You are on page 1of 5

Ramkumar Oruganti

GE Global Research,
Bangalore 560066, India

Adarsh Shukla
GE Global Research,
Bangalore 560066, India

Sachin Nalawade
GE Global Research,
Bangalore 560066, India

Sanket Sarkar
GE Global Research,
Bangalore 560066, India

K. G. V. Sivakumar
GE Global Research,
Bangalore 560066, India

T. Vishwanath
GE Global Research,
Bangalore 560066, India
A Microstructure-Based Model
Sanjay Sondhi
GE Global Research,
for Creep of Gamma Prime
Bangalore 560066, India
Strengthened Nickel-Based
Andrew Wessman
GE-Aviation,
Cincinnati, OH 45069
Superalloys
This paper outlines a microstructure-based model relating gamma prime microstructure
Daniel Wei and grain size of Ni-base alloys to their creep behavior. The ability of the model to
GE-Aviation, explain creep of multiple superalloys with a single equation and parameter set is demon-
Cincinnati, OH 45069 strated. The only parameters that are changed from alloy to alloy are related to the
gamma prime characteristics and grain size. This model also allows prediction of creep
Andrew Powell performance as a function of heat treatment and explains some apparently contradictory
GE-Aviation, data from the literature. [DOI: 10.1115/1.4040554]
Cincinnati, OH 45069
Keywords: superalloy, creep, model, gamma prime, particle spacing, continuum damage
Kenneth Bain mechanics
GE-Aviation,
Cincinnati, OH 45069

Jon Schaeffer
GE-Power,
Greenville, SC 29615

Arthur Peck
GE-Power,
Greenville, SC 29615

Michael Arnett
GE-Power,
Greenville, SC 29615

Girish Shastry
GE-Power,
Bangalore 560066, India

Francesco Mastromatteo
BHGE,
Florence 50127, Italy

Contributed by the Materials Division of ASME for publication in the JOURNAL OF


ENGINEERING MATERIALS AND TECHNOLOGY. Manuscript received September 12, 2017;
final manuscript received May 22, 2018; published online July 5, 2018. Assoc.
Editor: Antonios Kontsos.

Journal of Engineering Materials and Technology JANUARY 2019, Vol. 141 / 011001-1
C 2019 by ASME
Copyright V

Downloaded From: http://materialstechnology.asmedigitalcollection.asme.org/ on 07/25/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Introduction to account for it or whether it has a deeper significance.
We are now taking this a step further and proposing
Availability of a physics-based predictive model for creep can
that this constraint is the sole and fundamental reason
enable the following: (1) rapid alloy design for better creep resist-
for the spacing effect and this leads to an apparent
ance, (2) design of heat treatments for optimum creep perform-
increase in the c0 volume fraction.
ance, (3) reliable creep lifing of components at the design stage,
(2) The creep mechanism is thermally activated release of dis-
and (4) remaining life assessment of creep limited turbine compo-
locations from nodes rather than from neighboring c0 par-
nents. The model that came closest to a realistic assessment of
ticles [11,12]. This dislocation network is formed in the
creep in superalloys is that of Dyson [1]. This model with no fit-
attempt to accommodate the plastic inhomogeneity men-
ting parameters predicted the creep behavior of Nimonic 90 at
tioned previously.
higher stresses, but underestimated life at low stresses. The applic-
(3) The apparent c0 volume fraction will decrease as particles
ability of the model to other superalloys was not discussed in the
coarsen and reach a lower limit defined by the thermody-
paper though the general “Dyson approach” had been tried with
namic volume fraction.
limited success on other superalloys [2,3]. Subsequently, other
(4) In addition, grain boundary sliding needs to be incorporated
models from Ma et al. for CMSX-4 [4], Zhu et al. for single crys-
for fine grained superalloys (parts manufactured through the
tal Ni-base superalloys [5], Karthikeyan et al. for disk superalloys
cast and wrought, powder metallurgy, or additive routes).
[6], and others were also published though none of them explained
the creep behavior over a wide range of stresses and temperatures The original creep rate equation used by Dyson had the follow-
even for a single alloy. ing general form:
This paper outlines a model that works for a wide range of super-
   
alloys and is based on some modifications to the Dyson approach. It D QC ðr  rB Þkb2
has been used to predict the creep behavior of 14 superalloys. e_ g ¼ A exp  sinh (1)
Di RT MkT

