 5.0
50 0.7063 0.2164 1.5711 0.4814 3.3
critical nucleus. Several treatments of the nucleation 55 0.9801 0.4038 1.7831 0.7346 2.43
of voids on grain boundaries, made up of spherical 60 1.3090 0.6718 2.0153 1.0343 1.95
65 1.6915 1.0297 2.2722 1.3832
segments, have been published [8, 16, 21]. The criti- 70 2.1236 1.4828 2.5593 1.7869
cal radius of curvature for nucleation is r* = 2gs/ 75 2.5989 2.0316 2.8837 2.2543
(Sr+P), and the critical free energy for forming a 80 3.1087 2.6722 3.2548 2.7978
85 3.6426 3.3957 3.6845 3.4348
nucleus is G* = (Sr+P)r*3f(a)/2 = 4g3s f(a)/(Sr+P)2. 90 4.1888 4.1888 4.1888 4.1888
Here Sr is the remote (average) stress normal to the
SHEWMON and ANDERSON: VOID NUCLEATION AND CRACKING AT GRAIN BOUNDARIES 4863

Table 2. Data used in evaluating void growth The unique aspect of Hondros' data is that it
a±Fe Cu
provides the only measurements of excess for sur-
face and grain boundary (hereinafter designated Gs
Tm(K) 1810 1356 and Gb) which are available for any steel, and
O (10ÿ29 m3) 1.18 1.2
E (GPa) 196 124
almost for any alloy system. Equations (3) and (4)
n 0.28 0.35
G0 (GPa)* 16.9 11.3
Dobd (m3/s) 2  10ÿ12 3  10ÿ15
Qb (kcal) 42.3 24.8
B (m3/s) (0.5Tm) 1.9  10ÿ21 4.26  10ÿ22
B (m3/s) (0.45Tm) 1.54  10ÿ22 6.1  10ÿ23
B (m3/s) (0.4Tm) 6.6  10ÿ24 5.3  10ÿ24

*G0 = G/2p(1 ÿ n) = E/4p(1 ÿ n2).

to support the cycle time of 10ÿ4/s. Since each


nucleus must form in an area of about 10 mm2, the
nucleation rate of 10ÿ3/s in 10 mm2 corresponds to
p = 108 m2/s. Using the data for a±Fe in Table 2,
this corresponds to
g3s f …a†=…Sr ‡ P †2 R0:57 J3 =GPa2 …2†

where Sr and P are in GPa. Hydrogen attack


assisted cracking in Cr±Mo steels occurs when Sr
and P are each about 0.7 GPa [23]. At high tem-
perature with clean iron interfaces, this requires
a = 798 [24]. This gives fv1 = 3.0, and requires that
gs=0.72 J/m2, while for pure iron, gs=2.1 J/m2. If
the normal stress on the boundary is doubled
through stress concentrations or triaxiality, gs can
be no greater than 0.90 J/m2 for this nucleation
rate, much below the published value of 2.1 J/m2.
Faced with the problem of satisfying equation (2),
we will argue that gs is reduced to these low values
by adsorption of impurities at the relatively low
temperatures of 750±8508K. The work of Hondros
provides unique data on the variation of gs and gb
for a range of Fe±P alloys. Figure 1(a) reproduces
his results [24] and shows the drop in gs and gb as
the concentration of phosphorous is increased, at
14238K. Hondros' data for phosphorous on iron
surfaces are reasonably ®t by the Langmuir adsorp-
tion isotherm in which a monolayer, and only a
monolayer, of phosphorous adsorbs on the
surface [24]. Thus, gs in Fig. 1(a) is related to the
atom fraction of phosphorous in the lattice, XP, by
the integral of the Gibbs adsorption isotherm:
dgs ˆ ÿ GP dmP  GP RTd ln…XP †
Gibbs Adsorption eqn: …3†

with
GP ˆ Gmax bs Ks XP =…1 ‡ bs Ks XP † and
Ks ˆ exp…ÿEs =RT † …4† Fig. 1. Surface, gs, and grain boundary, gb, energies for
Fe±P alloys vs ln(XP) at 14238K; data taken from
G is called the surface excess of the adsorbed solute Hondros [23]; (b) calculated values of gs and gb at 750,
and is expressed in gm-atom (g-at)/m2. Es is the 800 and 8508K with equilibrium adsorption, based largely
on published data. The change in gb with temperature is
adsorption energy, and bs is a constant related to
negligible, while gs changes appreciably; (c) calculated
the entropy of adsorption. A similar set of values of gs and gb similar to (b) except that only 80% of
equations hold for the grain boundary energy, gb. the equilibrium surface excess is adsorbed.
4864 SHEWMON and ANDERSON: VOID NUCLEATION AND CRACKING AT GRAIN BOUNDARIES

