Prediction of The Effective Elastic Modu
Prediction of The Effective Elastic Modu
Prediction of The Effective Elastic Modu
a r t i c l e i n f o a b s t r a c t
Article history: Statistical modeling is used to correlate geometric parameters of pores with their contributions to the
Received 7 October 2013 overall Young’s moduli of linearly elastic solids. The statistical model is based on individual pore
Received in revised form 21 March 2014 contribution parameters evaluated by finite element simulations for a small pore subset selected using
Available online 12 April 2014
the design of experiments approach, so there is no need to solve the elasticity problem for all pores in
the material. A polynomial relating pore geometric parameters to the contribution parameters is then
Keywords: fitted to the results of the simulations. We found a good correlation between normalized projected areas
Stochastic approach
of the pores on three coordinate planes and their contributions to the corresponding effective Young’s
Effective elastic moduli
Irregular pores
moduli. The model is applied and validated for two large sets of pore geometries obtained by X-ray
Micromechanical modeling microcomputed tomography of a carbon/carbon and a 3D woven carbon/epoxy composite specimens.
Ó 2014 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijsolstr.2014.03.042
0020-7683/Ó 2014 Elsevier Ltd. All rights reserved.
2688 B. Drach et al. / International Journal of Solids and Structures 51 (2014) 2687–2695
Zimmerman (1991), for example, showed that compressibility use the design of experiments (DoE) approach (Ryan, 2007;
of 2D pores correlates well with the undimensionalized ratio of Myers et al., 2009) to identify the list of input parameters (in our
pore perimeter P to its area A defined as P2 =A (a three-dimen- case, pore shape geometric parameters) that corresponds to the
sional analog of this parameter, the surface-to-volume ratio, was specified level of model prediction variance across the whole
used in the stochastic model of Drach et al., 2013). Zimmerman’s parameter space. We design the ‘‘experiment’’ (the combination
observation was later supported by numerical studies on several of irregular shapes for FE simulations) using an I-optimal design
irregular 2D shapes (Tsukrov and Novak, 2002). However, such a module in JMP software (SAS Institute Inc, 2010). The optimization
parameter cannot be considered universal. For example, substitu- criterion for I-optimal designs is minimization of the integrated
tion of smooth pore boundaries by the very jagged ones (as illus- variance of the model predictions over the entire design space. This
trated in Fig. 1) will significantly increase the surface-to-volume means that the expected width of confidence intervals for model
ratio without making big changes in the overall response. This predictions will be approximately the same across the full param-
immediately follows from the comparison (or ‘‘auxiliary’’) theorem eter space.
of Hill (1965) as interpreted by Kachanov and Sevostianov (2005). We apply this method to two sets of pore shapes obtained by
According to their interpretation, contributions of two pores with microcomputed tomography of two different material samples:
very different surface-to-volume ratios shown by solid lines in carbon/carbon and 3D woven carbon/epoxy composites. The exam-
Fig. 1 are bounded by the same contributions of pores shown with ples of pore shapes are given in Fig. 2. For both sets we compare
dashed lines. The influence of corrugated boundaries on the overall contributions of the pores to effective elastic properties deter-
compressibility of pores is also discussed in Ekneligoda and mined by the direct FEA simulations for each pore with the ones
Zimmerman (2008a). based on its projected areas. Note that we use these sets of pore
Another possible set of geometric parameters to characterize shapes only to check the hypothesis that statistical estimates based
contribution of a pore to the effective elastic properties is its prin- on the projected areas can be used instead of elasticity solutions.
cipal moments of inertia (PMIs). PMIs are defined as the eigen- Our predictions cannot be directly used for carbon/carbon and
values of the matrix of moments of inertia given by 3D woven carbon/epoxy composites, because in our analysis we
Iij ¼ V qðrÞðr 2 dij xi xj ÞdV, where qðrÞ is the material density of
R
assume the pores to be placed in an isotropic homogeneous mate-
the body for which PMIs are calculated (pores can be treated as rial which is obviously not the case for these composites. However,
homogeneous bodies with qðrÞ ¼ q ¼ 1), r is the distance to the we feel that the considered pore shape sets provide good data to
axis around which the moment is calculated, dij is the Kronecker test the hypothesis, as the pores in these two cases result from very
delta, xi are the coordinates x1 ; x2 ; x3 . PMIs have been previously different mechanical processes (porosity from incomplete filling by
used to approximate pores by the equivalent ellipsoids (Li et al., chemical vapor deposition of pyrolytic carbon vs. damage due to
1999; Drach et al., 2013). In Drach et al. (2013) we utilized statis- chemical and thermal shrinkage of epoxy) and thus have different
tical approaches to correlate pore principal moments of inertia to morphologies and shape distributions.
their contribution to the effective elastic moduli.
