Philosophical Magazine

ISSN: 0031-8086 (Print) (Online) Journal homepage: http://www.tandfonline.com/loi/tphm19

The growth of grain-boundary voids under stress

D. Hull & D. E. Rimmer
To cite this article: D. Hull & D. E. Rimmer (1959) The growth of grain-boundary voids under
stress, Philosophical Magazine, 4:42, 673-687, DOI: 10.1080/14786435908243264
To link to this article: http://dx.doi.org/10.1080/14786435908243264

Published online: 12 Oct 2010.

Submit your article to this journal

Article views: 197

View related articles

Citing articles: 444 View citing articles

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=tphm19
Download by: [Chalmers University of Technology]

Date: 27 November 2015, At: 05:32

The Growth of Grain-boundary Voids Under Streset

By D. HULL and D. E. Rwa d ~ ~ a
Atomic Energy Reeearch Establiehment, Harwell
Downloaded by [Chalmers University of Technology] at 05:32 27 November 2015

[Received January 1, 19591

ABSTRACT
The effect of combined hydrostatic preseure, P , and of b a r i d ternion,
on the rupture time of polycryetalline copper wire in the temperature
range 400' to 600'0 hes been determined. Au the epeoime11.9 broke by
intergranular freoture. due to the growth of voids along grain bounderies.
Preliminary experiments indicated that the voids grow by the nddition of
vacanckm under the action of the applied streee. A theory hes been
developed Blpsuming that failure multa from the growth of void nuolei on
the p i n boundary, whioh requires that the eotivation energy for fdm
that of grain boundary *on
and th8t the rupture t h e . t,. depends only
on ( 0 - P ) . The experiments show that the eotivation energy in cloee to the
expected value, and provided that u is conatant I, egreee with theory.
Changes in u. when ( 0 - P ) in oorntent, affect t, appreciably, end the reedta
suggest that many of the void nuolei are strew induoed.
u,

5 1. INTRODUCTION
INrecent years attention has been drawn to the mechanism of failure of
metala under certain creep conditions. Small voids have been observed
at grain boundaries, particularly those transverse t o the applied atreas
when specimens are tested at high temperatures ; rupture reeulte from the
growth and coalescence of these voids (Greenwood et al. 1964). This
type of failure haa been observed in Magnox cans used for containing
uranium fuel elements, and the present work w w undertaken to try to
understand the mechanism of void nucleation and growth.
There appear to be several methods whereby voids might be nucleated
on grain boundaries. Grain boundary aliding can open up a hole at the
junction of three boundaries or a ledge on a single boundary. Impurity
particles which lack cohesion with the lattice may act aa pre-existing voids.
A dislocation pile-up breakmg through to the boundary might a h be a
suitable nucleus. Of these four possibilities, all except the f h t must have
vacancies supplied to them if they are to grow. Balluf3 and Seigle (1957)
have shown that sufficient vacancies can be produced at transverse grain
boundaries because of the tension acting across the boundary, and euggeet
that this method of void growth is more likely to occur t h m that baaed
on condensation of vacancies produced during strain (Mac& 1956).
If the latter mechanism is operative voids would be produced by both

t Communicated by the Authors.

P.M.

2x

D. Hull and D. E.Rimmer on the

Downloaded by [Chalmers University of Technology] at 05:32 27 November 2015

674

tensile and compressive atrains. However, no vacancies can be generated

at a grain boundary by applying a compreeeive stress across it. It wm
deoided, therefore; to carry out a eeriee of short time creep testa in
which a uniaxial tensile s t r e ~ aand a hydrostatic pressure were applied
simultaneously.
In the first past of this paper we outline a mechanism of rupture baaed on
the growth of void nuolei on a grain boundary by the addition of vacancies
tom the boundary. The effeot of hydrostatic preaeure and uniaxial
tensile stress on the time for failure of suoh 8 system ia odoulated. The
reeulta of experiments deeigned to test this model are given and these are
diSCusSd.

$2.

