www.nature.com/npjcompumats
ARTICLE
OPEN
Predicting temperature-dependent ultimate strengths of
body-centered-cubic (BCC) high-entropy alloys
B. Steingrimsson
1✉
, X. Fan2, X. Yang3, M. C. Gao
4
, Y. Zhang
5,6,7 ✉
and P. K. Liaw
2
This paper presents a bilinear log model, for predicting temperature-dependent ultimate strength of high-entropy alloys (HEAs)
based on 21 HEA compositions. We consider the break temperature, Tbreak, introduced in the model, an important parameter for
design of materials with attractive high-temperature properties, one warranting inclusion in alloy specifications. For reliable
operation, the operating temperature of alloys may need to stay below Tbreak. We introduce a technique of global optimization, one
enabling concurrent optimization of model parameters over low-temperature and high-temperature regimes. Furthermore, we
suggest a general framework for joint optimization of alloy properties, capable of accounting for physics-based dependencies, and
show how a special case can be formulated to address the identification of HEAs offering attractive ultimate strength. We advocate
for the selection of an optimization technique suitable for the problem at hand and the data available, and for properly accounting
for the underlying sources of variations.
1234567890():,;
npj Computational Materials (2021)7:152 ; https://doi.org/10.1038/s41524-021-00623-4
INTRODUCTION
Metallic structural materials with excellent mechanical properties
have been widely used in a variety of operating conditions and
often applied under constant or static loads. Engineering
components under either loading conditions are usually required
to exhibit high strength. Thus, it is important to be able to design
advanced materials with favorable strength properties. Highentropy alloys (HEAs) have drawn great attention in the recent
decade due to their excellent mechanical properties and vast
compositional space, which makes them suitable for this
purpose1. A key objective is to suggest a framework for joint
optimization of mechanical properties, to introduce—in context
with such a framework—compositions of HEAs yielding high
ultimate strengths (USs), and to conduct experimental verification of our findings.
Figure 1 outlines the multiple sources that impact the mechanical
properties of HEAs, and highlights dependence between them. It is
worth noting that improvements in the US may come at the
expense of other properties (hence, framework for joint optimization). For example, there usually is a trade-off between the ductility
and the strength of alloys. Sources of variations in US may involve
difference in compositions, microstructures, parameters of postfabrication processes, or defect levels.
In contrast to traditional alloys containing only one or two
principal elements, multi-principal-element alloys, also referred to
as HEAs, have been developed and studied in the recent decade1–7.
Carefully designed HEAs with either single or multiple phases have
shown encouraging mechanical properties, compared to conventional alloys8–16.
Data analytics and machine learning (ML) can help with rapid
screening, i.e., expedite identification of HEAs exhibiting given
properties of interest17. But as opposed to specifically applying
ML, (narrowly) defined in terms of single-layer or multi-layer
neural networks17, Bayesian graphical models, support vector
machines, or decision trees, to identification of HEA compositions
of interest, we reformulate the task in the broader context of
engineering optimization. We recommend picking an optimization technique suitable for the application at hand and data
available. But we certainly include ML in the consideration. For
background material on ML, refer to17.
Effective application of ML may require a large number of data
points. If you have such data, then ML can help you organize the
data in a meaningful fashion and extract complex, hidden
relationships17. But in the case of experimental data on HEA
compositions with attractive strength properties (the present state
of affairs), we are working in a domain of relatively limited data, a
domain where traditional ML may exhibit limitations. Producing
high-quality experimental data is usually both time consuming
and expensive. In case of such limited data, it is essential to make
the most of the underlying physics, i.e., to account for underlying
physical dependencies, in the prediction model. Occam’s razor
and Bayesian learning provide tools for quantifying the notion of
limited data in this context17.
Our approach is in part based on observations of Agrawal
et al.18. Table 2 and Figure 5 of18 illustrate that there is at most a
difference of a few percentages between the techniques applied
to predict the fatigue strength of stainless steels. Table 2 of18
shows that both simple linear regression and pace regression
yield the coefficient of determination, R2, of 0.963, while an
artificial neural network, a traditional ML technique, results
in R2 of 0.972.
In terms of an important contribution, this study presents a
method capable of yielding consistency among the predictions of
HEA compositions with attractive US, empirical rules of thermodynamics19,20, and experimental results. We accomplish this goal,
despite relatively limited data available, and the corresponding
selection of a simple prediction algorithm (multi-variate regression).
1
Imagars LLC, 2062 Thorncroft Drive Suite 1214, Hillsboro, OR 97124, USA. 2Department of Materials Science and Engineering, The University of Tennessee, Knoxville, TN 37996, USA.
University of Chinese Academy of Sciences, Center of Materials Science and Optoelectronics Engineering, Beijing 100049, China. 4National Energy Technology Laboratory, 1450 Queen
Ave. SW, Albany, OR 97321, USA. 5Beijing Advanced Innovation Center of Materials Genome Engineering, State Key Laboratory for Advanced Metals and Materials, University of Science
and Technology Beijing, Beijing 100083, China. 6Qinghai Provincial Engineering Research Center of High Performance Light Metal Alloys and Forming, Qinghai University, Xining
810016, China. 7Shunde Graduate School, University of Science and Technology Beijing, Foshan 528399, China. ✉email: baldur@imagars.com; drzhangy@ustb.edu.cn
3
Published in partnership with the Shanghai Institute of Ceramics of the Chinese Academy of Sciences
B. Steingrimsson et al.
2
1234567890():,;
Fig. 1 High-level depiction of role of optimization techniques for inferring the features of HEAs, including the composition,
microstructure, heat treatment, and process, from mechanical properties, including ultimate strength, desired. The color coding provides
the insight into the extent to which the sources are separate, and yet interconnected to a certain extent34,35.
Then, we present in Fig. 2 elements of a physics-based model for
predicting the US, a model that accounts for physical dependencies
as a priori information.
But more importantly, we introduce a bilinear log model for
predicting USs across temperatures. The model consists of
separate exponentials, for a low-temperature and a hightemperature regime, with a break temperature, Tbreak, in
between. The model accounts for the underlying physics, in
particular diffusion processes required to initiate phase transformations in the high-temperature regime21. Furthermore, we
show how piecewise linear regression can be employed to
extend the model beyond two exponentials and yield accurate
fit, in case of a non-convex objective function caused by hump
(s) in the data. Previous models for the temperature dependence of yield strengths (YSs) only accounted for a single
exponential22,23. Hence, there was no break temperature, Tbreak.
