You are on page 1of 9

Materials Science and Engineering A 387–389 (2004) 576–584

A model for creep based on microstructural length scale evolution


Glenn S. Daehn∗ , Holger Brehm1 , Huyong Lee, Byeong-Soo Lim2
Department of Materials Science and Engineering, The Ohio State University, Columbus, OH 43210, USA

Received 9 September 2003; received in revised form 23 December 2003

Abstract

This paper seeks to quantitatively link recovery and plastic deformation to develop a model for creep. A simple approach with one length
scale coarsening equation is postulated in this paper to provide a descriptive and unified framework to understand recovery. We propose
that recovery is a general dislocation-level coarsening process whereby the length scale, λ, is refined by dislocation generation by plastic
deformation and is increased concurrently by coarsening processes. Coarsening relations generally take the form:

d(λmc ) = KR(T) dt

where R(T) is the rate equation for the fundamental rate controlling step in coarsening, K a free constant, dt a time increment and mc is the
coarsening exponent. Arguments are presented that mc should be in the range of 3–4 for dislocation or subgrain coarsening. The coarsening
equation postulated is consistent with the compared data sets in the following ways: (i) temporal evolution of one length scale λ; (ii) temperature
dependence of recovery rate; (iii) adequacy of single parameter in the proper description of strength change. The coarsening equation is coupled
with standard arguments for modeling plastic deformation. Combining these we can easily justify the form of the empirically derived Dorn
creep equation:
 n
γ̇ τ
=B
D(T) µ
where the mobility of the recovering feature, R(T) should typically scale with self diffusivity, D(T), and the value of the steady-state creep
exponent, n is 2 + mc − 2c where c is a constant related to dislocation generation that should be in the range of 0–0.5. Hence, this approach
predicts creep as being controlled by self diffusion and that the steady-state stress exponent should be on the order of 4–6. All the parameters
in the creep model can be estimated from non-creep data. Comparison with the reported steady-state creep data of pure metals shows good
agreement suggesting recovery modeled as coarsening is a fundamental element in steady-state creep.
© 2004 Elsevier B.V. All rights reserved.

Keywords: Creep modeling; Microstructural evolution; Recovery

1. Goals and motivation compared at the same homologous temperature. In 1962,


Oleg Sherby published a paper [1] showing that if one plots
There is a remarkable commonality in the way materi- the steady-state creep strain rate divided by diffusivity ver-
als plastically deform. At low temperature almost all sim- sus creep stress normalized by elastic modulus, there is a
ple cubic metals have remarkable ductility and strong strain remarkable collapse of a large amount of experimental data.
hardening. Also, if stress is normalized by elastic modulus, Twenty-one different metals or phases are represented on
many materials show very similar stress–strain curves when the diagram. Specifically:

∗ Corresponding author. Tel.: +1-614-292-6779; • Scaling by diffusivity removes the temperature depen-
fax: +1-614-292-1537. dence, and data from multiple temperatures collapse to a
E-mail address: daehn@er6.eng.ohio-state.edu (G.S. Daehn). single curve.
1 Presently at Fraunhofer Gesellschaft, IWM Freiburg, Freiburg, Ger-

many.
• All the curves are relatively straight between stresses of
2 Visiting Professor from Sungkyunkwan University, Suwon, Republic 10−5 E and about 5 × 10−3 E, where E represents elastic
of Korea. modulus.

0921-5093/$ – see front matter © 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.msea.2003.12.082
G.S. Daehn et al. / Materials Science and Engineering A 387–389 (2004) 576–584 577