THEOBYOF RUPTIJREBY VOID GROWTH

The initial experimental results suggested that the mechanism propoeed

by BalluB and Seigle (1967) waa probably the operative one for vacancy
formation. This process must certainly be occurring, r e g d e s e of any
other mode of void growth whioh may also operate, and i t permits an
analysie in terms of parametere which may be estimated with maeonable
precbion.
In this section we ahall derive an approximate expression for the rupture
time, in order to show the sigdicant features of this model. A fairly
rigorous solution for a system having a aimple geometry is given in the
Appendix.
2.1. Estimate of Rupture Time
The specimen is taken to be in the form of a wire and subjected to a
uniaxial streas, u, along ita length and a hydrostatio pmsure P. Void
nuclei of an unspecified nature are aeaumed to lie on the grain boundaries.
po will denote the radiua of a typioal nucleus and a their mean eeparation.
At the temperature at which the testa are being made (400O to 600"o)
grain-boundary diffueion ia more efficient than lattice diffusion in transporting the surface atom of the void to the grain boundary. "his may
be shown briefly by the following calculation :
The ratio of the number of atom traneferred from the void to the
to that by grain-boundary diffusion is given
boundary by lattice -on
approximately by

where D, and D, are reepeotively lattice and grain-boundary diffusion

coefficients, p ia the void radius and Sz the grain-boundary width.
D,. SZ 3 x 10-l6 cm* eec-l for d v e r at 6 0 0 ' ~ (Hoffman and Turnbull
1961), and a a i d a r value is to be expected for copper. At the aame
temperature D, (for copper) 2 x 1O-l' cms sec-l. Even if p = 10-0 om,
which must be an extreme upper limit to its possible size, the oontribution
from lattice diffueion ia only 6% of the total, so that we are justified in

Downloaded by [Chalmers University of Technology] at 05:32 27 November 2015

ckowth of Qrain-bwndary Voids

676

neglecting it. If there ie considerable enhancement of lattice diffusion

under stress, this argument may no longer be valid ; the resultti, however,
does not occur.
seem to indioate that lattioe &ion
In view of this i t is presumed that the atoms are released from within a
ring round each void, whioh lie8 in, and has the same thickness, 62, as the
grain boundmy.
The rate of growth of a void is determined by the gradient of chemical
potential, Vf,in the plane of the grain boundary, since the diffusion flux,j ,
is given by (Herring 1950)

where k = Boltzmann'e constant, T = absolute temperature, = atomic

volume, and Vf is to be evaluated at the void surfme.
In the grain boundary, the potential is given by (Herring 1960)

f = -ao,51
where u,,is the normal tension acting acroo88 the boundary. At a point
well away from any void it may be aaumed that this is roughly equal to
the resultant of the externally applied stresees and ha8 a meximum value
( a- P)on a grain boundary normal to the direction of u.
On the surface of a void f = - 2yQ/p, where y is the surface tension of the
metal.
Hence, aa a rough approximation we may take,

giving

This shows that only those nuclei for which

Po'

2.r

a-P

will grow. For voide smaller than this critical size the applied strese is not
sufficient to overcome the eurfaoe tension and sintering oocurs. Bssuming
that there are some nuclei which can grow,the larger ones will determine
the rupture time and it then is reasonable to neglect 2y/p with respect to
(a-P).
The area of the void surface from which W o n tskee plaoe is 27rpSz,
so that the rate of increase in void volume is given by

dV

(Du8z)(o-P)~27rp"

- N

kTa

dt

If the bubble retains spherical shape by surface diihion, we have

dp
-rv

dt

(Du6z)(u-P)Q
2kTap
2x2

D. Hull and D. E. Rimmer onthe

676

- +.

To obtain the rupture time, f,, this expression must be integrated between
and p
h u m i n g that a Bpo this givea

p=po

flu'

Downloaded by [Chalmers University of Technology] at 05:32 27 November 2015

" ~ ( D , & ) (o P)O

It is shown in the Appendix that for an array of equally sized nuclei
lying on a square lattice the rupture time is given almost exactly by
t, =