Model
where e_ g is the intragrain strain rate, r is the applied stress, T is
The model of Dyson has been the basis of most of the models the temperature, rB is the back stress, A is a constant, D/Di rep-
published in the literature and the present case is no different. The resents the increase of mobile dislocation density, QC is the
three basic tenets that underlie the microstructure-based creep creep activation energy, M is the Taylor’s factor, k is the Boltz-
model of Dyson are the following: mann constant, and R is the gas constant. kb2 is the activation
(1) c0 particles have two roles to play volume related to the hold–release process and k is the spacing
(a) Holding dislocations between neighboring c0 particles of the obstacles which actually hold and release dislocations.
and releasing them through thermally activated climb: These obstacles, assumed to be c0 particles by most researchers,
Since the spacing of the particles controls the release are now taken to be dislocation nodes rather than particles.
rate, creep strength decreases continuously with Since dislocation spacing can be a function of stress and tem-
increasing c0 spacing with no obvious lower limit. perature, it is represented by an expression of the following
(b) Composite strengthening due to plastic inhomogeneity form [11]:
between the creeping matrix and the noncreeping c0
particles [7]: Consequently, the strengthening effect of k / KS expðQS =RTÞðr  rB Þ0:5 (2)
c0 particles, quantified by the back-stress generated in
the matrix, is determined by their volume fraction [8]. Combining Eqs. (1) and (2), we have
(2) The process of load transfer from the creeping matrix to the
hard particles (in situations much below the yield strength)   "   pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi#
causes primary creep. D QC QS ðr  rB Þ
e_ g ¼ A exp  sinh Ks exp  (3)
(3) Tertiary creep is primarily due to increase in mobile dislo- Di RT RT T
cation density.
(4) Secondary creep is only an inflection point between pri-
mary and tertiary creep and typically superalloys do not where and Ks and Qs are activation parameters related to the dislo-
show any extended period of steady-state creep. cation node spacing (the factors M and k are also subsumed within
the Ks parameter for the sake of convenience). The Arrhenius
These tenets form a good theoretical framework but do not lead
term within the sin h function arises because of the temperature
to a model that is able to represent the creep behavior of superal-
dependence of the dislocation node spacing. An important point to
loys over all stress and temperature ranges. The common approach
be noted here is that the particle spacing no longer explicitly
of researchers to arrive at better concordance between model pre-
occurs associated with the applied stress. It is this fundamental
dictions and experimental data has been to add more mechanisms
change that enables modeling of creep in superalloys without too
and equations to the basic approach outlined above [3–6]. Our
many other assumptions.
proposal is that a few small but important changes to the princi-
The additional creep strain contribution arising from grain
ples outlined above lead to a much better creep model with very
boundary sliding in fine grained materials is accounted for by the
few additional assumptions. These changes are as follows:
following well-known equation [13]:
(1) On the role of c0 particles:
(a) c0 particles have only one role to play—composite Kgb rm
e_ gb ¼ (4)
strengthening due to plastic inhomogeneity as men- dn
tioned earlier.
(b) When c0 particles are spaced close together, the con- where d is the average grain diameter and Kgb is temperature
straints placed on matrix deformation by such closely dependent and m and n are constants. The total creep rate is the
spaced particles [1,9,10] cause an increase in the appa- sum of Eqs. (3) and (4).
rent volume fraction of c0 and consequently an increase Back stress arises from stress transfer from the creeping c
in back stress. matrix to the elastically deforming c0 particles. As mentioned ear-
While Dyson did propose that closely spaced particles lier, this process of stress transfer causes primary creep and a sim-
in high c0 volume fraction alloys would lead to higher plified version of the following Dyson equation [1] describing this
values of back stress, he left open the question of how process is used:

011001-2 / Vol. 141, JANUARY 2019 Transactions of the ASME

Downloaded From: http://materialstechnology.asmedigitalcollection.asme.org/ on 07/25/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


f 0 rB 1
r_ B ¼ E_e 1    (5)
1f @ 2f A
r
1 þ 2f

where E is the elastic modulus of the c0 particles and f is the vol-


ume fraction of the strengthening phase. f, which we will refer to
as “apparent volume fraction” is a sum of the thermodynamic
volume fraction of c0 and the additional volume fraction arising
from constraint.
Equations (3)–(5) when integrated together will be able to
reproduce primary and secondary creep. Secondary creep in this
formulation occurs when the maximum possible stress has been
transferred from the creeping matrix to the hard particles. If no
other process intervenes, creep will occur at a constant rate end-
lessly. But this is not the case with superalloys, as pointed out ear-
lier. Following Dyson [1], tertiary creep is assumed to occur due
to increase in mobile dislocation density and is represented by the
following equation:
D_ ¼ K1 e_ þ K2 e_e (6)
where K1 and K2 are constants. Additionally, weakening will
occur due to loss of constraint resulting from coarsening of c0 . A
kinetic equation describing the resulting loss of apparent c0 vol-
ume fraction can be derived from measurements on samples
exposed to high temperature. A simple kinetic equation with the
following general form then results:

f_ ¼ gðT; t; f Þ (7)

Equation (7) is designed such that f decreases with increasing time


and temperature and eventually reaches a limiting value equal to
the thermodynamic volume fraction.

Parameters and Calibration


Two microstructural parameters of the model change from alloy
to alloy and these are grain size d and the apparent c0 volume frac-
tion f, both of which can be determined from micrographs. The
parameters for Eq. (7) (apparent c0 volume fraction evolution) can
be determined separately for each alloy of interest depending on Fig. 1 (a) Experimental and predicted strain versus time for
the level of predictive precision required. DS-GTD111 a directionally solidified Ni-base superalloy with
The parameters A, Qc, Ks, and Qs in Eq. (3) are calibrated only 55% c0 volume fraction. (b) Comparison of predictions from
once using short-term creep data (<1000 h) of one alloy (DS- the Dyson model and the current model with experimental data
on DS-GTD111.
GTD111). Thereafter, these creep mechanism-related parameters
are maintained constant for all other alloys. That this works rea-
sonably well points to the strong possibility that the fundamental includes data, predictions from the current model, and predictions
creep mechanism has remained the same over multiple genera- from Dyson’s modeling approach. It is clear from this figure that
tions of superalloys and over a wide range of stresses and temper- the problem of underestimation at lower stresses that is character-
atures (below yield). istic of the Dyson model is avoided by the current model. As
Parameters in Eq. (6) and Di were determined from X-ray peak pointed out earlier, this is primarily enabled by the new constitu-
broadening measurements on interrupted creep samples of DS- tive approach embodied in Eq. (3)—that the fundamental creep
GTD111. These parameters were found to work very well for all mechanism is hold–release of dislocations from networks and not
the alloys and hence were maintained unchanged. neighboring c0 particles.
Thus, it can be seen that the parameters that need fitting to Comparisons of experimental versus predicted lifetimes for vari-
creep data are very few and remain unchanged for different alloys. ous strains in four superalloys are shown in Figs. 2(a)–2(d). For
The only parameters that change from alloy to alloy are micro- these predictions, the average value of the apparent c0 volume frac-
structure related and measurable. Also, this approach does not tion for each alloy was used. It is seen that the model is able to pre-
need complex assumptions about bimodal and trimodal c0 distribu- dict the average behavior of a wide range of superalloys (powder/
tions as in the case of Ref. [3]. wrought, directionally solidified, single crystal) over a wide range
of stresses and temperatures with no significant systematic bias.
There are very large differences in chemical composition (Table 1)
Predictions going from R65, a cast and wrought polycrystalline Ni-base superal-
Equations (3) through Eq. (7) can be integrated to obtain creep loy with about 40% c0 volume fraction and grain size of 5–10 lm, to
strain as a function of time. An example of such a plot is shown in Rene’ N5, a cast single crystal superalloy with about 70% c0 volume
Fig. 1(a) for DS-GTD111. It is seen that the model is able to fraction. This appears to imply that, to a large extent, the improve-
match the experimental data reasonably well within experimental ments in creep resistance in Ni-base superalloys over the past few
error. Isostrain plots for different stress–temperature combinations decades can be attributed to the increases in the apparent c0 volume
can be derived from such creep strain versus time plots and one fraction (thermodynamic þ constraint). This proposition is worth
such plot for DS-GTD111 is shown in Fig. 1(b). Figure 1(b) investigating since it provides pointers for future alloy design.