• All the curves roughly follow the form: same way with strain and time with a fixed factor between
ε̇  σ n them. Under these assumptions the model that follows can
≈K (1) be equivalently applied to dislocation strengthening or sub-
D E
grain strengthening without comment as to which model for
where the K values are relatively constant and n is typi-
strength is ‘correct’.
cally in the range of 4–6.
In either event we assume that a single parameter can be
This degree of data collapse is remarkable especially be- used to describe structure. That parameter could be either
cause the materials represented have a wide range of crystal length scale, λ, λs , or the dislocation density (here expressed
structures (f.c.c., b.c.c., h.c.p.), stacking fault energies, slip as line length per volume), ρ. From a simple dimensional
systems, etc. argument the average slip distance can be related to the
The vast majority of the models in the literature that seek inter-dislocation spacing as
to understand the creep behavior of metals (in particular the g
stress exponent) are based on specific mechanistic models λ= √ (2)
ρ
and detailed assumptions about how dislocations are gener-
ated, moved and are annihilated. Because a nearly common The factor g will depend upon the spatial distribution of the
pattern is seen in many materials with many different mi- dislocations. A lower bound on g is set by an array of dis-
crostructural and dislocation-level details, it seems clear that locations that are all parallel, lying in the same direction.
there is something going on that is more general. In this case g is unity. For a more typical nearly isotropic
The purpose of this model is to understand the creep be- example, the dislocations may lie along the √ x, y and z direc-
havior of metals in a very general without reliance on spe- tions in square array. Here, the g value is 3. When we use
cific mechanisms. In this spirit, assumptions will be kept as this relationship to describe yield, it will be the dislocations
simple and general as possible (at great risk of over simpli- with the longest inter-obstacle spacing that will tend to be
fying the situation). The model is intended to apply to sim- released, also dislocations will tend to cluster. For these two

ple, single phase, nominally pure metals at temperatures of reasons, reasonable g values are in the range of about 3
about 0.3Tm and above (where Tm is the absolute melting to 20. A standard value of g = 4 will be chosen presently.
temperature). With these assumptions, dislocations will tend With this single parameter description of structure, we
to spend much time at relatively robust pinning points. And shall model creep from the three elements that are undeni-
once they are released they will move quickly to the next ably present: (1) release of dislocations from obstacles; (2)
set of pinning points. Barriers of small activation energy refinement of the dislocation network and subgrains due to
(such as Peierls barriers) will not significantly contribute to dislocation motion; (3) recovery (coarsening) of the dislo-
strength. cation structure. In describing these processes we will be
much more guided by data than any pre-supposed theories.

2. Fundamental processes and structural description


3. Dislocation release
Here, we want to use the simplest defensible description
of the material microstructure. Obviously it is heterogeneous We assume dislocations are held by discrete pins and when
at several length scales. The relative roles of grain bound- they are released from a pinning point they glide easily over
aries, subgrain boundaries and forest dislocations in material a distance on the order of λ. Implicitly assumed through-
strengthening have been debated extensively. As the present out this treatment is that the material strength is derived
focus is on behavior at relatively high homologous temper- wholly from such pins and interactions from strengthening
atures, and it is known that above about 0.4Tm , fine-grained mechanisms such as Peierls stresses for solute drag do not
materials are generally weaker than coarse grained materi- contribute to strength. Thus, this model is not expected to
als. Therefore, grain size strengthening will not be consid- apply at low homologous temperatures where very low en-
ered further here. The issue with subgrains is more compli- ergy barrier interactions contribute significantly to strength.
cated. Subgrains are known to refine with increasing levels Two types of models for dislocation release are considered
of stress and strain and fine subgrain sizes also correlate in the present treatment. The Arrhenius treatment is more
with higher material strength. However, it is quite difficult to versatile and correct, however, it does not easily produce
separate the possible separate roles of subgrains and forest a clean analytical form. The athermal treatment gives very
dislocations, and this topic has been discussed extensively similar results and also gives a form that is easily analyti-
[2]. In this treatment we assume that strength comes from cally tractable.
dislocation density rather than from subgrain size. However,
we identify two unique length scales. The first, λ, repre- 3.1. Arrhenius treatment of dislocation release
sents the average inter-dislocation spacing. The second, λs ,
is the average subgrain size. Typically λs is about 3–30 times A pinned dislocation puts a force, f, on an obstacle: f =
greater than λ and we will assume that both will scale the τbλ, where τ is the externally applied shear stress and b is
578 G.S. Daehn et al. / Materials Science and Engineering A 387–389 (2004) 576–584