. . . (11)
The approximate result differs by a factor of 47T from the first term in the
bracket in eqn. (11). The effect of nucleus size, which waa neglected
above, appeara in the third term and is usually negligible exoept for valuea
of po only alightly greater than 2y/(u- P).
'It would appear that the square array is about the best simple periodio
approximation that can be made to a random arrangement of nuclei,
but in order to estimate the effect of the aaaumed spatial distribution on
the result i t waa decided to oarry out an analyeis similar to that given in
the Appendix, but for a single void f w h g atom into a concentric ring
of grain-boundary having the same initial area, i.e. (aP-vpo2). A closepacked array of such system is an alternative approximation that may
be made to a real grain boundary. UBing a partioular set of parametem
the time for failure on thie model waa about 70% of the breaking time
ilsing the square lattice, The longer time taken by the square lattice
i s due to the fact that the region of grain boundary aseociated with each
void is of a more irregular shape than that of the other model (in which
circular symmetry is ueed), and is therefore the more realistic.
It therefore seem meeonable to uae eqn. (11) for the interpretation of
experimental reeults, with the following proviso. A randomly disperaed
array of mean aeparation a should bring about fracture in a shorter time
than the uniform set considered in obtaining eqn. (1 l), aa some will join
up quickly and increaae the looal stress concentration. A value of u
determined from an experimental result will therefore be somewhat less
than the mean distance between nuclei.
2.2. Theoretical Predictions
The eignificant features of eqn. (11) are: (i) The activation energy of
the procesa is that for grain-boundary diffusion. (ii) The quantitiea u
and P only occur in the form (a- P)so that for identical epecimena at the
mme temperature the breaking-time is a function only of (a-2'). If
( 0 - P) 2y/po the three terms on the right-hand side of eqn. (1 1) are in the
approximate ratio

Takmg y = l@dyne cm-I, a - 2 x 1 0 4 cm, the first term dominates

Qrozdh of Qrain-boundary

Void.9

677

Downloaded by [Chalmers University of Technology] at 05:32 27 November 2015

if ( u - P ) 6~x 10' dyne cm-L= 750 p.s.i. In the experiments to be

described this is true, although significant contributiona come from the
other two terms. Thus the shape of the t, versus ( u - P ) curve is
approximately hyperbolic down to (a- P)= 2y/p0, at which point the third
term becomes infinite. Physically this is the point at which the applied
stress is balanced by the surface tension of the void nucleus, 80 that the
voids never grow. Figure 1 shows the expected variation oft, with (u- 2').
Fig. 1

I
I

I
I
I

$lpo

(Z}

Theoretical Variation oft, with (u- P ) .

$ 3 . EXPERIMENTAL
DETAILS
T h e tests were made on a relatively impure copper (99.8%) which

mntained small copper oxide inclusions in short stringers along the direction
of drawing. The specimens were in the form of wire 0.020 in. diameter,
which waa etched in dilute nitric acid to reduce the diameter to between
0.013 and 0.010 in. along a 1 in. gauge length. Copper beede were welded
onto the ends of the wire specimens to hold them in the apparatus. The
specimens were annealed at 600'0 for one hour before testing. Combined
hydrostatic and uniaxial tensile streases were applied using the apparatus
illustrated in fig. 2. T h e tensile streaaee were obtained using a spring D
attached to one end of the epecimen C and held in a silica tube B, approximately 12 in. long. This aeeembly wae iliserted inside a pressure bomb A
(Fennell et d.1958) with an internal diameter 6/16 in., capable of holdmg
pressures of 6000 p.s.i. for a few h o w . All the testa were made in argon.
The time to h t u r e waa meaaured by meana of an electric clock circuit

D.Hull and D. E.R i m e r on the

678

through the specimen. The preeeure bomb waa inserted in a furnace tube
and temperatures between 400' and 600'0 maintained to +3Oc. The
creep strain was not measured during the test, but measurements on
fractured specimens showed that the creep strain waa lees than 3 to 5%.
'she specimens were examined metallographica,lly after fracture by
sectioning along their length followed by polishing on fine alumina.

Downloaded by [Chalmers University of Technology] at 05:32 27 November 2015

Fig. 2

Diegrem of epperetue dfor applying simultaneous hydrostetio preeeure

8nd b 8 i d tension.

9 4. EXPERIMENTAL
OBSEBVATIONS
4.1.

Ee'd

of Arm-hydrostatic Cmpre&ve Streaeee

An experiment waa carried out to determine whether voids were formed

in specimene tested with o = P . In this cam there is no resolved strees
along the tensile axis, but there are compressive streaaes normal to this
axis. The apecimena were teeted a t 480"c with a tensile stress sufficient
to break the specimens in less than 20min (without the hydrostatio
preaaure). The apecimena did not fracture after 8 hours and subsequent
micro-examination revealed that no voida had formed.
4.2. Hetallography