We consider the break point critical for the optimization of the
high-temperature properties of alloys. For reliable operation, the
temperature of turbine blades made out of refractory alloys may
need to stay below Tbreak. Once above Tbreak, materials can lose
strength rapidly due to rapid diffusion, leading to easy
dislocation motion and dissolution of strengthening phases21.
We consider Tbreak an important parameter for the design of
materials with attractive high-temperature properties, one
warranting inclusion in alloy specifications. Hence, it is
important to accurately estimate Tbreak, e.g., using the global
optimization approach presented.
RESULTS
Room temperature
Figure 26 of17 summarizes the rational for initially restricting
the analysis to room-temperature data. As illustrated in the
figure, the US exhibits significant dependence on the temperature. While all compositions in Figure 26 of17 contained a bodycentered-cubic (BCC) phase, and were subjected to some type
of annealing, the US at 1000 °C can be ~1/8th (~12%) to ~1/3rd
(~33%) of the US at room temperature. With this fact in mind,
and to maintain consistency across compositions, we elected
to first apply the optimization framework to US at room
temperature only.
Our original data set, listed in Table 13 of17, contains some 24
compositions that yield relatively high US at room temperature.
npj Computational Materials (2021) 152
To accommodate the elements involved, we derived two feature
sets, hereafter referred to as A and B, from the original data set in
Table 13 of17:
Feature Vector A ¼ xA ¼ ½%Al;
%Mo;
%Nb;
%Ti;
%V;
%Ta;
%Zr;
%Hf ;
(1)
Feature Vector B ¼ xB ¼ ½%Al;
%Mo;
%Nb;
%Ti;
%V;
%Ta;
%Zr;
%Hf;
%Cr:
(2)
We have available 19 instances of Feature Vector A, and 22 of
Feature Vector B. While the set of input data may seem small,
we will show that the data suffice for meaningful prediction,
provided that a suitable optimization technique is selected17,24.
In terms of data curation, we concluded that the US values,
except for MoNbTiVxZr (from25), were recorded with high
enough fidelity to warrant inclusion. For revised US measurements for MoNbTiV0.75Zr, refer to Table 1. For catching
suspicious recordings of the US, one can employ proportionality
relationships with the YS as a guideline. At least (or about) half
of the references associated an uncertainty interval with the US
values reported, with ΔUS usually within the range of 1% of the
US reported.
In order to develop insight into the causes of variations in the
US for the pure elements comprising Feature Vectors, A and B,
and for the identification of a model for predicting compositions yielding high US, we point to Figures 24 and 25 of17.
Figure 24 of17 shows that processing conditions and purity can
contribute to variations in US of Al of ~3x and of ~4x in the US
of Co. Figure 25 of17 similarly illustrates that processing
methods can have significant influence on the US of V and Cr.
For V, the variations in the US are ~2x, and for Cr, we are looking
at variations of up to ~5x. This trend suggests that the inputs
listed in Eq. (22) are indeed able to account for the variations in
the US observed. But for a relative comparison of US across
compositions, for the same heat-treatment process and defect
levels, and at a fixed (room) temperature, the model
US ¼ h ðcompositionÞ
(3)
may suffice. There may be additional sources of variations, such
as the test mode applied. But according to our tentative analysis,
presented in Supplementary Table 2, the variations in the YS
observed, based on the test mode applied, tend to be relatively
small. For the prediction presented in Supplementary Fig. 3, and
Published in partnership with the Shanghai Institute of Ceramics of the Chinese Academy of Sciences
B. Steingrimsson et al.
3
Fig. 2 Underlying physical dependencies structured in a fashion resembling a neural network. Our intent is to construct models capturing
the underlying physics. This abstracted model shows that the microstructure formed depends on the heat-treatment process applied,
manufacturing, processing as well as the composition34. In case that an artificial neural network (ANN) is deemed suitable for the application
at hand, we suggest employing custom kernel functions consistent with the underlying physics, for the purpose of attaining tighter coupling,
better prediction, and extracting the most out of the—usually limited—input data available. Note that the composition can include trace-level
elements (impurities), in addition to the principal components.
Table 1. Compression yield strength, σy, maximum strength, σmax, and fracture strain, εf, of the reference and predicted compositions at room
temperature.
Composition
Sample diameter (mm)
Strain rate (s–1)
σy (MPa)
σmax (MPa)
Al0.5Mo0.5NbTa0.5TiZr
Al0.5Mo0.5Nb1.5Ta0.5Zr1.5
3
3
1 × 10–3
1 × 10–3
1786
1791
1910
2024
7.6
7.6
MoNbTiV0.75Zr
3
1 × 10–3
1675
2427
25.1
MoNbTiV0.75Zr
5
2 × 10–4
1599 ± 40
1780 ± 70
Mo1.25Nb1.25Ti0.5V0.5Zr1.25
3
1 × 10–3
1705
2013
εf (%)
8.6 ± 2.8
10.0
All the alloys show high strength and obvious plastic deformation. With the similar fracture strain (7.6%), the predicted Al0.5Mo0.5Nb1.5Ta0.5Zr1.5 exhibit higher
σy and σmax of 1791 and 2024 MPa, respectively, compared with the reference composition, Al0.5Mo0.5NbTa0.5TiZr, which has σy and σmax as 1786 and 1910 MPa,
respectively. MoNbTiV0.75Zr and Mo1.25Nb1.25Ti0.5V0.5Zr1.25 have similar YSs (1675 and 1705 MPa, respectively), but the reference composition exhibits the
higher maximum strength due to the high fracture strain and strain hardening. Note the corrected US values for MoNbTiV0.75Zr relative to25. The uncertainty
intervals in YS and US for the 5-mm diameter MoNbTiV0.75Zr were derived from measurements of three separate samples.
the experimental verification outlined in Fig. 5, we employ the
prediction model of Eq. (3).
Given the relatively small size of the data set in Table 13 of17,
it appears that we may not be ready for traditional ML models.