thus allow one to estimate the shift in gs vs XP at


lower temperature, and thus to estimate the value
of gs and fv1 needed to describe nucleation between
7508 and 8508K) in steel. Figure 1(a) shows both gs
and gb start to change with ln(XP) at 14238K. That
is, the surface excesses, Gs and Gb, both start to rise
signi®cantly above zero at the same composition.
This gives one set of points needed to establish the
constants in equation (4). As averages for the sur-
face excess, Hondros' work gives Gs=2.3  10ÿ5 g-
at/m2 and Gb=1.1  10ÿ5 g-at/m2 [24, 25].
Grabke reports values for Es and Eb for
many binary Fe±X alloys using polycrystalline
samples [26]. He reports Es is ÿ190 kJ/mol for sul-
Fig. 2. Calculated variation of surface excess Gs and Gb
fur in iron, ÿ180 kJ/mol for phosphorous, ÿ110 kJ/ for phosphorous with ln(XP) for 750, 800 and 8508K.
mol for nitrogen, and ÿ85 kJ/mol for carbon. For Used to calculate curves in Fig. 1(b).
grain boundaries he reports Eb= ÿ 34.3 kJ/mol on
the fracture surface of Fe±P alloys, and Tatsumi
750, 800, and 8508K. Looking at equation (3) one
found Eb= ÿ 38 kJ/mol [27]. Presumably the frac-
sees that both Es>Eb and Gs>Gb contribute to the
ture surface exposes the grain boundaries that have
greater variation of gs with temperature than gb.
adsorbed the highest concentrations of phosphor-
The range of phosphorous concentrations studied
ous. We will take Eb= ÿ 35 kJ/mol, which requires
by Hondros was ÿ11 < ln(XP) < ÿ5. The concen-
bb=150 to ®t Hondros' high temperature data. For
trations in commercial steels might be
a free surface in Fe±P alloys, Es= ÿ 180 kJ/mol
ÿ9 < ln(XP) < ÿ7. Figure 3 shows the temperature
and this requires bs=0.0008 to ®t the data. The
stress lines at which nucleation will occur at 108 m2/
resulting curves for gs and gb are shown in Fig. 1(a)
s. There are three solid lines which correspond to
for 14238K.
nucleation stress for ln(XP) of ÿ9 (0.007wt%), ÿ8
In the Cr±Mo steels that show crack growth in
and ÿ7 (0.039wt%) at Gs=0.8Geq. The three nearly
hydrogen, bubbles/cracks develop early on only a
horizontal dashed lines corresponding to three
small fraction of the grain boundaries [20]. That is,
di€erent contact angles (ratios of gb/gs). These are
nucleation is appreciably easier on some boundaries
a = 608(gb/gs=1), a = 458(gb/gs=Z2), and a = 308
than on others. In the model leading to equation (1),
(gb/gs=Z3). Not shown is a = 0.08 (gb/gs=2) which
nucleation would be fastest on boundaries with
would lie somewhat below the zero stress line.
higher gb, and on which voids would expose sur-
Figure 1(b) shows that as XP rises, and as the tem-
faces with a lower gs. There are no data on the vari-
perature falls, gs drops below gb. This reduces the
ation of gsand Gmax
s with surface orientation. There
nucleation barrier to a low level as the condition
are limited data available on Eb for speci®c grain
for grain boundary ``wetting'' by voids is
boundaries. Lejcek and Hofman studied two
approached, i.e. as (2gsÿgb) approaches zero. Figure
speci®c boundaries in steel [28] and reported values
3 re¯ects this by the stress needed to nucleate voids
of Eb about 1/3 of 35 kJ/mol. This suggests that the
becoming very small. No lines are shown for the
energy and surface excess of grain boundaries vary
more than those for free surfaces. We have chosen
to approximate the energy of boundaries at which
nucleation would ®rst occur by taking
Gmax
s =2.53  10ÿ5 g-at/m2 (10% greater than
Hondros' average) and Gmax b =0.9  10
ÿ5
g-at/m2
(20% less than Hondros'). Figure 1(b) shows gs and
gb vs ln(XP) using these equilibrium of values of G
and the same values of E and b used to ®t
Hondros' data. But with these values of gs there is
no nucleation barrier for voids under the conditions
of interest. Figure 1(c) shows comparable curves for
a state the surface would pass through before it
reaches equilibrium, that is, for Gs equal to 80% of
the equilibrium value. This gives a ®nite nucleation
barrier. Both sets of curves show that values of gs
di€er appreciably for the three di€erent tempera- Fig. 3. Normal stress for a void/bubble nucleation rate of
108 m2/s on grain boundaries of Cr±Mo pressure vessel
tures of 750, 800, and 8508K, while there is little steels for values of ln(XP) = ÿ 9, ÿ8 and ÿ7. The nearly
change in gb over this temperature range. Figure 2 horizontal dotted lines represent the stress for contact
gives the variation of Gb and Gs with ln(XP) for angles of a = 30, 45, and 608.
SHEWMON and ANDERSON: VOID NUCLEATION AND CRACKING AT GRAIN BOUNDARIES 4865