It was confirmed that all the specimens festad had ruptured by the
formation of voids on grain boundaries. Figure 3 (Pl. 77) shows the
general appearance of a specimen, the majority of voids having formed on
boundariee normal to the applied strew. A few stringers of oxide particles
are also visible. Figure 4 (Pl. 77) is a high magnification photograph
which shows clearly the voids along grain boundaries. Voids do not form
on twin boundariea, even when the boundariea are transverse to the
applied stress. T h e photographs show that the voids in fractured specimens
are cloae together and often coalesce. The e k e and number of voids
varied with tbe applied stress. This is illustrated in the photographs
(figs. 5 and 6 (Pl. 78)) of two specimens teeted at the tame temperature,
the f h t streaaed at 4730 p.8.i. and the second at 3000 p.6.i. At the high
stress a large number of small voids have formed compared with a smaller
number of larger voids at the low streaa.

G-rowth of Gain-boundary Vo'ids

87 9

4.3. Activation Energy of the Proceaa Leuding lo Rupture

A series of specimem was tested at a constant uniaxial tension and the
time to fracture meaeured for a range of testing temperatures between
370" and 490Oc. The results, which are plotted in fig. 7, give an activation

Downloaded by [Chalmers University of Technology] at 05:32 27 November 2015

Fig.7

Variation of log 1, with 1/T for a=4730p.e.i., P=O.

energy of 26 000 k 3000 calslmole. This figure comparea well with the
activation energy for grain-boundary difFwion determined for eilver
(20200 csls/mole) (Hoffman and Turnbull 1961).

D. Hull and D. E. Rimmer on the

680

Downloaded by [Chalmers University of Technology] at 05:32 27 November 2015

Fig. 8

PIKSSURL

P,

p.nJ.

Variation of 1, with P. Experimental points with theoretical C U T V ~ B .T=41Ooo.

Fig. 0

I
4000

5 0 0
TENSION

9 p.s.1.

boo0

Variation of t , with (I for constant values of ((I - P).

(Taken from smoothed experimental resultti at 410% )

Growth of Gain-baundury Voids

68 1

4.4. Effect of Strem (u)and Pressure ( P )at Conslanl Temperature

'rhe variation of the time to rupture (t,) with hydrostatic pressure for

Downloaded by [Chalmers University of Technology] at 05:32 27 November 2015

a number of values of u waa obtained at 410"c and the results are plotted
in fig. 8. The variation of rupture time with stress for a number of values
of ( u - P ) has been obtained from these results and is plotted in fig. 9.

Equation (11) predicts that the rupture time &odd be dependent only
on (u- P),i.e. a variation of u should not affect the rupture time providing
( 0 - P ) remains constant. As shown in fig. 9, the rupture time alters
appreciably with change in u and this is discussed below in 9 6.

5 5.

DISCUSSION

I n Q 4.1 an experiment designed to test the theory of void growth by

vacancy condensation under strain waa described. It waa shown that
although a certain stress applied in tension would easily break the specimen,
when applied in compression it could not do EO. Further, in the latter cme
no visible voids were produced. This suggests that the vacancies for
void growth are produced by the tension across the grain boundaries
rather than deformation inside the grain. It seemed reasonable therefore
to try and correlate the experimental results with a theory baaed on the
grain-boundary diffusion model from which eqn. (11) was derived.
The activation energy of the processes leading to rupture in the temperature range used corresponds with that for grain-boundary diffusion
rather than lattice diffusion. This supporte one of the basic wumptions
of the the0ry that the voids grow by the dihsion of atoms from the void
surfrtce along the grain boundary.
Examination of the reaulta in fig. 8 compared with eqn. (11) shows that
the theory cannot explain all the experimental results, since if (u- P)were
the only stress factor affecting t, then for a constant value of (a- P),e.g.
3000 p.s.i., the breaking time should be constant, whereas at u = 4730p.s.i.
t, = 2.2 x 10' MC, and at u = 6060 p.s.i., t, = 1-25x 10' E ~ C . However, we
shall consider first how eqn. (11) agrees with the effect of ( 0 - P ) on t,
providing u is constant, and then examine the variation of t, with u.
Figure 8 shows some observed values oft, a8 a function of P for a number
of values of u. Equation ( l l ) , with which it is to be compared, contains
two unknown^, a and p o . The appropriate value to use for po is difficult
to determine. It is however possible to find a lower limit for it because it is
known that l , + a as (a- P)+2y/p0 (EM Q 2.2). Figure 8 shows that f, is
still finite when (u- P)= 2000 p.6.i. It is unlikely that this, the smallest
experimental value, is by coincidence very close to the critical one. If we
take 1600 p.6.i. = 108 dyne cm-2 aa the critical value