the magnitude of Burgers’ vector. At each obstacle, at each other simple pure metals. In using this form, there should
local vibration, there is a probability the dislocation will be a slow decrease in s with increasing temperature.
break free from the obstacle and glide. That probability, Prel
is given as
  4. Dislocation accumulation
−G∗
Prel = exp (3)
kT
Asbhy [5] produced a compelling schematic plot of dislo-
where the energy barrier G∗ is given in the form commonly cation density as a function of plastic shear strain. His plot
used by Kocks et al. [3]: was intended to schematically indicate alloy behavior, and
  p q was based on the copper single crystal data of Basinski and
∗ ∗ f
G = F 1 − (4) Basinski [4]. Since this time this plot has been schematically
k̂ reproduced other places and interpreted as more general than
The terms F∗ , k̂, p and q are all essentially parameters that originally intended. This plot also provides the motivation
describe a given barrier in a single material. F∗ is in en- for examining if there is a general pattern for dislocation
ergy terms the unstressed size of the barrier that must be sur- accumulation due to plastic flow.
mounted for dislocation release. This can be easily normal- It is appealing to assume that free dislocations become
ized to varied materials by expressing it as F ∗ = s1 µb3 . immobilized after moving some distance related to the slip
This has been used previously as a normalizing scheme distance, λ, after which new dislocations must be formed to
where s1 values on the order of 0.02, 0.25 and 10 correspond enable further deformation. In such a case we can write:
to barriers such as solute interactions, dislocation junctions dρ k
= = Mρ0.5 (7)
and precipitate interactions, respectively, for example. The dγ λ
dislocation obstacle can be overcome without any thermal
where k or M represent material constants which must be fit
activation if the force on the obstacle exceeds the athermal
to either a model of dislocation generation or experimental
breaking force, k̂. Again k̂ is normalized to the intrinsic
data.
materials properties. Here, we take k̂ = s2 µb2 . With this
During plastic deformation there is also dynamic recov-
normalizing scheme materials can be described in a rather
ery that will reduce dislocation density, essentially while it
generic way with the free parameters being s1 , s2 , p and
is generated. From a practical point of view we can re-cast
q. Variations in these parameters can be directly correlated
Eq. (7) as being the total change in dislocation density (in-
with variations in the micromechanics of flow.
cluding dynamic recovery) due to a plastic strain increment.
When viewed this way, that there is considerable experi-
3.2. Athermal treatment of dislocation release
mental data (which was reviewed nicely by Gilman [6]) that
shows for many polycrystalline systems dislocation density
Eq. (4) represents a well-founded but analytically inconve-
accumulates linearly with strain and there is a similar pattern
nient way of describing a material’s flow stress as a function
for many metals over a range of temperatures. A compila-
of temperature and obstacle properties. For computational
tion of such data is shown in Fig. 1. A generic differential
convenience we will find it useful to introduce an athermal
form that will fit both this observation as well as Eq. (7) is
description for flow. Basically we assume that when the force
on an obstacle exceeds some athermal breaking force it will dρ
= Mρc (8)
be overcome. We can derive the material flow stress as dγ
sµb where M is the multiplication coefficient and c is the breed-
τf = (5)
λ ing exponent. Using this form and assuming a material has
Here, s represents a single-parameter description of an initial dislocation density ρ0 at a strain of zero, the dislo-
the junction strength. In fact this parameter should be cation density as a function of strain (in the absence of any
temperature-dependent, but it is very weakly temperature- losses due to static recovery, that are discussed later) can be
dependent, relative to diffusivity, for example. If dislocation written as
density, ρ, is used as the structural metric instead of λ, the ρ = [Mγ(1 − c) + ρ0
(1−c) 1/(1−c)
] (9)
equation can be re-written as
  As can be seen in Fig. 1, the data from many simple metal-
s √ √
τf = µb ρ = αµb ρ (6) lic systems at relatively low homologous temperatures clus-
g
ters nicely, and several systems can be fit using M = 2 ×
The latter equation is Taylor’s equation for plastic flow resis- 1015 m−2 and c = 0. It is impressive that the data col-
tance. There is a considerable amount of data confirming the lapses this neatly, especially since the systems studied in-
validity of this equation for f.c.c. metals at low homologous clude many very different metals examined at temperatures
temperatures. The value of α is found to be approximately ranging from room temperature to creep conditions. The ex-
0.3 [4]. Also, the equation quite reasonably fits b.c.c. and ponent of c = 0 will lead to parabolic hardening. We assume
G.S. Daehn et al. / Materials Science and Engineering A 387–389 (2004) 576–584 579