Journal of Engineering Materials and Technology JANUARY 2019, Vol. 141 / 011001-3

Downloaded From: http://materialstechnology.asmedigitalcollection.asme.org/ on 07/25/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Fig. 2 Predicted time-to strain versus experimental time-to-strain for four superalloys with different compositions
and c0 volume fractions. The data points include strains from 0.2% to 5%. The dashed lines indicate 610% bounds.
One model with microstructural inputs and no other change in parameters is able to predict the behavior of all the
alloys (a) Rene’ 65, (b) DS-GTD111, (c) GTD444, and (d) Rene’ N5.

From the 610% bounds shown in Fig. 2, it is seen that that the fraction was known beforehand for subsets of samples. This sec-
prediction error while large for short times decreases significantly tion also illustrates how initial differences in apparent volume
for longer test times. The large prediction error at short times is fraction do not matter in the long term (the definition of long term
because the average apparent volume fraction for each alloy rather is of course temperature dependent).
than the value for each sample was used. The short-term predic-
tion accuracy would have been much higher if the individual
apparent volume fractions for each sample had been used. But
obtaining this parameter for each sample is a prohibitive exercise
and in any case the original samples were unavailable for much of
the data.
In the long term, on the other hand, the apparent volume frac-
tion decreases and reaches a limiting value beyond which no fur-
ther change occurs. Thus, the long term predictions are not
sensitive to initial variations in the apparent volume fraction lead-
ing to better predictive accuracy.
The section Predicting the Effect of Initial c0 Microstructure
illustrates a more controlled data set where the apparent volume

Table 1 Chemical compositions of various alloys represented


in this paper (numbers in wt %)

Ni Cr Co Fe Mo W Al Ti Nb Ta Re

Rene 65 Bal 16.0 13.0 1 4.0 4.0 2.1 3.7 0.7 — —


DS-GTD111 Bal 14.0 9.5 — 1.5 3.8 3.0 4.9 — 2.8 — Fig. 3 Experimental data and predictions for DS-GTD111 with
GTD444 Bal 9.8 7.5 — 1.5 6.0 4.2 3.5 0.5 4.8 — two different initial c0 distributions. Both data converge to a
Rene N5 Bal 7.0 7.5 — 1.5 5.0 6.2 — — 7.0 3.0 common trajectory after a certain amount of time. The point of
convergence shifts to lower times as temperature increases.