Models, such as artificial neural networks, decision trees,
support vector machines, Bayesian networks, or genetic algorithms, tend to be effective in organizing and extracting
complex patterns from large sets of data17. But for the
application and limited data set at hand, it makes sense to
select a simple linear-prediction model, multi-variate linear
regression, to begin with, and build from there. As suggested by
Agrawal et al.18, changing the method may not vary the results
that much. According to Figure 5 of18 and Table 2 of18, the linear
regression yields R2 of 0.963, when predicting the fatigue
strength of a stainless steel, compared to R2 of 0.972 for the
artificial neural networks.
Our approach, outlined in17, assumes starting with a simple
model, multi-variate linear regression, and accounting for the
input sources that contribute to variations in the US observed.
The approach then involves expanding the model, and adding
non-linearities, based on the underlying physics, and as necessitated by the application at hand and data available.
When applying the multi-variate linear regression, we solve an
unconstrained optimization problem of the form
minb kX b
yk22 :
(4)
Here, y represents a vector of US values, b denotes a vector of
regression coefficients, and X symbolizes a training set of
compositions (a stacked version of x vectors17, derived from
Table 13 of17). This unconstrained optimization problem has a
well-known, closed-form solution17:
1
(5)
b ¼ XT X XT y:
When the training set is very small, the inverse (XTX)–1 may not
exist. In that case, we recommend replacing (XTX)–1 with (XTX)+,
the pseudoinverse26.
The observations reported in Supplementary Fig. 2—and more
extensively in17—strengthen our belief in that the prediction
accuracy, measured in terms of R2 and the standard deviation
normalized per data point, is primarily limited by the quality of
(variance in) the input data. These limitations in the prediction
accuracy are consistent with the variations observed in Table 13
and Figures 24 and 25 of17. These observations further suggest
Published in partnership with the Shanghai Institute of Ceramics of the Chinese Academy of Sciences
npj Computational Materials (2021) 152
B. Steingrimsson et al.
4
that multi-variate regression is indeed a suitable technique for
this application.
Elevated temperatures
In an effort to identify compositions exhibiting the ability to
retain strength at high temperatures, we present Fig. 3. In case
of high-temperature applications, we are looking to derive a
model of the form
US ¼ hðcomposition; TÞ
(6)
for the prediction of the US across temperatures. In addition to
the temperature dependence of pure tungsten and the HEAs,
the temperature dependence of the commercial alloys [the
Mo-rich Titanium-Zirconium-Molybdenum alloy and the Nb-rich
C-103 alloy] is also of interest.
Looking at Fig. 3, one first notices that the strength vs.
temperature data definitely do not look linear. Hence, the multivariate linear regression may no longer be the preferred
approach. Second, the temperature dependence does come
across as approximately exponential, but not exactly. It seems
to entail a high-temperature and a low-temperature regime.
Third, one may shy away from employing an automated ML
suite, such as the Tree-Based Pipeline Optimization Tool27,
because of limited ability of such black-box models to provide
much needed insights into the underlying physics. One is
motivated to make the most of the limited data available, by
incorporating important a priori information about the underlying physics into the model structure, for purpose of deriving
such insights. Fourth, Fig. 3a, e.g., the high-temperature data
points for MoNbTaW, highlights the need for data curation.
Fig. 3 Identification of compositions with the ability to retain strengths at high temperatures. Panel (a) shows the quality of fit for the
bilinear log model in the linear domain. Panel (b) depicts the quality of fit for the bilinear log model in the logarithmic domain. Panels (c) and
(d) illustrate how experimental measurements of temperature-dependent yield strength naturally conform to two linear regions, when viewed
in the logarithmic domain. Panel (c) is reproduced from21 with permission.
npj Computational Materials (2021) 152
Published in partnership with the Shanghai Institute of Ceramics of the Chinese Academy of Sciences
B. Steingrimsson et al.
5
Motivated by Fig. 3c, d, together with physics-based insights
from21, we model the temperature dependence of the US(T), in
terms of a bilinear log model, parametrized by the melting
temperature, Tm, as follows:
USðT Þ ¼ minðlogðUS1 ðT ÞÞ;
logðUS2 ðT ÞÞÞ;
(7)
US1 ðT Þ ¼ expð C1 T=Tm þ C2 Þ;
0 < T < Tbreak ;
(8)
US2 ðT Þ ¼ expð C3 T=Tm þ C4 Þ;
Tbreak < T < Tm :
(9)
There is an additional physics (diffusion) induced constraint on
Tbreak21:
0:35 t T break =T m t 0:55;
(10)
and a continuity constraint between the low-temperature and the
high-temperature regimes:
US1 ðTbreak Þ ¼ US2 ðTbreak Þ;
T break ¼
ðC 4
ðC 3
(11)
C2Þ
Tm
C1Þ
(12)
A conceptually simple approach for fitting the model in Eqs. (7)–
(12) to the US data available consists of first deriving the constant
coefficients, C1 and C2, by applying linear regression to data points
available to the lowest temperature region (0 < T < 0.35 Tm) as well
as to the intermediate region (0.35 Tm ≤ T ≤ 0.55 Tm). One can then
derive the constants, C3 and C4 by applying linear regression to
data points available to the intermediate (0.35 Tm ≤ T ≤ 0.55 Tm)
and high-temperature (T > 0.55 Tm) regions. Note that Tbreak does
not need to be known in advance. According to Eq. (12), this
inherent property of a given alloy comes out of the model as the
break point between the two linear regions. The model consists of
only four independent parameters, C1, C2, C3, and C4, which simply
can be estimated by applying linear regression separately to lowtemperature and high-temperature regimes, even to a fairly small
data set. Note, furthermore, that for a new alloy system, Tm, does
not need to be known experimentally in advance either; a rough
estimate for Tm can be obtained, using “the rule of mixing”, and a
more refined estimate obtained, employing Calculation of Phase
Diagram (CALPHAD) simulations17.
A superior approach for deriving the coefficients, C1, C2, C3, and
C4, involves concurrent optimization over the low-temperature
and high-temperature regimes using global optimization. Here,
we seek to minimize
(13)
minC1 ;C2 ;C3 ;C4 norm2; i ðUSðT i Þ y i Þ2 ;
where yi represents the measured US values,
USðTi Þ ¼ minðlogðUS1 ðTi ÞÞ;
logðUS2 ðTi ÞÞÞ;
(14)
and US1(Ti) and US2(Ti) are modeled through Eqs. (8) and (9),
respectively. Matlab provides a function, fminunc(), for solving this
type of unconstrained minimization over a generic function. The
results in Fig. 3a, b were derived, using the global optimization
approach, applied separately to individual alloys, for the purpose
of obtaining a tighter fit and more accurate estimation of Tbreak,
than for separate optimization over the low and high-temperature
regimes.