Fig. 1. Compilation of data for dislocation density as a function of imposed plastic strain for a number of annealed metallic systems. The line marked
‘assumed density’ shows the equation used in the present model. The original data for this figure comes from Refs. [7–12].

that this is the native behavior of the material. Of course and found that stress does not seem to change the character
Stage II hardening (which is not robustly seen in polycrys- of the recovery, but it can change its rate. They found re-
tals, except at very low homologous temperatures) demands covery rate increases monotonically with stress and in their
linear hardening and would require an exponent, c of 0.5. study could be as much as four times as fast as stress-free
Ashby [5] has discussed how geometrically necessary dislo- recovery. Presently we assume that only static recovery is
cations will accumulate linearly with imposed strain (as the responsible for the softening of the material with increasing
bulk of the present data indicate). time at temperature, but note that applied stress may play a
In the present model, dislocation accumulation is assumed role in speeding the process. The materials community re-
to take place in a manner that is described by Eq. (8) where ally does not have a singular basic approach to recovery. In-
the values of M and c seem to cluster around 2 × 1015 m−2 stead distinct dislocation-based models that are appropriate
and 0, respectively, for many simple metals. We do not wish to the given situation are usually used [14]. Here, we will
to present any theories for this; instead we will rely on the assume that dislocations exist in a network structure and this
trends seen in the experimental data. network will coarsen with time in a self-similar way. There
may also be a subgrain structure and this is assumed to have
a size that scales proportionately with the dislocation link
5. Recovery (substructural coarsening) length, λ. With these assumptions one can treat recovery as
a generalized coarsening process and leverage the advances
Clearly at elevated temperature, dislocation density will made over the past few decades in this area [15].
decrease with increasing time and temperature. Also sub- Again λ represents the characteristic microstructural di-
grains will coarsen. Plastic deformation may well have a de- mension. For simplicity and brevity we will essentially ex-
cisive role in hastening these recovery processes. Raymond tend the derivation of grain coarsening developed by Burke
and Dorn performed an especially careful study of the effects and Turnbull [16]. Let us start by summarizing their deriva-
of stress on the recovery kinetics of aluminum in creep [13] tion. They assumed that for a self-similar array of grains the
580 G.S. Daehn et al. / Materials Science and Engineering A 387–389 (2004) 576–584

pressure difference from one side of a boundary to another concentration is given by Xv . Summarizing for transport by
scales with σ/λ, where σ represents surface tension and λ bulk diffusion:
is the grain size dimension (because this is dimensionally dXv
the equation for gas pressure in a sphere with surface ten- M ∝ Jvac ∝ D (12)

sion, σ) and with regard to kinetic mechanism Burke and
Turnbull assumed differential pressure drives boundary mo- Here, we recognize that the mobility, M, term will have both
tion in at rate proportional to the boundary mobility, M. Or temperature and length scale dependence. If we substitute
specifically: this into the relations above, the coarsening exponents will
σ  all rise by 1. Hence, for coarsening of surface area per vol-