case of Gs=Geq since the stress needed for nuclea- thus need to occur at higher temperatures than
tion in this case is zero at the temperatures of inter- dimpled grain boundary fracture in creep because
est. That is, adsorption assisted void nucleation di€usion of the segregating element would be faster.
occurs very rapidly before the equilibrium amount There is some experimental evidence to support
of segregation has built up. This is discussed further this.
below. Additional phosphorous, present above with
Two e€ects indicated in Fig. 3 are noteworthy. equilibrium grain boundary segregation, is needed
There are no experimental measurements of gs or gb for void nucleation since when a void nucleus forms
at 750±8508K to compare with the predictions of an additional amount of solute roughly equal to
Fig. 1(b,c) and no prospect of ever getting any. (2Gmax
s ÿGmax
b ) must accumulate for each unit area
Thus, it is appropriate to consider the ``robustness'' of grain boundary replaced by void surfaces. Since
of the reasoning that leads to it. The Gibbs adsorp- Gmax
s is twice Gmax
b for this case, the amount of
tion equation (3) is well established. The use of the phosphorous in the region must increase by a factor
Langmuir±McClean equation (4) is less well estab- of 4. However, the time for this accumulation
lished, but it has been recommended by Hondros should be comparable to the time required for the
and Seah as best for Fe±P alloys [25]. If it is in vacancy aggregation needed to form the void
error, they indicate the true values of Gs and Gb (at nucleus so it should not limit or inhibit void nuclea-
T < 14238K) would be even larger than the high tion.
temperature values we use. This would mean that The other e€ect of concern is the fact that other
the true values of gs and gb are lower than our esti- elements may segregate to the surface more strongly
mates, and our conclusion about easy nucleation than phosphorous. For example, sulfur adsorbs
and cracking would be unchanged. The main issues more strongly than phosphorous [26], and
that might change our conclusion are: Luckman [29] reports that sulfur displaced phos-
phorous on free surfaces in their work, at these
1. Is equilibrium attainable in the time available?
temperatures. Misra shows that Cr and N can
2. Will the many elements that can adsorb at sur-
adsorb strongly to surfaces in steels. It may well be
faces in steel (both alloying and tramp elements)
true that other elements will displace phosphorous
invalidate these conclusions which were based on
on steel, but this will not invalidate our main con-
data for pure Fe±P alloys?
clusion that nucleation is made possible by adsorp-
Equilibrium segregation on the grain boundary tion at surfaces since for such a solute displacement
must occur by di€usion through the lattice, and lat- to occur ``spontaneously'' (in the thermodynamic
tice di€usion can be quite slow at temperatures less use of this word) the adsorption of the new element
than half the melting temperature (7508K/ (and displacement of phosphorous) must reduce the
18108K = 0.41Tm). Two di€erent studies have been surface energy even further than does phosphorous
made of the kinetics of development of an adsorbed adsorption alone. Thus, our conclusion about solute
layer of phosphorous in commercial steels at rela- adsorption allowing void nucleation at low stresses
tively low temperatures. Luckman et al. [29] studied is not invalidated but strengthened.
the development of an adsorbed layer on Fe±P The literature lends strong support to our con-
alloys and on commercial steels, including a clusion that trace amounts of certain impurities will
2.25Cr±1Mo steel. At 8738K they found it took decrease creep ductility, increase void density in
2 ks to reach 0.9Geq. Using the activation energy of creep, and cause grain boundary separation under
Viefhaus and Richarz [30] for this process (230 kJ/ stress at elevated temperature. In some of these
mol) the time at 7738K would be 120 ks, and at cases the data indicate strong adsorption. In other
7338K, 950 ks. Misra et al. [31] measured the rise of cases one can follow Seah's work [32] and take low
phosphorous on the grain boundaries of a solid solubility in the base metal as a strong indi-
NiCrMoV steel with 0.24C and 0.01wt% P, and cation of adsorption.
found the phosphorous concentration in the bound- Grant and Mullendore [33] report the absence of
aries peaked after 80 ks at 7738K, which compares grain boundary cracking/voids in high purity alumi-
reasonably with the extrapolated estimate of 120 ks num creep specimens but its presence in 2S (com-
given above. This suggests that the grain boundary mercially pure) Al, or high purity Al±Mg alloys.
surface layer should come to equilibrium with the Ransley and Talbot reported that 0.001% Na
lattice in about a day at 7738K and about a week severely impaired the hot working of Al±Mg alloys
at 7338K. These times are long but are shorter than due to the growth of grain boundary cavities [34].
the times for void formation in creep, and would (Sodium has virtually no solubility in solid alumi-
provide another explanation of why grain boundary num, and very little in liquid aluminum, so it
voids are observed primarily in third stage creep. should adsorb strongly at aluminum surfaces.)
They are longer than the times available in stress Gaertner and Grabke report that phosphorous ad-
relief cracking (SRC). This would suggest that ditions at the level of 0.01% clearly reduce creep
dimpled grain boundary fracture in SRC occurs ductility in Ni±20Cr and Nimonic 80A alloys, and
with less than equilibrium segregation and would that it adsorbs to grain boundaries and fracture
4866 SHEWMON and ANDERSON: VOID NUCLEATION AND CRACKING AT GRAIN BOUNDARIES

surfaces [35]. Tipler and Hopkins [36] compared (800±8758K) the cracking mode changed to a
creep cavitation in heats of commercial and high dimpled grain boundary fracture. In the heat with
purity Cr±Mo±V steels at 5508C (8238K). The two no phosphorous doping, a dimpled grain boundary
heats di€ered by roughly a factor of 2 in S, P, and fracture was found at the higher temperature, but
Sn (all of which Hondros and Seah have shown no cracking was found in the lower temperature
adsorb strongly [24]), and by a factor of 10 in As, range. Bika et al. found similar behavior in low
Sb, and N, which have low solid solubility. The re- alloy steel doped with 0.012% S [44].
duction in trace elements resulted in a substantial Figure 4 repeats the 7508K lines of Fig. 1(c), but
gain in stress rupture ductility and life, in delayed adds a line for twice the energy of a surface in equi-
cavitation and it reduced cavity density by two librium with the matrix concentration of phosphor-
orders of magnitude. Tipler and McLean showed ous. The point where wr=0 is shown on Fig. 4 and
that small additions of antimony to copper substan- occurs at about ln(XP) = ÿ 7. If the phosphorous
tially increased creep cavitation and reduced creep concentration in the matrix is less than this, exter-
ductility to less than 1/4 that for pure copper [37]. nal work is needed for decohesion. It is found that
They also showed that this antimony addition an applied stress enhances the rate of crack
reduced the surface energy of copper to 1/4 or 1/5 growth [43, 44]. A rough estimate of the work such
that of pure copper [38]. Gibbons et al. [39, 40] a crack opening stress might do is the local (crack
have shown that the creep ductility of Ni based tip) normal stress (0.7 to 1.5 GPa) times the separ-
superalloys is signi®cantly reduced by 10±20 ppm of ation at the maximum of attraction between two
Pb or Te. White et al. studied creep in Ni±20Cr new surfaces, taken as 0.5 nm. This product gives
and reported that with 0.002% sulfur the ductility work equal to 0.35 to 0.75 J/m2 so that an applied
was quite limited, voids formed easily and sulfur stress could aid grain boundary decohesion in the
was adsorbed on the exposed fracture surfaces range 0.75 J/m2>wr>0. This range is shown on the
(prior grain boundaries) [41]. If 0.11% Zr was ®gure. The value of ln(XP) = ÿ 7 is where Hippsley
added, the sulfur was collected in a grain boundary et al. found the smooth grain boundary fracture. It
precipitate and, rather than the precipitate acceler- is very likely that the stress on the sample allowed
ating void nucleation as the theories of Raj or fracture to occur with less than full equilibrium seg-
Argon would suggest, the sulfur rich precipitate regation, that is, over the range of local lattice con-
increased elongation to 150%. It also changed the centrations of ÿ10 < ln(XP) < ÿ7. McMahon and
fracture mode from grain boundary (with free sul- Vitek [47] have also treated such cracking and point
fur) to transgranular (with sulfur tied up by zirco- out that the grain boundary crack could be quite
nium in a precipitate) [41, 42]. sharp and thus the stress concentration at the crack
These examples give a fair indication of the many tip would be still higher than our assumed upper
cases in which impurity adsorption at surfaces value of 1.5 GPa.
enhances the formation of grain boundary voids A requirement for the validity of a value of wr
and reduces creep ductility. near zero is that enough phosphorous di€use to the
advancing crack to raise the local excess from Gb,
the excess on the grain boundary prior to the separ-
2. SMOOTH GRAIN BOUNDARY FRACTURE ation, to about 2Gs for the two surfaces. This is an
increase of about 4. Di€erent authors have treated
The reversible work of decohesion for separating this di€usion problem, postulating that the crack
a grain boundary into two solid-vapor surfaces is growth rate is limited by solute di€usion to the
roughly wr=2gsÿgb. So far we have dwelt on the
idea that as wr becomes quite low it becomes easier
to nucleate voids. However, it also becomes easier
(requires less work) to simply separate the grains at
the boundary without any voids being formed.
Several studies of grain boundary hot cracking in
steel [4, 43±45] have demonstrated a di€usion-con-
trolled decohesion mechanism that gives a smooth
(as opposed to dimpled) grain boundary fracture.
As one example, Hippsley et al. [4] studied stress
relief cracking in a controlled purity 2.25Cr±1Mo
steel doped with 0.054wt% P (lnXP= ÿ 7) and in
an undoped heat of the same steel with 0.007wt%
P(lnXP= ÿ 9). They found two distinct cracking
regimes in the P-doped alloy during a simulated
post weld heating cycle. At low temperatures (700± Fig. 4. 7508K lines of Fig. 1(c) with lines indicating the
7508K) a smooth grain boundary fracture mode reversible work needed for grain boundary separation
was produced, whereas at higher temperatures (2gsÿgb).
SHEWMON and ANDERSON: VOID NUCLEATION AND CRACKING AT GRAIN BOUNDARIES 4867