It is worth noting that previous models for the temperature
dependence of the YS only accounted for a single exponential22,23.
Hence, there was no break temperature, Tbreak. We consider the
break point critical for the optimization of the high-temperature
properties of alloys. For reliable operation, the temperature of
turbine blades made out of refractory alloys may need to stay
below Tbreak (not accounting for coatings17). Once the temperature of the turbine blades exceeds Tbreak, undesirable phase
transformations (e.g., dissolution of strengthening precipitates)
may start to take place21, and the alloy may begin to lose
structural integrity. Here, the second exponential, modeled
through US2(T), may prove detrimental. Turbine blades made
out of certain alloys, such as Ni-based superalloys, should only be
operated above Tbreak, if supported by extensive experimental test
results. We consider the break point, Tbreak, an important
parameter for the design of materials with attractive hightemperature properties, one warranting inclusion in alloy
specifications. Hence, it is important to be able to accurately
estimate Tbreak, e.g., using the global optimization approach
presented [Eq. (13)].
Senkov et al. provide physics-based foundation for the
prediction model in Eqs. (7)–(12)21. According to Senkov et al.,
the diffusion processes required to initiate phase transformations
generally become noticeable at temperatures T > 0.4 Tm, while at
T < 0.4 Tm the phase transformations are kinetically restricted21.
The atoms cannot move out of the lattice, and no phase
transformations can take place. This trend applies to lowentropy alloys, medium-entropy alloys, and HEAs, and serves to
explain the two exponentials. A solid-solution strengthening
model by Rao et al.28,29 does not take into account diffusion
effects and agrees well with the experimental data only at
relatively low temperatures, where diffusion-controlling deformation mechanisms can be ignored. Then, as the temperature
increases, the chemical bonds between the elements become
softer. The diffusion-controlled regime generally occurs
above ~0.5–0.6 Tm. It can be distinguished from the lowtemperature regime by a more rapid drop in strength with
increasing temperature, because dislocations are able to move
more easily around obstacles30.
A related model for the prediction of YS over temperature was
presented by Wu et al.22. The authors separately analyzed the
temperature dependencies of the YS and strain hardening of a
family of equi-atomic binary, ternary, and quaternary alloys based
on the elements, Fe, Ni, Co, Cr, and Mn, which had been shown to
form single-phase FCC solid solutions. The authors presented a
model with a single exponential for the overall YS, σy (T), of the
form22
T
σy ðT Þ ¼ σa exp
(15)
þ σb ;
C
where σ a, C, and σ b were fitting coefficients. The authors showed
that lattice friction appeared to be the predominant component of
the temperature-dependent YS, possibly because of the Peierls
barrier height decreasing with increasing temperature, due to a
thermally induced increase in dislocation width. Note, while
similar to the YS, we are here modeling the US.
According to Maresca et al., the YS of the solid-solution BCC
matrix alloy constitutes the major part of the alloy and can be
estimated by23:
"
#
1
kB T ε_ 0 0:91
;
(16)
τ y ðT; ε_ Þ ¼ τ y0 exp
ln
0:55 ΔEb ε_
where τy0 is the zero-temperature flow stress, ΔEb is the energy
barrier for dislocation movement, T is the absolute temperature, ε_
is the strain rate, and kB is the Boltzmann constant. For accurate
modeling of the YS, it is important to consider dislocations, atomic
and volume misfits.
Depending on the grain sizes and compositions involved, a
trilinear log model may yield a better fit for certain alloys31:
USðT Þ ¼ minðUS1 ðT Þ;
US2 ðT Þ; US3 ðT ÞÞ;
(17)
US1 ðT Þ ¼ expð C1 T=Tm þ C2 Þ;
0 < T < Tbreak1 ;
(18)
US2 ðT Þ ¼ expð C3 T=Tm þ C4 Þ;
Tbreak1 < T < Tbreak2 ;
(19)
US3 ðT Þ ¼ expð C5 T=Tm þ C6 Þ;
Tbreak2 < T < Tm ;
(20)
Published in partnership with the Shanghai Institute of Ceramics of the Chinese Academy of Sciences
npj Computational Materials (2021) 152
B. Steingrimsson et al.
6
Table 2.
No.
Quantification of ability of compositions to retain ultimate strength at high temperatures.
Alloy
Solvus temperature (oC)
Tbreak (oC)
C3: slope for high-temp.
regime in Fig. 3b
MSE: two-exponential
MSE: single exponential
Log
Log
Linear
Linear
1
Al0.3NbTa0.8Ti1.4V0.2Zr1.3
2043
800
13.77
1.7e–05
2
Al0.3NbTaTi1.4Zr1.3
2088
800
4.95 (only 3 data points)
0.039
31
0.130
2588
0.034
3
4
Al0.4Hf0.6NbTaTiZr
Al0.5NbTa0.8Ti1.5V0.2Zr
2124
1992
927.3
800
12.90
13.25
4.4e–05
9.2e–11
2455
102
0
0.146
0.151
72,312
47,797
5
AlCr0.5NbTiV
1704
769.3
18.30
6
AlCrNbTiV
1725
798.9
19.35
7.4e–09
0
0.504
186,447
6.0e–05
57
0.501
7
AlCr1.5NbTiV
1741
806.3
20.00
6.2e–08
203,776
0
0.211
143,719
8
AlCrMoNbTi
1867
943.7
15.80
0.002
1706
0.230
106,796
9
AlMo0.5NbTa0.5TiZr
1896
770.1
8.93
0.001
1195
0.228
502,343
8.19
4.2e–08
22
31,363
10
AlNb1.5Ta0.5Ti1.5Zr0.5
1863
779.5
0.080
47,474
11
12
AlNbTiV
AlNbTiVZr
1679
1714
Need more data
Need more data
0.007
0.070
38,994
78,813
13
AlNbTiVZr0.1
1683
Need more data
0.001
737
14
AlNbTiVZr0.25
1689
Need more data
0.012
90,556
15
AlNbTiVZr0.5
1698
Need more data (only 2 data points)
0.000
0
16
AlNbTiVZr1.5
1727
Need more data (only 2 data points)
0.000
0
17
CrMo0.5NbTa0.5TiZr
2145
923.5
12.86
5.5e–08
0.175
97,285
18
HfMoNbTiZr
2297
948.0
13.96
0.002
19
20
HfNbSi0.5TiVZr
MoNbTaVW
1973
2690
527.5
970.2
17.81
4.85
1.2e–07
0.001
21
MoNbTaW
2885
1124.8
2.75–7.82
Average (No. 1–10 and 17–21):
0
969
0.162
94,093
0
1229
0.304
0.017
236,176
30,751
1.3e–11
14
0.045
36,023
2.9e–3
528
0.195
122,587
Tbreak refers to the breaking point between bilinear log models, defined in Fig. 3.