= A MP = A M (10) ume with bulk diffusion control, the coarsening exponent
dt λ will rise from 2 to 3. This is the reason Ostwald ripening
Upon integration one obtains the classical equation for has the well-known coarsening exponent of 3 [20,21]. From
parabolic grain growth: this point on we will use the symbol M to designate viscous
mobility of a boundary or line defect (motion by short-range
λ2 − λ20 = AMt (11)
diffusion across the boundary) and R(T) will be generally
This equation will apply equally well to subgrain coarsening used for the rate controlling step in the coarsening process,
by boundary mobility, where the rate of boundary migration including both diffusion and viscous mobility.
is proportional to the differential pressure on it. We may take this one step further and consider pipe dif-
In the present treatment we are at least as interested in the fusion. Here Deff = ρDp . Since ρ scales with λ−2 , a pro-
coarsening of dislocation arrays as in grains or subgrains. cessing coarsening by pipe-diffusion transport will have an
For the case of coarsening a planar array of dislocations exponent 2 higher than that through bulk diffusion.
(such as those that may make up a low angle boundary), it Based on all of this we can establish a general coarsening
is the force/length on the dislocation line that is the driv- equation:
ing force for motion. Again the driving force will scale with
d(λmc ) = KR(T) dt (13)
T/λ, where T is the dislocation line tension and λ is the mi-
crostrucural length scale (i.e. proportional to the mean radius The coarsening exponents take on values between 2 and 6
of curvature for a self-similar system). One can follow the depending upon system morphology and mechanism. The
same approach as above to again develop parabolic network rate controlling step R(T) can be either viscous mobility or
coarsening. diffusion limited. The term K must be determined from ex-
Now let us consider the dimensionality of a 3D network of periment or a theory based upon specific mechanistic as-
lines, such as the array of dislocations and nodes described sumptions. The possible coarsening/recovery mechanisms
by Öström and Lagenborg [17,18]. Here, the driving force are summarized in Table 1.
is reduction in line length per volume, instead of length per
area as above. As will be shown later in a formal dimensional 5.1. Experimental assessment of the coarsening model
analysis approach [19], the driving force now scales with
T/λ2 . Now if dislocations move by a viscous mechanism There have been a number of experimental studies on the
equation (11) will have coarsening exponent, of 3 instead of recovery of simple metals since the 1950s. Many studies
2. have shown at homologous temperatures above 0.4 or so, the
To this point we have described the coarsening of three activation energies for coarsening are experimentally indis-
different structure types: (sub)grains, planar dislocation ar- tinguishable with those for diffusion (for a review see [22],
rays and volumetric dislocation arrays. The former two have also [13,23]). However, there have been relatively few stud-
a coarsening exponent of 2 while the latter one has a coars- ies where substructural length scales or dislocation densities
ening exponent of 3. Now let us change the kinetic path and have been measured directly to test the coarsening exponent
assume defect mobility controls the rate of coarsening. Now or temporal form of Eq. (13). If Eq. (13) is plotted on log–log
assume long-range transport will control the mobility of the axes, it should take on a form like that as shown in Fig. 2. At
defect (vacancy). Defect velocity will scale with vacancy very short times the initial length scale is present in the ma-
flux that is described by Fick’s first law, where the vacancy terial. Also by physical constraints, this cannot be less than

Table 1
Appropriate terms for coarsening exponent, mc , and rate controlling rate, R(T), for the generalized coarsening equation d(λmc ) = KR(T) dt
Rate step/geometry Grains/subgrains 2D dislocation array 3D dislocation network

Viscous motion mc = 2, R(T) = Mb (T) mc = 2, R(T) = Mg (T) mc = 3, R(T) = Mg (T)


LR lattice diffusion mc = 3, R(T) = Dl (T) mc = 3, R(T) = Dl (T) mc = 4, R(T) = Dl (T)
LR pipe diffusion mc = 5a , R(T) = Dp (T) mc = 4b , R(T) = Dp (T) mc = 6, R(T) = Dp (T)
a Assumed that the dislocation spacing will scale with (sub)grain size, both proportional to λ.
b Pipe diffusion increases mc by 1 vs. bulk diffusion because area fraction of dislocations scales with 1/λ.
G.S. Daehn et al. / Materials Science and Engineering A 387–389 (2004) 576–584 581

Fig. 2. Schematic behavior of Eq. (13) including the lower bound based
on the fact that λ must be significantly larger than b and the upper
bound based on the observation that even at very long annealing times
dislocation density seldom drops below 1010 m−2 .
Fig. 3. Oden’s data in the form of characteristic inter-dislocation spacing
as a function of the product of diffusivity and time. The line shown
a few Burgers’ vectors. At very long times it is noted that indicates m = 3 and a K value of 9.1 × 10−7 m.
materials will coarsen to a ‘frustration length’. Dislocation
densities seldom become less than about 1010 m−2 , even af-
ter extensive annealing. Also studies of grain growth show Cermak for a similar alloy is used to normalize with time
that at very long time grain growth will stop with a grain di- [25] and sets R(T). The specific diffusion coefficient used
ameter that is typically related to sample size. These consid- is 6.4 × 10−18 m2 /s. Fitting to the data yields a K value of
erations place the upper and lower bounds shown in Fig. 2. 9.1 × 10−7 m.
Between these bounds Eq. (13) is expected to be apparent. The recent work of Huang and Humphreys [26] allows ex-
Two especially appropriate studies have been carried out amination of both the temporal and temperature dependence,
that allow examination of the temporal and (in one case) but concentrates on subgrain growth instead of network
thermal dependences of Eq. (13). First Oden et al. [24] used coarsening. For a crystal deformed 70% at room tempera-
TEM to investigate the diffusional coarsening of a disloca- ture and recovered at varied temperatures they find coarsen-
tion network developed by creep after the stress is removed. ing exponents between 4 and 2.8 for temperatures between
This study was carried out in a 20% Cr–35% Ni stainless 250 and 400 ◦ C. Their original data is plotted in Fig. 4 in
steel at 700 ◦ C after creep at 127 MPa. Fig. 3 shows their log–log form with points added at very short times to indi-
data replotted in the form of Fig. 2 using g = 4 to ob- cate the original subgrain size of the material of 0.9 ␮m. The
tain λ from dislocation density. Here, the diffusion data of data are also plotted with the product of time and diffusion