opening crack [45, 46]. However, there is some con- ond set of voids can form on the boundary away
fusion about how the stress maintains the low from the triple line and, if it can, at what distance
solute potential around the advancing crack tip. We the new set will form. The nucleation and growth
believe the appropriate boundary condition must be of the second, and successive, row of voids would
obtained roughly as follows. If the stress concen- then lead to the propagation of voids away from
tration at the crack tip were sucient to open a the triple line that is observed in cross-sections of
crack with no additional phosphorous di€using to samples developing a dimpled grain boundary frac-
the surface, then the surface excess on the new sur- ture. Such sequential nucleation can result from the
faces formed would be given by Gs=Gb/2, or about stress maximum that develops in front of the wedge
4.5  10ÿ5 g-at/m2. Figure 2 indicates this is in equi- of material that di€uses into the grain boundary
librium with a lattice concentration of near the triple line as the new voids grow. The
ln(XP) = ÿ 19 at 8008K. Thus, in a sample with a stress calculation/di€usion problem given below in-
bulk concentration of ln(XP) = ÿ 7, the growing dicates that the normal stress reaches a maximum
crack would act as a sink with the lattice concen- on the grain boundary at some distance in front of
tration next to the crack tip ranging over the growing voids. It treats an isolated void on a
ÿ19 < ln(XP) < ÿ7 depending on the stress avail- grain boundary which thickens only near the void
able and needed to open the crack. and does not thicken far from the void. This is basi-
The ®ndings of Hippsley et al. that smooth grain cally di€erent from a Hull±Rimmer model which
boundary fracture in phosphorous-doped Cr±Mo assumes a uniform thickening rate for the entire
steels can occur 1008K below the temperature grain boundary.
where dimpled fracture occurs indicates that phos- Consider the case of a boundary with a remote
phorous di€usion (on grain boundaries and through normal stress of Sr and internal pressure P, acting
the lattice) is fast enough to adsorb and aid grain on a void/bubble with a cross-section as shown in
boundary fracture at a temperature about 1008K Fig. 5(a). We assume:
lower than that required for grain boundary void
1. all di€usion occurs along the grain boundary, so
formation through vacancy di€usion.
that the volume of the void equals the volume
di€used into the grain boundary;
3. SEQUENTIAL VOID NUCLEATION AND STRESS
DRIVEN DIFFUSION

We have shown that adsorbing impurities can


greatly ease the nucleation of grain boundary voids.
This leaves unanswered the question of how the
regular array of voids is formed on the dimpled
grain boundary fracture surfaces produced in creep,
hydrogen attack, or stress relief cracking. In metals
undergoing creep, void nucleation often starts at
grain boundary triple points. (In Cr±Mo pressure
vessel steels, the sequence of bubble nucleation may
also start on grain boundaries at the end of an in-
clusion stringer, or in some bandedÐimpurity
richÐarea.) We will treat the case of nucleation
along a triple line with one boundary normal to the
stress since the geometry is well de®ned. However, a
similar, more qualitative, argument could also be
given for nucleation starting at other catalysts like
stringers or lines of inclusions.
Consider the line along which three grain bound-
aries meet. For the low contact angles of relevance
here, nucleation will always occur more easily, i.e.
earlier, at a triple line than on a grain boundary.
This follows from equation (1) and the fact that fv3
is always less than fv1 (Table 1), and usually much
less. In many cases local shear stress relaxation by Fig. 5. (a) Grain boundary void showing coordinates,
sliding along the boundaries at the triple line that remote stress, Sr, internal pressure, P, and the dislocations
are diagonal to the applied stress will also aid used to describe the void in the analysis; (b) schematic dia-
nucleation too, but this is not necessary. Once gram of voids in a grain boundary with a normal stress
acting on it. The ®rst voids form on the triple line formed
nucleation starts along a triple line, a string of by intersection of three grain boundaries. The next gener-
voids will quickly form along the triple line. The ation of voids form on the grain boundary away from the
critical questions in propagation are whether a sec- triple line.
4868 SHEWMON and ANDERSON: VOID NUCLEATION AND CRACKING AT GRAIN BOUNDARIES