Given an anomalous yield stress phenomenon in a CMSX-4,
single-crystal, Nickel-based superalloy, three exponentials are
needed for accurate modeling in case of Heat Treatment A, but
four exponentials in case of Heat Treatment B, according to
Supplementary Fig. 4. This phenomenon manifests itself as a
hump between the low-temperature and high-temperature
regimes, found in superalloys strengthened by L12-ordered
intermetallics. Here, the increased strength of γ′ phase with
temperature is explained by thermally activated cross-slip of
dislocations from {111} planes to {100} planes. Supplementary
Tables 4 and 5 present a practical approach to model selection
suitable for this case. We stop increasing the model order, once
the mean squared error (MSE) starts to taper off. Supplementary
Figs. 5 and 6 capture an application of piecewise linear regression
needed to address challenges imposed by non-convexity of the
objective function possible in this case. Here we expand the
parameter set such as to explicitly include the break temperatures.
Supplementary Figs. 7 and 8 contain Matlab pseudo-code for the
bilinear log model (a convex case) and a trilinear log model
(a possibly non-convex case).
In Supplementary Note 7, we provide physics-based reasons
explaining why a bilinear model will likely suffice for refractory
HEAs32. We also address the number of data points needed for
modeling. Since the hump between the low-temperature and
high-temperature regimes originates from the γ′ phase (which is a
L12 phase, i.e., ordered FCC structure), and since most refractory
HEAs contain BCC or hexagonal-closed-packed phases, with totally
different dislocation systems, it is unlikely that cross-slip from
{111} to {100} planes will happen in refractory HEAs32. Hence, we
expect the bilinear log model (two exponentials) to suffice for
most refractory HEAs.
npj Computational Materials (2021) 152
Table 2 further characterizes the ability of the 21 compositions
under consideration to retain strength at high temperatures, both
in terms of a high break temperature, Tbreak, and a small slope, C3.
Table 2 also compares the modeling accuracy of the bilinear log
model to that of a single exponential. It is not surprising that the
composition, MoNbTaW, which consists of strongly refractory
elements (Mo, Nb, Ta, and W), i.e., elements with the melting point
above 2200 °C33, yields the highest Tbreak of 1124 °C. This evidence
serves to validate the model. MoNbTaVW, which includes one
weakly refractory element (V), i.e., an element with a melting point
above 1850 °C, results in the smallest slope, C3, of 4.85, compared
to 7.82 for MoNbTaW. But this observation assumes omitting the
data point at T = 1600 °C, as a result of data curation. MoNbTaW
will result in the lowest slope (C3 = 2.75), if this data point is
included. This trend underscores the importance of considering
dislocations, interactions between elements, volume or lattice
misfit, and atomic mismatch23, when designing materials with
attractive high-temperature properties, in addition to the melting
points of the constituent elements. Similarly, it is not surprising
that the composition, AlMo0.5NbTa0.5TiZr, which additionally
contains the weakly refractory elements, Ti and Zr, also results
in a small slope (C3 = 4.95). While AlMo0.5NbTa0.5TiZr does seem to
offer relatively favorable high-temperature properties, it is worth
noting that the estimation of its slope is only based on three
data points.
In terms of the modeling accuracy, the bilinear log model yields
the average MSE of 0.003 in the log domain, for composition No.
1–10 and 17–21 from Table 2, compared to 0.195 for the model
with a single exponential. In the linear domain, this trend
translates to MSE of 528 for the bilinear log model, compared to
that of 122,588 for the single-exponential model. For the case of
composition No. 18 from Table 2 (HfMoNbTiZr), Fig. 4 provides
Published in partnership with the Shanghai Institute of Ceramics of the Chinese Academy of Sciences
B. Steingrimsson et al.
7
Fig. 5 Engineering stress–strain curves of the reference (dashed
lines) and predicted (solid lines) compositions. For estimation of
the error margins, refer to Table 1.
Fig. 4 Quantification of modeling accuracy of the bilinear log
model, for the composition, No. 18, from Table 2 (HfMoNbTiZr),
and comparison of the modeling accuracy to that of a model with
a single exponential. Panel (a) compares the modeling accuracy of
the bilinear log model to that of a single exponential in the
logarithmic domain, whereas panel (b) presents corresponding
comparison for the linear domain.
graphical insight as to why the bilinear log model yields a better
match to the data available than a model consisting of a single
exponential. Supplementary Figs. 9–23 provide similar diagrams
for the other 20 alloy compositions from Table 2.
DISCUSSION AND CONCLUSION
For interpretation of the prediction results, we refer to Section 4.7
of17. To analyze consistency with experimental verification, we
point to Supplementary Table 2, which captures the outcomes
from applying the empirical rules of19,20 to the formation of the
compositions predicted, Al0.5Mo0.5Nb1.5Ta0.5Zr1.5, Mo1.25Nb1.25Ti0.5V0.75Zr, Mo1.25Nb1.25Ti0.5V0.5Zr1.25, and MoNbZr. We expect
Al0.5Mo0.5Nb1.5Ta0.5Zr1.5 to be a stable composition with two types
of phases (BCC1 + BCC2). Supplementary Table 2 suggests that
the compositions have high chance of forming a solid-solution
main phase with ordered solid-solution precipitates.