Fig. 4. Data of Huang and Humphreys [26] plotted on log–log axes both with time and the product of time and diffusivity. Notice that when normalized
the data collapse nicely and are fit reasonably well with Eq. (13) with a value of m = 3 and K = 4 × 10−5 m, as is plotted on the graph.
582 G.S. Daehn et al. / Materials Science and Engineering A 387–389 (2004) 576–584

in Fig. 4 (right). The characteristic line here has an m-value to be 0.1, and substituting in the parameters listed already
near 3 and K ∼ 4 × 10−5 m, using the self diffusivity of and assuming a temperature of 1000 K the subgrain growth
aluminum as R(T). If subgrains and interior dislocation net- constant can be estimated as KN–H ∼ 3 × 109 . This is in the
works coarsen in a geometrically similar way, the coarsen- range of 200–2000 times slower than the process observed
ing constant for subgrains should be about 10 times greater here from data and it takes on the same form with respect
than that for networks, and this is approximately seen here to temperature dependence, grain size and time.
from these two very different data sets.

5.2. A mechanistic comment 6. Integrating the processes

Here, we argue substructures coarsen by long-range va- If we use a rate-independent formulation for dislocation
cancy transport. The argument for subgrain growth con- release, our model can be summarized as the set of equations
trolled by long-range diffusion is that in the motion of low shown in Table 2. A more detailed version of this derivation
angle grain boundaries (which are describable as arrays is available elsewhere [27].
of dislocations), depending on their structure and curva- To develop this model in a general way, generic values
ture, some boundaries must emit vacancies to move in the for constants are used. We note that slip Burgers’ vectors for
driven direction, while others must absorb them. This will a wide range of pure structural metals again fall in a fairly
set up a vacancy flux that is in spirit similar to that for tight range. The values are almost universally between 0.25
Nabarro–Herring diffusional creep. However, the rate of and 0.43 nm. We will use the reference value of 0.3 nm here.
grain boundary motion at a given vacancy flux can be much Elastic moduli and diffusivity do vary widely, but these are
greater than that in diffusional creep because now instead of used as normalizing parameters in the final results. With
atom flux having to provide all the matter to move the bound- these assumptions, we can make direct predictions of the
ary, many fewer atoms can move the boundary by feeding plastic behavior of simple metals. The most obvious way to
the excess edge dislocations on the boundary, allowing it to do this is to tack the change in λ with strain and time due
move forward. to refinement and coarsening to develop a stress strain law.
It is instructive to compare our subgrain coarsening model The full solution to this is somewhat complex and not very
to that for Nabarro–Herring creep, which states: illuminating, however, there are two limits of this that are
  quite useful. So long as the initial dislocation density is low,
DL σΩ the material will initially strain harden (due to dislocation
ε̇ = 10 2 (14)
d kT accumulation) with a strain hardening exponent, N (defined
in σ = cεN ) of
This can be developed into a form similar to that studied
here by again making some reasonable order-of-magnitude 1
N= (16)
estimates. The grain size, d, is taken as the subgrain size 2(1 − c)
λs . The atomic volume Ω, is taken as b3 , and the driving Thus, if c = 0.5, linear hardening will be obtained, but a
stress, σ, is taken as a fraction of the pressure of an iso- value of N = 0.5, indicating parabolic hardening will result
lated subgrain, of 3f(γ/λ). We can solve directly for the ex- if c = 0, as we propose it is here. This is reasonable, because
pected grain growth rate expected if grains were to grow by parabolic hardening is often seen in the low-temperature
Nabarro–Herring creep: strain hardening of annealed metals. We will show in later
  papers that the lower strain hardening exponents associated
λ̇ DL γb3
ε̇ = = 30f 3 (15) with higher temperature deformation can be explained based
λ λ kT
on the natural stochastic strength heterogeneity within and
Putting this into the context of Eq. (13), we can estimate the between grains.
K value that would be expected if subgrains were to grow by As deformation continues, the structure refines and this
Nabarro–Herring creep and this would place a lower bound increases the rate of coarsening. A steady-state condition
on K. If we assume the surface tension, γ, to be 0.5 J/m2 , but will be reached where the rate of dislocation accumulation
the fraction, f, of this pressure driving coarsening, is taken and reduction balance. The steady-state characteristic spac-