2. the chemical potential of atoms in the boundary l = (Bt)1/3. The result is that
is determined by the local normal stress, Sn,
through the equation m = ÿ OSn;
^ x†
dY… ^ ^ x† d2 S~n …x†
^
x^ ÿ Y… ^ ˆ3 …cyl:†;
3. no stress relaxation occurs by lattice creep; and dx^ dx^ 2
4. the void surface and the grain boundary meet ^ r†  ~ 2 ~ 
dY…^ ^ r† ˆ 3 dSn …^r† ‡ r^ d Sn …^r† …axi:†: …8†
with the equilibrium contact angle, a, throughout r^2 ÿ r^Y…^
d^r d^r d^r2
the growth process.
Two void geometries are considered. The ®rst is An implicit assumption in such a scaling is that a^ cav
that of a cylindrical void with the z-axis lying paral- is a constant, so that the void edge, acav, scales line-
lel to a grain boundary triple line (junction), with arly with l and the cavity growth is self similar.
void edges located at x = + acav, ÿacav, and the This assumption is reasonable for the small voids
grain boundary normal along the y-direction. treated here where surface di€usion can maintain
Matter from the void di€uses along the grain the equilibrium angle a and until the voids grow to
boundary in the x-direction, where |x|>acav. A uni- where cavity interaction destroys self-similarity. A
form remote tension, Sr, is applied normal to the consequence of this rescaling is that the solution for
boundary and the resulting wedge pro®le and stress the wedge opening is normalized by l. An impli-
distribution are independent of z. cation of this rescaling is that if Y(x*,t*) and
The second is that of an axisymmetric void with Sn(x*,t*) are known, then at some multiple, c, of
axis of rotation normal to the grain boundary and that time, Y(c1/3x*,ct*) = c1/3Y(x*,t*) and Sn(c1/
3
x*,ct*) = Sn(x*,t*). Thus, the rescaling maps dis-
with the void edge located at r = acav Matter from the
tributions of wedge height and stress over all time
void di€uses along the grain boundary in the r-direc- ^ x† ~ x†.
into the functions Y… ^ and S… ^
tion, where r>acav. A uniform remote tension, Sr, is ^ in the
The stress at some position …x^ ˆ a, ^ y^ ˆ h†
applied normal to the boundary, and a pressure P ^
^ y^ ˆ h† in the axisymmetric
cylindrical case or …^r ˆ a,
may also develop within the void. The resulting wedge
case may be written as
pro®le and stress distribution are independent of
angular position y within the grain boundary. X
Mÿ1
S~n …a, ^ ˆ S~r ‡
^ h† ~ a,
g… ^ a^ i ,h^i †
^ h;
A governing equation for each geometry is that
iˆ1
the ¯ux, Jb, along the boundary is given by
Db d @ m Db d @ m  X
M‡N
Jb ˆ ÿ …cyl:†; Jb ˆ ÿ …axi:†: …5† ~ a,
‡ g… ^ a^i , ÿ h^i † b^i ‡
^ h; ~ a,
g… ^ a^i ,0†b^i : …9†
^ h;
OkT @ x OkT @ r iˆM

If Y is designated the thickness of the wedge of dif- The ®rst term is the remote applied stress. The sec-
fused material along the grain boundary, then con- ond term approximates the stress ®eld of the void,
servation of matter requires that by inserting into an in®nite, homogeneous, elastic
ÿ 
@ Y…x,t† @ Jb @ Y…r,t† O @ rJb material a quantity of M-1 dislocation loops posi-
@t
ˆ ÿO
@x
…cyl:†;
@t
ˆÿ
r @r
…axi:†: tioned at coordinates …a^i ,h^i † along the surface of the
…6† upper half of the void, and a symmetric distribution
at …a^i , ÿ h^i † in the lower half, as depicted in
When equations (5) and (6) are combined to elimin- Fig. 5(a). The loops have Burgers vectors bi normal
ate Jb and the stress-dependent chemical potential to the grain boundary. The last term contains one
in assumption 2 above is used to eliminate m, the additional loop for the void edge lying in the grain
following second order di€erential equations result: boundary, and the remaining loops from
i = M + 1 to M + N approximate the wedge of
@ Y…x,t† @ 2 S~n
ˆ ÿB …cyl:†; di€used material outside the void.
@t @x2 The function g~ provides Sn/G' at some position
  ^ due to a dislocation loop at position …a^ i ,h^i † with
@ Y…r,t† 1 @ S~n @ 2 S~n ^ h†
…a,
ˆ ÿB ‡ 2 …axi:† …7a† unit Burgers vector. For the cylindrical geometry,
@t r @r @r
where …a ÿ ai †‰3…h ÿ hi †2 ‡ …a ÿ ai †2 Š
~
g…a,h;a i ,hi † ˆ
‰…h ÿ hi †2 ‡ …a ÿ ai †2 Š2
Db dOG 0 G …†
Bˆ , G0 ˆ , …e† ˆ 0 …7b†
kT 2p…1 ÿ † G …a ‡ ai †‰3…h ÿ hi †2 ‡ …a ‡ ai †2 Š
ÿ …cyl:†
where G is the elastic shear modulus, and n is ‰…h ÿ hi †2 ‡ …a ‡ ai †2 Š2
Poisson's ratio. …10a†
The form of equation (7a) permits the opening
and stress distributions around the void to be and for the axisymmetric geometry,
expressed as a function of position x^ in the cylindri- ~
1 @ U…a,h;ai ,hi †
cal case and r^ in the spherical case, where (^) ~
g…a,h;ai ,hi † ˆ ÿ …axi:† …10b†
2pa @a
denotes normalization with respect to the length
SHEWMON and ANDERSON: VOID NUCLEATION AND CRACKING AT GRAIN BOUNDARIES 4869

where the interaction energy between two circular


prismatic dislocation loops is given by [48]
   
~ m 1=2 2 ÿ m p
U…a,h;ai ,hi † ˆ paai 2 K… m†
aai m
  3=2
2 p m
ÿ E… m† ‡ …h ÿ hi †2
m aai
 