011001-4 / Vol. 141, JANUARY 2019 Transactions of the ASME

Downloaded From: http://materialstechnology.asmedigitalcollection.asme.org/ on 07/25/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Predicting the Effect of Initial c0 Microstructure. Figure 3 will completely disappear after a fixed time and tempera-
shows creep data of DS-GTD111 obtained as part of this study. ture exposure with the lower limit of creep resistance being
The two sets of data correspond to peak aged and overaged sam- defined by the thermodynamic volume fraction of c0 . For
ples. Overaging was carried out to dissolve tertiary c0 (<100 nm) turbine components operating at temperatures in the range
and coarsen the secondary c0 particles (>400 nm). The thermody- of >850  C, this limit is reached in a few tens of hours to a
namic c0 volume fraction of this alloy is 55%, which is also the few hundred hours.
apparent volume fraction of the overaged material (since the par- (4) An important conclusion from this work is that c0 size/
ticles are coarse and far apart with no constraint effect). The peak spacing alone cannot be used to predict remaining life since
aged samples on the other hand had an apparent c0 volume fraction it ceases to have any role on creep resistance beyond a cer-
of 62%, which includes the thermodynamic volume fraction and tain value. Having said this, remaining life assessment is
the additional volume fraction arising from constraint. The data possible if other microstructural parameters, such as dislo-
show that while the peak aged material starts off stronger at high cation density and its evolution, are also taken into account.
stresses, the apparent volume fraction erodes with time until its This will be the subject of a future publication.
performance reaches that of the overaged material (apparent vol-
ume fraction ¼ 55%). Beyond this point of convergence, there is Acknowledgment
no more weakening. The point of convergence of the peak-aged
and overaged material shifts to lower times as the temperature Funding from the Advanced Technology Program of GE Global
increases. The overaged material defines the lower limit of creep research (Steve Duclos, Bernard Bewlay, Michael Idelchik), GE
capability of this material. All these aspects are captured accu- Aviation, GE Power, and BHGE are gratefully acknowledged.
rately by the current model (lines in the plot). Vinod Kumar, Technology Leader-GE Global Research is
acknowledged for constant support both funding and otherwise.
Discussions with Debashis Kar, Swapnil Patil, Timothy Hanlon,
Explanation of Literature Data. The model explains some Chen Shen (all from GE-Global Research), Srinivasan Swamina-
apparent contradictions seen in superalloy creep literature. One than, Justin Bennett, and Jeff Williams (all from GE-Aviation)
such instance is as follows: Stevens and Flewitt’s experiments on and Khaled Abdel-Tawab, Steven Zimmerman, Liming Xu (all
IN738 [14] showed that the creep capability of this material from GE-Power) are acknowledged. Yogesh Potdar (GE-Global
showed a strong dependence on the prior aging treatment.1 On the Research) is acknowledged for careful reading of the manuscript
other hand, Tipler and Peck quoted by Dyson and McLean [15] and valuable suggestions. The journal reviewers are also acknowl-
showed that standard aged samples and overaged samples had edged for their diligent examination of the manuscript, which
almost identical lives (about 8000 h).2 This apparent contradiction helped improve the overall quality of the paper. Finally, we would
can be explained using the framework proposed in this paper: The like to place on record our deep gratitude to Dr. Brian Dyson for
high stresses used in Stevens and Flewitt’s tests ensured that the his unstinting offer of advice and guidance at various stages of the
test durations were much shorter than the tests of Tipler and Peck. project. We benefited greatly from his erudition and deep under-
Therefore, the differences in initial apparent volume fraction standing of high temperature mechanical behavior.
between the standard aged and overaged samples sustained over
the duration of their tests leading to different lives. Tipler and
References
Peck’s tests, on the other hand, ran much longer and the initial dif-
[1] Dyson, B. F., 2009, “Microstructure Based Creep Constitutive Model for Pre-
ferences in apparent volume fraction disappeared early leading cipitation Strengthened Alloys: Theory and Application,” Mater. Sci. Technol.,