which is reasonable. Using this figure, any one of the experimental results

682

D. Hull and D. E. Rimmer on the

Downloaded by [Chalmers University of Technology] at 05:32 27 November 2015

giving t, for specified values of u and P will enable a value for a to be calcula-

ted by eqn. (11). Thie may then be used to predict valuee of 1, for comparison with the remaining experimental values. If results for which P = 0
are used to determine a, the term containing po is small and the assumptions
made in ita eetimation have little effect on the caloulated value of a.
It waa found that for each value of (I, a different value of a is needed
ta fit the experimental point at P = 0. However, if for any given u,the
appropriate value of a is ued to derive the t,(P) relation, this agrees well
with the experimental pointe. The full lines in fig. 8 are the theoretioal
ourves, using the following valuee of a :

I I-.

Table Giving the Distance Between Voida, a, for

Values of Applied Teneion, u

) * ; ;=

6400
6ooo

1-1

Growth of Qrain-bvundary Voids

6 83

Downloaded by [Chalmers University of Technology] at 05:32 27 November 2015

In concluaion, the experiments show that the grain-boundary voids

which are frequently observed to be the wu88 of oreep failure, grow by the
diffusion of surface atom to the surrounding grain boundary under the
action of the applied stress. It appears that many of the void nuclei are
indnoed by the streas, but their exact nature is uncertain.

ACKNOWLEDQMENTS
It b a pleaeure to thank Rofeesor A. H. Cottrell, Dr.W. M.Lomer and
Mr. R.S.Barnea for many useful discussions, and Measrs. D. C.Wynne and
B. Hudson for help with some of the experimental work.
REFEEENOES
B A ~ . , LR.
~ W.,
, and SEIOLE,
L.L., 1967, Ada Met., 5,449.
CHEN,C. W., end &cm..m, E. S., 1966, Acta Met., 4, 666.
Fmmm,, J., boom, N. H., and BARNES,R. S., 1968, Meld Tram., 25,
332.

G m s , R. D.,1956, Act0 Md.,4,98.

GREENWOOD,J.N.,Mn.r.RR,D.R.,and S m ,J. W.,1964, A d a M d . , 2,260.
-0,
C., 1950, J . appl. Phya., 21,437.
HO~MAN
R., E., and TOBNBm, D., 1961, J . appl. Phy8., 22, 634.
Ma-,
E. S., 1966, Tram. Amer. I&. min. (d.)
Zngr.9, 206, 106.

APPENDIX
The Rupture Time for a 8qwzre Array of E q d y Sized NotcIei
Figure 10 shows a square array of spherical voids with repeat distance a
and having grown to a radius p, lying on a grain-boundary normal to the
direction of (I. The centre of a void is taken aa the origin of reotangular
and polar axes.
In order to calculate Vf at the void surface it is necessary to define the
problem more exactly than in 5 2, and we shall make the following additional
aesumptions :
(i) The ring of void surface lying in the grain boundary, which acts
aa a source of diiuaing atoms has a radial width w, about equal to that of
one atomio layer (see fig. 10). The atoms are released uniformly from
within tbis volume.
(ii) The atoms are deposited uniformly over the grain boundary, so
aa to minimize lattice strains.
(iii) The void retains its apherioal shape throughout its growth. It
must be pointed out that whilst there is mechanism available (i.e. surface
difhion) for doing this, a sphere is not the equilibrium shape for a void
under the conditions that have been postulated. Owing to the presence
of the other voids the natural shape of ita grain-boundary cross section
is a deformed circle with bulges towards the oornem of the unit square.

684

D. Hull and D. E. Rimmer on the

Downloaded by [Chalmers University of Technology] at 05:32 27 November 2015

These become more predominant &B the void grows. In order to retain
a circular cross section it is necessary to impose small restraining forces
on the void. In view of the other approximationsnecessary in this problem
i t has been thought permissibleto aasume that the bubbles remain spherical,
yet ignore the forces holdmg them to this shape. Although the effect
of this on the h a 1result is unimportant, it willbe seen later that it produces
some slight inconsistencies in the mathematics which have been accepted
in view of the great simplification in the working that haa resulted.
Fig. 10

-UNIT

CELL

Cross section through voids lying in the

I
0

plane of the grain boundary.