1 p 2ÿm p
 ÿ K… m† ‡ E… m†
m 2m…1 ÿ m†

…11a†
p p
F… m† and E… m† are complete elliptic integrals of
Fig. 6. Ratio of grain boundary normal stress, Sn, to
the ®rst and second kind [49], where remote stress, Sr, as a function of distance from the void
4aai radius, expresses in units of l = (Bt)1/3 for various values
mˆ : …11b† of Sr/G' and P/G'. The inset expands the region of the
…a ‡ ai †2 ‡ …h ÿ hi †2 maximum. All are for a contact angle of a = 308.
Solutions are obtained by specifying the void angle
various curves re¯ect the di€erences in cylindrical
(a), the void dimension (a^ cav ), and the fraction (P/Sr)
and axisymmetric geometries, and magnitude of
of internal void pressure. A unit applied stress is
loading, expressed as Sr/G' and P/G', and given
assumed initially, so that Sr=1, and all bi are deter-
beside each curve in the inset. Figure 7 shows the
mined uniquely by satisfying the di€erential
shape of the corresponding wedge of material that
equation (8) at points i = M + 1 to M + N along the
has di€used into the grain boundary around the
grain boundary, and by satisfying that the stress pre- void. Several points are clear:
dicted from equation (9) is ÿP/Sr at M locations
within the void. The form of equation (9) reveals that 1. a peak in normal stress occurs at about
when Sn= ÿ P/Sr in the void, P/Sr can be added to 3.8(Bt)1/3 away from the cavity for all curves;
both sides of the equation so that Sn=0 in the void 2. this peak is much larger for the cylindrical than
and Sr=1 + P/Sr . Typically M = 30 and N = 60 axisymmetric voidÐabout (1.22±1.23) (Sr+P) vs
are used to obtain answers for stress and wedge height (1.03±1.04) (Sr+P);
to within 1%. 3. the wedge of di€used material decreases to zero
The Sr corresponding to the prescribed a, a^cav , at about 4(Bt)1/3, the approximate location of
and P/Sr is determined uniquely by requiring that the stress peak;
the volume of the wedge along the grain boundary 4. the void radius, acav/(Bt)1/3, increases appreciably
equals that of the void. This condition is im- with applied stress;
plemented easily since the bi are linearly dependent 5. the cavity size and stress peak increase appreci-
on Sr. Thus, if Sr were doubled for this combi- ably as the cavity tip angle, 2a, is decreased; and
6. the ratio acav/(Bt)1/3 is proportional to the rate
nation of a, a^ cav , and P/Sr, the values of all bi
of expansion of the void. It increases with
would double and hence the volume of the wedge
increasing driving force, i.e. Sr+P, and decreas-
would double. The equality of the void and wedge
ing contact angle a. This higher velocity of acav
volumes thus uniquely determines the remote stress
stems from (corresponds to) a steeper wedge of
needed.
inserted material.
 
a^ cav 2
…a ÿ sinacosa†
sina
S~r ˆ …cyl:†;
X
M‡N
^
bi …a^i ÿ a^cav †
iˆM‡1
 3
a^ cav
fv1…a†
sina
S~r ˆ …axi:†: …12†
X
M‡N
b^i p…a^ 2i ÿ a^2cav †
iˆM‡1

The b^i used in equation (12) is the solution for a


unit remote stress.
Figure 6 shows normal stress on the grain bound-
ary as a function of distance away from the cavity Fig. 7. Shape of wedge of material di€using into grain
(void) edge, measured in units of l = (Bt)1/3. The boundary ahead of void for Fig. 6 parameters.
4870 SHEWMON and ANDERSON: VOID NUCLEATION AND CRACKING AT GRAIN BOUNDARIES

With this understanding of the stress maximum, the triple line would be at a distance of
we now turn to the question of sequential nuclea- 1000 nm3 l mm from the triple line. The time
tion. If the stress is sucient to cause void nuclea- required for this much di€usion is (1 mm)3/B, or
tion at some point along a triple line, a void will 104 s in steel at 7708K. This time is reasonable, as
grow. Material di€using from the void will insert a is the spacing, though the observed spacing is
wedge of material that locally unloads the boundary often somewhat greater than 1 mm. It is quite
and develops a stress maximum, Smx, around the possible that the observed void spacing on a frac-
void. This will lead to new nucleation events at ture surface is coarser due to void coalescence on
some distance on either side of the initial void, and the grain boundary before fracture.
ultimately to a row of voids along the triple line Once a new line of voids is growing o€ of the tri-
with a spacing d. For short times, i.e. (Bt)1/3<d, the ple line, the growing voids will unload a widening
stress ®eld around each void is given by the axisym- region on the edge of the boundary facet they are
metric solution. For the case of (Bt)1/3>>d, the on, thus raising the remote stress on the boundary
grain boundary wedges along the triple line will be in front of this new set of voids. Again, the stress in
in equilibrium with the voids, though the voids will front of the voids will rise as the stress ®elds from
still be separate, and the grain boundary normal adjacent voids overlap and grow into a more con-
stress around the triple line will be given by the tinuous front producing a stress more characteristic
cylindrical solution. Since the voids would still be of our cylindrical case than the axisymmetric one.
isolated, an internal pressure would supplement the This produces a stress adequate to again cause
applied stress just as it does for the ``axi.'' case of nucleation of a new rows of voids. The time
P/G'>0 in Figs 6 and 7. required to get this set of voids should be compar-
The nucleation barrier, G*, for voids on triple able to that for the ®rst since the required di€usion
lines is always much lower than that on grain distance to attain a cylindrical geometry is again
boundaries since fv3 is always less than fv1 (see about (Bt')1/3>>d, where t' is the time after the
Table 1). In fact, for contact angles of 308and less, nucleation of the previous row of voids. Thus,
G* on the triple line is zero. Thus, for the contact when the conditions are such that the ®rst row of
angles of interest here, say a R 458, voids will form voids o€ the triple line can nucleate, the process
easily along the triple line and the critical step limit- will repeat itself and the developing array of voids
ing the development of voids across a grain bound- should move across the facet. This sort of a model
ary facet will be the nucleation and growth of the is consistent with the wedge shaped cracks, and
®rst row of voids on the grain boundary but away rows of voids, that are often observed in samples
from the triple line. The stress o€ the triple line will going into third stage creep [1].
rise to a maximum at a distance of about 3.8(Bt)1/3
from the cavity tip and have a value of about
1.23Sr/G'. Thus, G* is reduced to about G*/(1.23)2 4. HYDROGEN ASSISTED CRACK GROWTH IN
STEEL
in a region which expands in width as t1/3 and
moves away from the triple line as t1/3. Two studies have shown that an applied stress
The next row of voids would be expected to will increase the local rate of hydrogen attack (high
develop in this higher stressed region, well away pressure hydrogenÐmethaneÐassisted cracking) in
from the ®rst row of voids (Fig. 5(b) and Fig. 6), hot carbon steel [50] and Cr±Mo steel [20]. A creep
indicates the separation between the two rows study found that hydrogen pressure can greatly
would be a distance of about 10acav. The distance reduce the creep ductility of Cr±Mo steels in the
between the new line of voids and the triple line same temperature range [51]. The fracture surface
cannot be calculated with any precision, but an formed is always a dimpled, grain boundary frac-
estimate can be obtained as follows. When an iso- ture with a dimple spacing of a few microns in Cr±
lated void ®rst forms in steel in a region stressed Mo steels [20]. Both cracking studies [20, 50] found
to 100 ksi (0.7 MPa), r*/sin(a) = acav* is the order distributed cracking on isolated grain boundaries at
of 1 nm. This cavity embryo must grow to roughly a distance of 20±50 grain diameters in front of the
10 times that size before the capillarity force main crack of a compact tension specimen. The
restraining the growth of the void, 2gs/r, becomes density of such crack segments increased as the
negligible and Smx reaches its steady-state value main crack root was approached. Chen [20] reports
around the void (this is the ``axi'' case). Now a threshold for accelerating crack growth of
acav=10acav* 3 10 nm. Figure 6 indicates that KI=20 MPaZm, and measured macroscopic crack
acav/(Bt)1/330.3 at Smx so the distance between growth rates of 1±25  10ÿ3 mm/s at 7508K.
voids is 3.8(Bt)1/3/0.3, or a distance of 11acav, or This type of crack growth requires very active
about 110 nm along the triple line, d. If we say nucleation of voids on the grain boundaries. The
the voids on the triple line must increase another observation that voids ®rst initiate on boundary
order of magnitude before we satisfy the condition segments well ahead of the main crack could re¯ect
(Bt)1/3>>d for developing the full cylindrical local conditions of impurity concentration and sur-
stress. Thus, the ®rst row of voids nucleated o€ face energy that favor void nucleation. Figure 8
SHEWMON and ANDERSON: VOID NUCLEATION AND CRACKING AT GRAIN BOUNDARIES 4871