Compression tests were conducted on both the reference and
predicted compositions in the as-cast condition. Figure 5
summarizes the engineering stress–strain curves of the predicted
compositions, including Al0.5Mo0.5Nb1.5Ta0.5Zr1.5 and Mo1.25Nb1.25Ti0.5V0.5Zr1.25, in comparison to the respective references,
Al0.5Mo0.5NbTa0.5TiZr and MoNbTiV0.75Zr. The compression properties of these alloys, such as the YS, σy, maximum strength, σmax,
and fracture strain, εf, are listed in Table 1. We conclude from the
experimental results in Fig. 5 that the candidate compositions,
Al0.5Mo0.5Nb1.5Ta0.5Zr1.5, Mo1.25Nb1.25Ti0.5V0.75Zr, and Mo1.25Nb1.25Ti0.5V0.5Zr1.25, indeed exhibit higher strengths than the
respective reference, hence confirming the outcome of our two
sets of prediction.
Figure 6 shows the energy dispersive X-Ray spectroscopy (EDX)
mapping for the predicted alloys. It can be noted that both of the
predicted compositions feature typical dendrite-inter-dendrite
microstructure, which indicates elemental segregation during
solidification with high cooling rates. Al0.5Mo0.5Nb1.5Ta0.5Zr1.5
exhibits segregation in all the five elements. The X-ray diffraction
(XRD) results in Fig. 7 indicate that both of the predicted alloys
contain two BCC phases. These results are consistent with known
properties of BCC and FCC phases, in terms of the BCC phases
usually helping improve material strength, but the FCC phases
helping improve ductility.
In this study, we proposed a bilinear log model for predicting
the US of HEAs across temperatures and evaluated its effectiveness for 21 compositions. We considered the break temperature,
Tbreak, an important parameter for design of materials with
attractive high-temperature properties, one warranting inclusion
in alloy specifications. For reliable operation, the operating
temperature for the corresponding alloys may need to stay below
Tbreak. Previous models for temperature dependence of the YS
only accounted for a single exponential. Hence, there was no
break temperature.
We, further, suggested a general methodology for joint
optimization, a methodology capable of accounting for physicsbased dependencies, and presented the maximization of the US
as an initial step toward the joint optimization of mechanical
properties. We applied an optimization technique suitable for the
problem under study, linear regression analysis, to a data set of
modest size from the literature, to predict HEA compositions
yielding the exceptional US at room temperature. For accurate
prediction, we recommended picking an optimization technique
appropriate for the application at hand and the data available and
carefully accounting for the underlying sources of variations.
Despite relatively limited data and a simple prediction algorithm17,
we were able to attain the goal of successfully predicting HEA
compositions, exhibiting superior strength, compared to previous
work, and to demonstrate consistency of our prediction, both with
empirical rules (in Table 16 of17) and with an experimental finding
(per Fig. 5). In this way, we successfully addressed the research
objective of predicting compositions of HEAs yielding the best
strength, and conducting experimental verification of our findings.
Next, one needs to account for the ductility. In case of the
Published in partnership with the Shanghai Institute of Ceramics of the Chinese Academy of Sciences
npj Computational Materials (2021) 152
B. Steingrimsson et al.
8
Fig. 6 EDX mapping and chemical compositions of the predicted compositions, Al0.5Mo0.5Nb1.5Ta0.5Zr1.5 and Mo1.25Nb1.25Ti0.5V0.5Zr1.25.
For panel (a), the brighter regions are rich in Mo, Nb, and Ta, while the darker regions are Al and Zr enriched. The other predicted composition,
Mo1.25Nb1.25Ti0.5V0.5Zr1.25, in panel (b) exhibits a nearly homogenized distribution of Ti and V. The dendrite regions are rich in Mo and Nb, and
the inter-dendrite regions are Zr enriched. The detailed elemental concentrations are also listed. A Zeiss EVO MA15 scanning electron
microscope with a back-scattered electron detector and Bruker xFlash 6130 energy dispersive X-ray spectroscopy was used for microstructural
and chemical composition analysis.
maximization of US and presence of a relatively small data set, we
recommended multi-variate linear regression as the method of
choice. In this case, the prediction rule is fairly general: One can
extrapolate in a direction of the gradient from the data point in
the training set exhibiting the highest US. As long as the step size
is selected as sufficiently small (only aiming for 5–10% increase in
US for a single step), the resulting prediction is considered much
superior to a trial-and-error approach. Sequential learning17 is
expected to greatly expedite the identification of alloys exhibiting
given mechanical properties of interest.
METHODS
Fig. 7 XRD patterns for the predicted compositions. The two BCC
phases could be related to the segregated microstructures
observed in the EDX maps, which may contribute to the high
strengths of the alloys due to the solid-solution strengthening
effect and the second phase strengthening effects. For comparison
and consistency, note that (according to Table 13 of17) the training
set consisted of a combination of single-phase and multi-phase—
mostly BCC compositions. A Panalytical Empyrean X-ray diffractometer, at Cu Kα radiation, was used to identify the crystal structure
of the alloys.
npj Computational Materials (2021) 152
While the primary emphasis here is on the US, the optimization of the
mechanical properties is assumed to take place within a framework for
joint optimization. We essentially take the US to represent the ultimate
compression strength, since Supplementary Table 1 contains 34 compression measurements, but only 1 tensile measurement. For further
explanations, refer to Supplementary Table 1.
The spider chart in Fig. 8 provides a high-level depiction of the content
of the database used for the present research. The database currently
captures mechanical properties of some 218 US recordings for HEA
configurations, many of which contain refractory elements, with 147
configurations measured at room temperature and 71 at elevated
temperatures. The US was measured using tensile testing for 29 of the
recordings. But the remaining 189 recordings were obtained through
compression testing.
Figure 2 captures an abstracted model of physical dependencies for the
prediction of the US. This model is an extension of the input sources
modeled in Eq. (21). Capturing of the physical dependencies helps greatly
in terms of the incorporation of a priori knowledge, derived from the
Published in partnership with the Shanghai Institute of Ceramics of the Chinese Academy of Sciences
B. Steingrimsson et al.
9
deriving the system model17. For a parametrized description of the terms
in Eq. (22), including composition, manufacturing processes, microstructures, and defects, and for support for the model in context with theory on
ML, refer to Supplementary Notes 3–5.
Experimental validation
The alloys predicted were prepared by arc-melting a mixture of pure
elements [purity >99.9 weight percent (wt.%)] in a Ti-gettered argon
atmosphere. The ingots were flipped and remelted for at least five times to
achieve homogenized elemental distributions. The ingots were cast into a
water-cooled copper hearth and then cut into desired shapes for further
experiments. Compression tests were performed on a computer-controlled
uniaxial mechanical testing system with a servo hydraulic load frame at
default strain rate of 1 × 10–3 s–1.