Table 2
Phenomenological equations and constants in a first-order general plasticity model
Phenomenon Equation {Range} and/or assumed value Basis

Structure parameter (λ or ρ) Eq. (2) g = {1–20}, g = 4 1: theoretical min.; 20: large gaps in random structure
Athermal yield Eq. (5) s = {0.1–1.0}, s = 0.5 Strength compilations [1,18]
Dislocation accumulation Eq. (8) M ∼ 2 × 1015 , c = 0 Fig. 1
Recovery (structural coarsening) Eq. (13) k = 10−6 m, mc = 3 Figs. 3 and 4
Other constants b = 0.3 nm Small range
G.S. Daehn et al. / Materials Science and Engineering A 387–389 (2004) 576–584 583

ing can be determined by equating the coarsening rate and many creep-resistant materials) stress exponents over 5 are
the refinement rate, both in terms of λ, and solving for the expected.
equilibrium value [27]. The steady-state characteristic spac- This model can provide quantitative predictions of
ing is steady-state strain rates. Table 2 shows typical values for
 1/(2+mc −2c) the necessary model parameters. We believe that all of these
2KD(T)g2−2c
λss = (17) should be approximately correct for a wide range of metals
M γ̇ as is supported by the data compilations shown here and
With the athermal yield equation strain-rate can be written elsewhere. Fig. 5 shows a comparison of Sherby’s [1] “all
as a function of stress as the data in the world” compilation and on top of this the
 n prediction from this model is superposed. The constants
τ used in Table 2 were used as inputs. In another paper [28]
γ̇ = BD(T)
µ we have shown that so long as the activation energy for
   2+mc −2c
2Kg2−2c (bs)−(2+mc −2c) τ diffusion remains less than G∗ for dislocation release the
= D(T) (18) scaling that is shown here is preserved very nicely. Hence,
M µ
thermally activated dislocation release does not dramati-
One of the more significant findings from this investiga- cally complicate the picture. The line from our prediction
tion is that the material steady-state creep stress exponent falls near the center of the block of data. Deviations above
is equal to n = (2 + mc − 2c) where we believe mc to or below can be explained by the relative facility of coars-
be in the range of 3–4 for typical creep processes and c in ening processes changing with factors such as stacking fault
the range of 0–0.5, with zero showing the better fit to ex- energy and nascent dislocation and boundary structures.
perimental data. This immediately explains the commonly Also applied stress may increase the coarsening rate, which
observed steady-state creep stress exponent between 4 and may also elevate the predicted curve.
6. Exponent values considerably higher than 5 can also be
easily rationalized in this framework. If a microstructure is
developed in such a way that it resists coarsening, or pro- 7. Concluding remarks
vides higher coarsening exponents (as may be the case in
A simple approach with one length scale coarsening equa-
tion is postulated in this paper to provide a descriptive and
unified framework to understand recovery. Also a creep
model is presented based on a single length scale coarsening
and refinement processes which are assumed to be indepen-
dent. The coarsening equation and creep model are com-
pared with the reported experimental data and we find a very
good agreement, confirming a quantitative link between this
simple approach of coarsening and the phenomenological
creep behavior of pure metal.
The coarsening equation postulated is consistent with the
compared data sets in the following ways: (i) strain-based
evolution of one length scale λ; (ii) temperature dependence
of recovery rate and creep rate; (iii) quantitative consistency
of the recovery kinetics with creep rates of pure metal via
our model, but this is based on limited data.
This model carries with it two suggestions: (1) in design-
ing creep resistant materials it may be more effective to de-
sign for recovery resistance than for inhibition of dislocation
motion, however, the same strategies will typically work in
both cases; (2) a better understanding of the mechanisms
and phenomenology of recovery is called for.