methane bubbles after exposure to conditions giving


hydrogen attack [55]. Tin is one of the elements
that Hondros and Seah [25] have shown will adsorb
and markedly reduce the surface energy of steel.
Also, Knott et al. [43] have reported that phosphor-
ous, sulfur and other impurities aid in the develop-
ment of the same kind of dimpled grain boundary
cracking in post weld stress relief anneals.

AcknowledgementsÐWe would like to acknowledge help-


ful remarks by Roland DeWit of NIST, and the support
of AFOSR Grant F49620-96-1-0238.

Fig. 8. Normal stress for void/bubble nucleation rate for


ln(XP) = ÿ 9, and ÿ7 at 80% of equilibrium surface REFERENCES
excess, and ln(XP) = ÿ 9 for full equilibrium surface
excess. The solid lines are the same as in Fig. 3. The 1. Ashby, M. F., Gandhi, C. and Taplin, D. M. R., Acta
dashed lines are the nucleation stress when the voids con- metall., 1975, 23, 425.
tain the equilibrium methane pressure typically formed in 2. Hakkarainen, T. J., Wanagel, J. and Li, C.-Y., Metall.
Cr±Mo steels in 20 MPa of hydrogen gas. Trans., 1980, 11A, 2035.
3. Erwin, W. E., Proc. Am. Petro. Inst. (Re®n.), 1982,
61, 120.
4. Hippsley, C. A., Knott, J. F. and Edwards, B. C.,
shows the normal stress on the grain boundary
Acta metall., 1980, 28, 869.
required for the nucleation rate and values of 5. Hull, D. and Rimmer, D. E., Phil. Mag. A, 1959, 4,
ln(XP) used in Fig. 3, that is, for segregation at 673.
80% of Gmax on the surface. Also shown is the 6. Speight, M. V. and Harris, J. E., Metal Sci. J., 1967,
nucleation stress in the presence of internal methane 1, 83.
7. Raj, R. and Ashby, M. F., Acta metall., 1975, 23, 653.
pressure, the pressure being chosen as that in equili- 8. Argon, A. S., Chen, I.-W. and Lau, C. W., in Creep±
brium with a typical Cr±Mo pressure vessel steel Fatigue±Environmental Interactions, ed. R. M. Pelloux
and hydrogen at a pressure of 20 MPa [52, 53]. At and N. S. Stolo€. AIME, New York, 1980, p. 46.
7508K the stress required for void nucleation is 9. Chuang, T. J., Kagawa, K. I., Rice, J. R. and Sills, L.
B., Acta metall., 1979, 27, 265.
reduced by about 500 MPa. However, the observed
10. Vander Burg, M. W., Vander Giessen, E. and
increase in crack growth rate above the threshold is Brouwer, R. C., Acta metall., 1996, 44, 505.
so rapid that there must be an increase in the frac- 11. Vitek, V., Acta metall., 1978, 26, 1345.
tion of grain boundaries developing voids as well as 12. Chuang, T. J., J. Am. Ceram. Soc., 1982, 65, 93.
an increase in the nucleation rate on any given 13. Chuang, T. J., Chu, J. and Lee, S., Trans. ASME,
1996, 63, 796.
boundary. The nucleation of voids at an applied 14. Argon, A. S., in Recent Advances in Creep and
stress of zero (Fig. 8) would correspond to hydro- Fracture of Engineering Material and Structures, ed. B.
gen attack in these steels in the absence of an Wilshire and D. R. J. Owen. Pineridge Press, Swansea,
applied stress, as is observed. 1982, p. 1.
15. Riedel, H., in Material Science Technology, Vol. 6, ed.
Several workers have noted that the rate of R. W. Cahn, P. Haasen and E.J. Cramer. VCH
hydrogen attack (crack growth rate) increases with Verlag, New York, 1993, pp. 565±634.
temperature up to at least 8758K [20, 52, 54], while 16. Raj, R., Acta metall., 1978, 26, 995.
the nucleation stress curves given in Fig. 8 indicates 17. Chuang, T. J., J. Am. Ceram. Soc., 1982, 65, 93.
that the nucleation rate decreases as the tempera- 18. Vitek, V., Acta metall., 1978, 26, 1345.
19. Raj, R. and Baik, S., Metal Sci. J., 1980, 14, 385.
ture rises. We would interpret this to mean that at 20. Chen, L.-C. and Shewmon, P. G., Metall. Trans.,
low temperatures void nucleation and cracking 1995, 26A, 2317.
occurs without full equilibrium segregation, but 21. Evans, A. G., Rice, J. R. and Hirth, J. P., J. Am.
that at higher temperatures the surface excess gets Ceram. Soc., 1980, 63, 368.
22. Clemm, P. J. and Fisher, J. C., Acta metall., 1955, 3,
closer to the equilibrium value so that the nuclea- 70.
tion rate increases as the temperature rises. In sup- 23. Parthasarathy, T. A. and Shewmon, P. G., Metall.
port of this, Fig. 8 also shows the nucleation stress Trans., 1984, 15A, 2021±2027.
for 100% of Gmax and ln(XP) = ÿ 9. Here the 24. Hondros, E. D., Proc. R. Soc. A, 1965, 286, 479.
nucleation stress is near zero at 8808K and lower 25. Hondros, E. D. and Seah, M. P., Metall. Trans., 1977,
8A, 1363.
temperatures, thus demonstrating that the degree of 26. Grabke, H. J., ISIJ Int., 1989, 29, 529.
solute segregation can greatly in¯uence the rate of 27. Tatsumi, K., Okaumura, N. and Funaki, S., Grain
nucleation of voids. Boundary Structure and Related Phenomena, Proc.
There has been one systematic study of the e€ect JIMIS-4, 1986, p. 427.
28. Lejcek, P. and Hofmann, S., Acta metall. mater., 1991,
of impurities on the kinetics of hydrogen attack in 39, 2469.
these steels. Sakai and Haji found that 0.01% Sn in 29. Luckman, G., Didio, R. and Graham, W., Metall.
a 2.25Cr±1Mo steel greatly increased the density of Trans., 1981, 12A, 253.
4872 SHEWMON and ANDERSON: VOID NUCLEATION AND CRACKING AT GRAIN BOUNDARIES