eventually to similar lives. 25(2), pp. 213–220.
This result shows that at typical operation temperatures of for- [2] Basoalto, H., Sondhi, S. K., Dyson, B. F., and McLean, M., 2004, “A Generic
ward stage turbine blades (>850  C), the creep resistance of Microstructure-Explicit Model of Creep in Nickel-Base Superalloys,” Superal-
loys, K. A. Green, T. M. Pollock, J. Harada, T. E. Howson, R. C. Reed, J. J.
superalloys rapidly approaches their stable value. Beyond this sta- Schirra, and S. Walston, eds., TMS, Warrendale, PA, pp. 897–906.
bilization point, the creep resistance has no dependence on the c0 [3] Coakley, J., Dye, D., and Basoalto, H., 2011, “Creep and Creep Modeling of a
distribution.3 Multimodal Nickel-Base Superalloy,” Acta Mater., 59(3), pp. 854–863.
[4] Ma, A., Dye, D., and Reed, R. C., 2008, “A Model for the Creep Deformation
Behavior of Single Crystal Superalloy CMSX-4,” Acta Mater., 56(8), pp.
1657–1670.
Summary [5] Zhu, Z., Basoalto, H., Warnken, N., and Reed, R. C., 2012, “A Model for Creep
Deformation of Nickel-Based Single Crystal Superalloys,” Acta Mater., 60(12),
(1) The model outlined in this paper is able to predict the creep pp. 4888–4900.
behavior of multiple superalloys with two measurable [6] Karthikeyan, S., Unocic, R. R., Sarosi, P. M., Viswanathan, G. B., and Whitis,
microstructural inputs, the grain size and the apparent c0 D. D., 2006, “Modeling Microtwinning During Creep in Ni-Based Super-
alloys,” Scr. Mater., 54(6), pp. 1157–1162.
volume fraction (and its evolution with time). The apparent [7] Ashby, M. F., 1970, “The Deformation of Plastically Non-Homogeneous Mate-
c0 volume fraction is a combination of the thermodynamic rials,” Philos. Mag., 21(170), pp. 399–424.
volume fraction and that due to constraint. [8] Brown, L. M., and Stobbs, W. M., 1971, “The Work Hardening of Copper-
(2) An important corollary of this model is that a large part of Silica—Part I: A Model Based on Internal Stresses With No Plastic Relax-
ation,” Philos. Mag., 23(185), pp. 1185–1189.
the improvements in creep capability of superalloys over [9] Groh, S., Devincre, B., Kubin, L. P., Roos, A., Feyel, F., and Chaboche, J.-L.,
the past few decades can be attributed to increases in appa- 2005, “Size Effects in Metal Matrix Composites,” Mater. Sci. Eng., 400–401,
rent volume fraction of c0 . This proposition is worth investi- pp. 279–282.
gating further since it has implications for future alloy [10] Pollock, T. M., and Argon, A., 1992, “Creep Resistance of CMSX-3 Nickel-
Base Superalloy Single Crystals,” Acta Metall., 40(1), pp. 1–30.
design. [11] Oruganti, R., 2012, “A New Approach to Dislocation Creep,” Acta Mater.,
(3) The increase in creep capability arising from finer c0 size 60(4), pp. 1695–1702.
and spacing is explained by the increase in apparent c0 vol- [12] Daehn, G. S., Brehm, H., Lee, H., and Lim, B. S., 2004, “A Model for Creep
ume fraction. The approach also predicts that this increase Based on Microstructural Length Scale Evolution,” Mater. Sci. Eng. A,
387–389, pp. 576–584.
[13] Wadsworth, J., Ruano, A., and Sherby, D., 1999, “Deformation by Grain
1
Standard aged material was consistently stronger than overaged material when Boundary Sliding and Slip Creep Versus Diffusional Creep,” Creep Behavior of
tested at 750  C and 457 MPa with test times ranging from 200 hours to 1700 hours. Advanced Materials for the 21st Century, Rajiv S. Mishra, Amiya K. Mukher-
2
When tested at a higher temperature of 852  C and lower stress of 170 MPa. jee, and K. Linga Murty, eds., TMS, Warrendale, PA, pp. 425–439.
3
This is not to say that there will be no variability in rupture time. Rupture is [14] Stevens, R. A., and Flewitt, P. E. J., 1979, “The Effects of c0 Precipitate Coars-
determined by many factors such as the presence of defects (grain boundaries, ening During Isothermal Aging and Creep of the Nickel-Base Superalloy
porosity, brittle phases) and oxidation of grain boundaries. The model described here IN738,” Mater. Sci. Eng., 37(3), pp. 237–247.
deals only with creep deformation and not failure, an important but frequently [15] Dyson, B. F., and McLean, M., 1983, “Particle Coarsening and Tertiary Creep,”
ignored distinction. Acta Metall., 31(1), pp. 17–27.

Journal of Engineering Materials and Technology JANUARY 2019, Vol. 141 / 011001-5

Downloaded From: http://materialstechnology.asmedigitalcollection.asme.org/ on 07/25/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

You might also like