DATA AVAILABILITY
The data in this paper, including those in the Supplementary Figures, can be
requested by contacting the corresponding authors (baldur@imagars.com or
drzhangy@ustb.edu.cn).
Fig. 8 Characterization of the HEA data contained in the database
used for the present research.
underlying physics, and in terms of making the most of the—usually
limited—input data available.
Our intent is to accurately capture the input sources that contribute to
variations in the US observed (to variations in the output). In the present
research, we model the input combination as
CODE AVAILABILITY
Matlab comprises the software package primarily used for this study. Supplementary
Figs. 7 and 8 contain Matlab source code for the objective functions optimized in
case of the bilinear or the trilinear log model, respectively.
Received: 27 February 2021; Accepted: 5 August 2021;
Input ¼ ðcomposition; T; process; defects; grain size; microstructureÞ:
(21)
Here, “defects” are defined broadly such as to include inhomogeneities,
impurities, dislocations, or unwanted features. “T” represents temperature.
Similarly, the term “microstructure” broadly represents microstructures, at
nano- or micro-scale, as well as phase properties. The term “process”
broadly refers both to manufacturing processes and post-processing.
Correspondingly, the term “grain size” refers to the distribution in grain
sizes. Section 4.4 of17 allows for dependence between input sources, and
Section 4.5 outlines the expected dependence of the US on the individual
input sources listed. Dependence amongst the inputs is further addressed
in Supplementary Note 2.
Methodology for maximization of the US
The overall methodology for predicting the US is presented in
Supplementary Fig. 2. We summarize the prediction model as follows:
US ¼ h½composition; T; process; defectsðprocess; T Þ; grainsðprocess; T Þ;
microstructureðprocess; T Þ:
(22)
If the US corresponding to a given input combination is known, one can
simply look up the known value. If the US corresponding to a given input
combination is not known, then a prediction step can be applied (e.g.,
interpolation or extrapolation).
The purpose of the data curation step in Supplementary Fig. 2 is to
ensure that input data to the prediction step are of the highest quality
possible17. Here, the intent is to look for outliers, suspected cases of
discrepancy, or incorrect data (data that one may not fully trust). Generally,
it is recommended to filter out data that have no relevance to the
application domain or the task at hand17.
The step in our methodology for maximization of the predicted US, ~y ,
assumes a generic model of the type
~y ¼ hð~
xÞ:
(23)
Here, the input vector, ~
x, can be considered as the definition of a feature
set comprising of parameters related to the compositions, temperature,
heat-treatment process, defect property, grain size, microstructure (phase
properties), manufacturing process, or post-processing, essentially all
the source parameters that impact the output quantity of interest. The
~. Artificial
transformation, h(·), can be a non-linear function of the input, x
intelligence and supervised ML are presented as one of the alternatives for
REFERENCES
1. Miracle, D. B. & Senkov, O. N. A critical review of high entropy alloys and related
concepts. Acta Mater. 122, 448–511 (2017).
2. Yeh, J.-W. et al. Nanostructured high‐entropy alloys with multiple principal elements: novel alloy design concepts and outcomes. Adv. Eng. Mater. 6, 299–303
(2004).
3. Cantor, B., Chang, I., Knight, P. & Vincent, A. Microstructural development in
equiatomic multicomponent alloys. Mater. Sci. Eng.: A 375, 213–218 (2004).
4. Gao, M. C., Yeh, J.-W., Liaw, P. K. & Zhang, Y. High Entropy Alloys – Fundamentals
and Applications (Springer, 2016).
5. Zhang, Y. et al. Microstructures and properties of high-entropy alloys. Prog. Mater.
Sci. 61, 1–93 (2014).
6. Senkov, O. N., Wilks, G. B., Miracle, D. B., Chuang, C. P. & Liaw, P. K. Refractory
high-entropy alloys. Intermetallics 18, 1758–1765 (2010).
7. George, E. P., Raabe, D. & Ritchie, R. O. High-entropy alloys. Nat. Rev. Mater. 4,
515–534 (2019).
8. Hemphill, M. A. et al. Fatigue behavior of Al 0.5 CoCrCuFeNi high entropy alloys.
Acta Mater. 60, 5723–5734 (2012).
9. Tang, Z. et al. Fatigue behavior of a wrought Al 0.5 CoCrCuFeNi two-phase highentropy alloy. Acta Mater. 99, 247–258 (2015).
10. Shukla, S., Wang, T., Cotton, S. & Mishra, R. S. Hierarchical microstructure for
improved fatigue properties in a eutectic high entropy alloy. Scr. Mater. 156,
105–109 (2018).
11. Liu, K., Nene, S. S., Frank, M., Sinha, S. & Mishra, R. S. Metastability-assisted fatigue
behavior in a friction stir processed dual-phase high entropy alloy. Mater. Res.
Lett. 6, 613–619 (2018).
12. Liu, K., Nene, S. S., Frank, M., Sinha, S. & Mishra, R. S. Extremely high fatigue
resistance in an ultrafine grained high entropy alloy. Appl. Mater. Today 15,
525–530 (2019).
13. Suzuki, K., Koyama, M. & Noguchi, H. Small fatigue crack growth in a high entropy
alloy. Procedia Struct. Integr. 13, 1065–1070 (2018).
14. Kim, Y.-K., Ham, G.-S., Kim, H. S. & Lee, K.-A. High-cycle fatigue and tensile
deformation behaviors of coarse-grained equiatomic CoCrFeMnNi high entropy
alloy and unexpected hardening behavior during cyclic loading. Intermetallics
111, https://doi.org/10.1016/j.intermet.2019.106486 (2019).
15. Kashaev, N. et al. Fatigue behaviour of a laser beam welded CoCrFeNiMn-type
high entropy alloy. Mater. Sci. Eng.: A 766, https://doi.org/10.1016/j.
msea.2019.138358 (2019).
16. Guennec, B. et al. Four-point bending fatigue behavior of an equimolar BCC
HfNbTaTiZr high-entropy alloy: macroscopic and microscopic viewpoints. Materialia 4, 348–360 (2018).