Let us consider a unit cell centred on the origin and defined by

1.1 < 4 4
IYI < !la-

Vy is a meaaure of the strength of a given point aa a source of or sink for

atoms. In view of assumption (i) and (ii) above it is determined (apart
from a scaling factor)at all points inside the unit cell since we have,
Vzf=O when O c r < p ,
Vy=A when p c r c p + w ,
(1)
V f = B when r > p + w ,
where A and B me related by the equation,
B(u"7+)+Aw*2?rp=O
. . . . . * (2)
as the total number of atoms is constant.

,1

.....

Growth of Gain-boundary Voids

But

685

Vy has periodicity a in x , y, and may be written in the form

+w
27r
Vy= 2 2 C,,,,,oosa(mz+ny)
..
+OD

m--

w m--

(3)

Downloaded by [Chalmers University of Technology] at 05:32 27 November 2015

where

from eqm. (1).

b=Q.w9p,

C,,,,=(Bor 0 ) -

+ns)u*

27rp(ma

a(mz+ nz)ur

27rApw J, 27rp(mL+ nn)ua)

a8

....

(4)

where the h t term is B if m =n = 0 and zero otherwise.

Hence
C
from eqn. (2). Otherwiee,
-Cmn
=-

2np(mr n2)l'*)

5')Jo (27p(mSa+na)uz
. . .

(6)

The most general integral of eqn. (3) which has the required symmetry is
f = - - a'
27r
cmn
cos-(mx+ny)+K
(6)
~rre
(ma+nz)
a
where K is a conatant, obtained from the condition,

. .

f=

-u,a.

. . . . . . -

*
(7)
a,,will vary over the unit oell owing to the streaa concentration set up round

the void. However, the total force over the unit cell due to the applied
Btreea and pressure must be (u-P)aa.
Hence integrating eqn. (7) over a unit cell and allowing for surfaoe
tension,
a[%py-(u-P)az]= I ~ : l + * f d z d y - I ~ I C f i d B d r .
-to

..

(8)

On physical grounds we may expect f to have the value - 2yQ/p at all

pointa ineide the void. If it were calculated from eqn. (6)thia result would
not be obtained becauee the restoring f o r m round the void have been
neglected. - (2yQ/p) will iu fact be the mean value inside the region and
for the purposes of carrying out the second integration in eqn. (8) we shall

086

D. Hull and D. E. Rimmer on the

take it aa constant. The origin is one of the positions which haa the mean
value, a fact which will be used later. Hence
C?[2npy- (U- P)ua]=K d

Downloaded by [Chalmers University of Technology] at 05:32 27 November 2015

giving

2yQ .mpa
+P

K = - O(U-P).

Beeuming a radial flow of atom at the void eurface which is reaaonable

if p a the diffusion flux is given by,

j = - A'(
f)
kTQ 5 r-p+w*
If we let 1-0
and evaluate (a!/&-) at r = p it must be remembered t h a t
(af/&-)is now diecontinuous at this point (see fig. 11)and the value required
is the maximum (numerically) which is equal to twice that obtained by
computing the Fourier s u m at r = p .
Fig. 11

_c

Approximate radial variation of grsdient of chemical potential.

Qrowth of Qrain-boundary Voids

687

The number of atom, per second, coming out of an elementary area

pddSz of void is,
2~,pdes~

af

Downloaded by [Chalmers University of Technology] at 05:32 27 November 2015

kTQ ( 5 ) ;
Therefore, the r a b at which atom leave the void ia given by,

The coefficiente Cmn are not known, but C,,,,/B is given in eqn. (5).
However, (LB strrted earlier the value off is known at the oentre of the void.
Substituting
f-

when z==y=O

into eqn. (9) givee

-=

dt

kT

where

The rate of change of void radius is,

dp=- R

.-=
dN

27rn(D,Sz)(u-P - 2 y / p ) S

dt h p '
dt
kTap
The time to h t u r e ie obtained by integrating this expression between
p =po and p = &z. S ie a function of p/a only, and haa been daulated for
a set of values in the expeoted range of integration. These suggest that
the approximation S= 1 may be used BB it only deviates sigdhantly from
this value for a short time just before the voide oodeeoe.
Thie givee,

where t, =time to failure.

On integrating and wuming po4a, we obtain

..

(11)