30. Viefhaus, H. and Richarz, B., Mater. Corr., 1995, 46, 44. Bika, D., Pfaendtner, J. A., Menyhard, M. and
306. McMahon, C. J., Acta metall., 1995, 44, 1895.
31. Misra, R. D. K. and Balasurbramanian, T. V., Acta 45. Hippsley, C. A., Rauh, H. and Bullough, R., Acta
metall. mater., 1990, 38, 2357. metall., 1984, 32, 1381.
32. Seah, M. P., J. Vac. Sci. Technol., 1980, 17, 16. 46. Bika, D. and McMahon, C. J. Jr, Acta metall., 1995,
33. Grant, N. J. and Mullendore, A. W., Deformation and 43, 1909.
Fracture at Elevated Temperature. MIT Press, 47. McMahon, C. J. and Vitek, V., Acta metall., 1979, 27,
Cambridge, MA, 1965, p. 195. 507.
34. Ransley, C. E. and Talbot, D., J. Inst. Metals, 1959, 48. Lardner, R. W., Mathematic Theory of Dislocation and
88, 150. Fracture. University of Toronto Press, Toronto, 1974,
35. Gaertner, E. and Grabke, H. J., Mater. Sci. Technol., p. 291 (Note: Lardner incorrectly replaced F(Zm) and
1991, 7, 111. E(Zm) with F(m) and E(m)).
36. Tipler, H. R. and Hopkins, B. E., Metal Sci. J., 1976, 49. Beyer, W. H., Standard Math. Tables, 27th edn. CRG
10, 47. Press, Boco Ratan, FL, 1988, p. 356.
37. Tipler, H. and McLean, D., Metal Sci. J., 1970, 4, 50. Shewmon, P. G. and Xue, Y.-H., Metall. Trans., 1991,
103. 22A, 2703.
38. C Inman, M., McLean, D. and Tipler, H., Proc. R. 51. Ciu€reda, A. R., Heckler, N. B. and Norris, E. B.,
Soc. A, 1963, 273, 538. Advanced Material for Pressure Vessel. Vessel Service
39. Thomas, G. and Gibbons, T., 1980 Proc. Seven with Hydrogen at High Temperature and Pressures, ed.
Springs Conf., ed. G. K. Tien et al. ASM, Metals M. Semchysen. ASME, New York, 1982, pp. 53±68.
Park, OH, p. 99. 52. Parthasarathy, T. A., Acta metall., 1985, 33, 1673.
40. Osgerby, S. and Gibbons, T. B., Mater. Sci. Engng, 53. Parthasarathy, T. A. and Shewmon, P. G., Metall.
1992, 157, 63. Trans., 1984, 15A, 2021.
41. White, C. L., Schneibel, J. N. and Padgett, R. A., 54. Shewmon, P. G. and Yu, Z.-S., Advanced Material
Metall. Trans., 1984, 14A, 595. for Pressure Vessel. Vessel Service with Hydrogen at
42. Kraai, D. A. and Floreen, S., Trans. Am. Inst. Min. High Temperature and Pressures, ed. M. Semchysen.
Engrs, 1964, 230, 833. ASME, New York, 1982, pp. 85±92.
43. Hippsley, C. A., Knott, J. F. and Edwards, B. C., 55. Sakai, T. and Kaji, H., Tetsu to Hagane, 1980, 66,
Acta metall., 1982, 30, 641. 1133.

Vous aimerez peut-être aussi