Published in partnership with the Shanghai Institute of Ceramics of the Chinese Academy of Sciences
npj Computational Materials (2021) 152
B. Steingrimsson et al.
10
17. Steingrimsson, B., Fan, X., Kulkarni, A., Gao, M. C. & Liaw, P. K. Machine learning and
data analytics for design and manufacturing of high-entropy materials exhibiting
mechanical or fatigue properties of interest, (eds Liaw, P. K. & Brechtl, J.) Chapter 4
In Fundamental Studies in High-Entropy Materials (Springer, 2021).
18. Agrawal, A. et al. Exploration of data science techniques to predict fatigue
strength of steel from composition and processing parameters. Integr. Mater.
Manuf. Innov. 3, 8 (2014).
19. Zhang, Y., Zhou, Y. J., Lin, J. P., Chen, G. L. & Liaw, P. K. Solid-solution phase
formation rules for multi-component alloys. Adv. Eng. Mater. 10, 534–538 (2008).
20. Feng, R. et al. Design of light-weight high-entropy alloys. Entropy 18, 333 (2016).
21. Senkov, O. N., Gorsse, S. & Miracle, D. B. High temperature strength of refractory
complex concentrated alloys. Acta Mater. 175, 394–405 (2019).
22. Wu, Z., Bei, H., Pharr, G. M. & George, E. P. Temperature dependence of the
mechanical properties of equiatomic solid solution alloys with face-centered
cubic crystal structures. Acta Mater. 81, 428–441 (2014).
23. Maresca, F. & Curtin, W. A. Mechanistic origin of high strength in refractory BCC
high entropy alloys up to 1900K. Acta Mater. 182, 235–249 (2020).
24. MacKay, D. J. Bayesian methods for neural networks: theory and applications,
Neural Networks Summer School, University of Cambridge Programme for
Industry http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.47.6409 (1995).
25. Zhang, Y., Yang, X. & Liaw, P. K. Alloy design and properties optimization of highentropy alloys. Jom 64, 830–838 (2012).
26. Ben-Israel, A. & Greville, T. N. Generalized Inverses: Theory and Applications, Vol. 15
(Springer Science & Business Media, 2003).
27. Le, T. T., Fu, W. & Moore, J. H. Scaling tree-based automated machine learning to
biomedical big data with a feature set selector. Bioinformatics 36, 250–256 (2020).
28. Rao, S. I. et al. Solution hardening in body-centered cubic quaternary alloys interpreted using Suzuki’s kink-solute interaction model. Scr. Mater. 165, 103–106 (2019).
29. Rao, S. I. et al. Modeling solution hardening in BCC refractory complex concentrated alloys: NbTiZr, Nb1.5TiZr0.5 and Nb0.5TiZr1.5. Acta Mater. 168, 222–236
(2019).
30. Caillard, D. & Martin, J.-L. Thermally Activated Mechanisms in Crystal Plasticity.
(Elsevier, 2003).
31. Otto, F. et al. The influences of temperature and microstructure on the tensile
properties of a CoCrFeMnNi high-entropy alloy. Acta Mater. 61, 5743–5755 (2013).
32. Diao, H. Y., Feng, R., Dahmen, K. A. & Liaw, P. K. Fundamental deformation
behavior in high-entropy alloys: an overview. Curr. Opin. Solid St. Mater. Sci. 21,
252–266 (2017).
33. Wilson, J. General behaviour of refractory metals. Behavior and Properties of
Refractory. (Stanford University Press, 1965).
34. Steingrimsson, B. A., Liaw, P. K., Fan, X. & Kulkami, A. A. (United States Patent
Application Publication, 2020).
35. Miracle, D. B. et al. ASM Handbook, Vol. 21 (ASM international Materials Park, OH,
2001).
ACKNOWLEDGEMENTS
X.F. and P.K.L. very much appreciate the support of the U.S. Army Research Office
Project (W911NF-13-1-0438 and W911NF-19-2-0049) with the program managers,
Drs M.P. Bakas, S.N. Mathaudhu, and D.M. Stepp, as well as the support from the
Bunch Fellowship. XF and PKL also would like to acknowledge funding from the State
of Tennessee and Tennessee Higher Education Commission (THEC) through their
support of the Center for Materials Processing (CMP). P.K.L., furthermore, thanks the
npj Computational Materials (2021) 152
support from the National Science Foundation (DMR-1611180 and 1809640) with the
program directors, Drs J. Yang, G. Shiflet, and D. Farkas. B.S. very much appreciates
the support from the National Science Foundation (IIP-1447395 and IIP-1632408),
with the program directors, Dr G. Larsen and R. Mehta, from the U.S. Air Force
(FA864921P0754), with J. Evans as the program manager, and from the U.S. Navy
(N6833521C0420), with Drs D. Shifler and J. Wolk as the program managers. M.C.G.
acknowledges the support of the US Department of Energy’s Fossil Energy
Crosscutting Technology Research Program. The authors also want to thank
Dr. G. Tewksbury for bringing to their attention suspicious recordings of the US
from the literature, which have prompted the data curation effort.
AUTHOR CONTRIBUTIONS
B.S., X.F., and P.K.L. conceived the project. B.S. performed the ML predictions, but
consulted with M.C.G. in the process. X.F. prepared the database and conducted
experimental verification of the ML predictions. All authors edited and proofread the
final manuscript and participated in discussions.
COMPETING INTERESTS
The authors declare no competing interests.
ADDITIONAL INFORMATION
Supplementary information The online version contains supplementary material
available at https://doi.org/10.1038/s41524-021-00623-4.
Correspondence and requests for materials should be addressed to B. Steingrimsson
or Y. Zhang.
Reprints and permission information is available at http://www.nature.com/
reprints
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims
in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons
Attribution 4.0 International License, which permits use, sharing,
adaptation, distribution and reproduction in any medium or format, as long as you give
appropriate credit to the original author(s) and the source, provide a link to the Creative
Commons license, and indicate if changes were made. The images or other third party
material in this article are included in the article’s Creative Commons license, unless
indicated otherwise in a credit line to the material. If material is not included in the
article’s Creative Commons license and your intended use is not permitted by statutory
regulation or exceeds the permitted use, you will need to obtain permission directly
from the copyright holder. To view a copy of this license, visit http://creativecommons.
org/licenses/by/4.0/.
© The Author(s) 2021
Published in partnership with the Shanghai Institute of Ceramics of the Chinese Academy of Sciences