Acknowledgements

This work was supported by the National Science Foun-


dation under award no. DMR-0080766 as part of the Center
Fig. 5. Sherby’s 1962 compilation with the results of this model with the for Accelerated Maturation of Materials (CAMM). A von
parameters from Table 2 superposed. The line has a slope of 5. Humboldt Foundation grant largely funded the work of Hol-
584 G.S. Daehn et al. / Materials Science and Engineering A 387–389 (2004) 576–584

ger Brehm and SAFE, ERC of Korea funded the work of [14] E. Nes, Acta Metall. Mater. 43 (1995) 2189.
Byeong Soo Lim. [15] L. Ratke, P.W. Voorhees, Growth and Coarsening, Springer-Verlag,
2002.
[16] J.E. Burke, D. Turnbull, Progress in Metal Physics, vol. 3, Pergamon
Press, 1952, p. 220.
References [17] P. Öström, R. Lagenborg, J. Eng. Mater. Technol., Trans ASME Ser.
H 98 (1976) 114.
[1] O.D. Sherby, Acta Mater. 10 (1962) 135–147. [18] P. Öström, R. Lagenborg, Res. Mech. 1 (1980) 59.
[2] M.E. Kassner, M.-T. Pérez-Prado, Prog. Mater. Sci. 45 (2000) 1–102. [19] P.M. Anderson, G.S. Daehn, in preparation.
[3] U.F. Kocks, A.S. Argon, M.F. Ashby, Prog. Mater. Sci. 19 (1975) 1. [20] I.M. Lifshitz, V.V. Slyozov, J. Phys. Chem. Solids 19 (1961) 35–50.
[4] S.J. Basinski, Z.S. Basinski, in: Recrystallization, Grain Growth and [21] C. Wagner, Z. Elektrochem. 65 (1961) 581–591.
Texture, ASM, 1965, pp. 1–44. [22] J.C.M. Li, Recrystallization, Grain Growth and Textures, ASM, 1965,
[5] M.F. Asbhy, in: A. Kelly, R.B. Nicholson (Eds.), Strengthening Chapter 2.
Mechanisms in Crystals, Applied Science Publishers, London, 1971, [23] R. Durard, J. Washburn, E.R. Parker, J. Met. 6 (1953) 1226–1229.
Chapter 3, pp. 137–192. [24] A. Oden, E. Lind, R. Lagenborg, Proceedings: Creep in Steel and
[6] J.J. Gilman, Micromechanics of Flow in Solid, McGraw-Hill, 1969, High Temperature Alloys, Sheffield, 20–22 September, 1972, The
p. 190. Metals Society, London, 1974, pp. 60–66.
[7] J.E. Bailey, P.B. Hirsch, Philos. Mag. 5 (1960) 485–497. [25] J. Cermak, Z. Metallkd. 81 (1990) 193–195.
[8] A. Lawley, H.L. Gaigher, Philos. Mag. 10 (1964) 15–33. [26] Y. Huang, F.J. Humphreys, Acta Mater. 48 (1999) 2017–2030.
[9] M.J. Hordon, B.L. Avverbach, Acta Met. 9 (1961) 247–249. [27] G.S. Daehn, H. Brehm, B.-S. Lim, in: D.R. Lesuer, T.S. Srivatsan,
[10] H.D. Livingston, Acta Met. 10 (1962) 229–239. E.M. Taleff (Eds.), Modeling the Performance of Engineering Struc-
[11] J.W. Edington, R.E. Smallman, Acta Met. 12 (1964) 1313–13238. tural Materials III, Proceedings of the TMS Fall Meeting, 2002,
[12] A. Orlova, H. Cadek, Philos. Mag. 28 (1973) 891. pp. 371–382.
[13] L. Raymond, J.E. Dorn, Trans. AIME 230 (1964) 560–567. [28] H. Brehm, G.S. Daehn, Mater. Trans. A 33 (2002) 363–371.

You might also like