In this paper we investigate the hypothesis that contribution 2. Micromechanical modeling approach utilizing statistical
of irregularly-shaped pores to the effective Young’s moduli of analysis
porous material can be evaluated based on their projected areas
(‘‘shadows’’) by statistical analysis of effective compliance Characterization of individual pore contributions to the effec-
parameters. We begin by selecting a certain subset of pores from tive material response is based on the pore compliance contribu-
the full dataset. Then, we perform finite element (FE) simulations tion tensor – H -tensor proposed in Kachanov et al. (1994). Note
for the pores in the subset to quantify their individual contribu- that a similar tensor in the context of a microcrack was earlier
tions to the effective compliance of the material. This data is given by Horii and Nemat-Nasser (1983). The fourth rank tensor
then used in regression analysis to construct an approximating H is defined as a set of proportionality coefficients between remo-
polynomial model for the dependence of pore’s contribution to tely applied homogeneous stress field r0 and the additional strain
the effective compliance on its projected area. To verify that the De generated in the material due to the presence of a cavity:
model works for the full dataset, the constructed model is then
checked against the FE simulations on the new subset of randomly De ¼ H : r 0 ð1Þ
chosen irregular shapes not used in the model fitting. where ‘‘:’’ denotes the contraction over two indices. For a material
Because the accuracy of the model predictions (width of the with a large number of pores, a proper representative volume ele-
error bars) depends on the number of direct simulations used in ment (RVE) (Hill, 1963; Nemat-Nasser and Hori, 1999) can be
the data fitting, it is important to choose an ‘‘optimal’’ subset selected, and the effective compliance tensor of the material is
which corresponds to the desired model prediction variance. We given by
(a) (b)
Fig. 1. Two pores having similar contributions to the overall elastic properties but significantly different surface-to-volume ratios. Dashed lines show upper and lower limits
of their contributions.
B. Drach et al. / International Journal of Solids and Structures 51 (2014) 2687–2695 2689
Fig. 2. Examples of pores extracted by microcomputed tomography from (a) carbon/carbon composite sample; (b) 3D woven carbon/epoxy composite sample.
S ¼ S0 þ HRVE example Drach et al., 2011), this method may not be practical
when contributions of tens of thousands of different pores are to
where S0 is the compliance tensor of the matrix material and HRVE is
be estimated.
the contribution from all pores present in the RVE.
Our method involves statistical analysis to derive a formula pre-
For dilute distribution of pores, the non-interaction approxima-
dicting contribution of a pore to effective properties based on its
tion can be used, and HRVE is found by direct summation of contri-
geometric parameters. First, we run a relatively small number of
butions from all individual pores in the RVE:
FEA simulations (100–200) for individual pores in a reference
HNI volume. The set of pores is chosen as an ‘‘optimal’’ subset of data
X
RVE ¼ HðiÞ ð2Þ
i using the DoE approach so that the pore shape parameters allow
to obtain the model with a specified level of variance across the
where HðiÞ is the compliance contribution tensor of the ith pore. For whole parameter space. Then a polynomial is fitted to the simula-
higher porosities when the non-interaction approximation is no tion results and estimates for the rest of the pores are inferred from
longer applicable, more advanced micromechanical schemes should the resulting empirical formula. Thus, the statistical analysis
be used. However, most of the first order micromechanical schemes results in a polynomial approximation model, which correlates
can be readily expressed in terms of the non-interaction approxi- pore contributions with their geometric parameters. If m geometric
mation, see Eroshkin and Tsukrov (2005). For example, predictions parameters are chosen, and the degree of approximating polyno-
for the overall pore compliance contribution tensor by the Mori– mial is n, the polynomial has the following general form
Tanaka method (Mori and Tanaka, 1973; Benveniste, 1987) in terms
of HNI
RVE is given by a simple formula Kachanov et al. (1994): PCC ¼ a0 þ a1 F 1 þ a2 F 2 þ þ am F m þ amþ1 F 21 þ amþ2 F 22 þ
4.5 4.5
4 4
3.5 3.5
E1 Actual
E1 Actual
3 3
2.5 2.5
2 2
1.5 1.5
1.5 2 2.5 3 3.5 4 4.5 1.5 2 2.5 3 3.5 4 4.5
E1 Predicted E1 Predicted
RSq=0.96 RMSE=0.1287 RSq=0.96 RMSE=0.0599
4.5 4.5
4 4
3 3
2.5 2.5
2 2
1.5 1.5
1.5 2 2.5 3 3.5 4 4.5 1.5 2 2.5 3 3.5 4 4.5
E2 Predicted E2 Predicted
RSq=0.95 RMSE=0.1366 RSq=0.93 RMSE=0.0907
4.5 4.5
4 4
3.5 3.5
E3 Actual
E3 Actual
3 3
2.5 2.5
2 2
1.5 1.5
1.5 2 2.5 3 3.5 4 4.5 1.5 2 2.5 3 3.5 4 4.5
E3 Predicted E3 Predicted
RSq=0.96 RMSE=0.084 RSq=0.97 RMSE=0.1493
(a) (b)
~i Actual) vs. model prediction results (E
Fig. 6. The results of response surface fitting presented as plots of direct calculation results (E ~i Predicted) for pore compliance
contribution parameters: (a) carbon/carbon composite dataset; (b) 3D woven carbon/epoxy composite dataset.
For validation of the proposed stochastic 3-factor PCC model Pore geometries were selected from the original datasets with all
given by formulae (9), the estimates of pore compliance contribu- available pores. The results of the validation experiment are shown
tion parameters for each dataset were compared to the direct FEA in Fig. 7. The values of R2 and RMSE for carbon/carbon composite
simulations performed on an independently chosen set of 100 are R2 ¼ 0:96; 0:95; 0:92 and RMSE ¼ 4:6%; 4:7%; 5:3% for E ~1 , E
~2
pores which were not used for the model development. For this and E~3 responses, correspondingly. For 3D woven carbon/epoxy
experiment, the normalized projected area parameters were composite, the values of R2 and RMSE are R2 ¼ 0:96; 0:92; 0:94
selected randomly with uniform distribution over the design space. and RMSE ¼ 2:7%; 4:0%; 4:7% for E ~2 and E
~1 , E ~3 responses,
B. Drach et al. / International Journal of Solids and Structures 51 (2014) 2687–2695 2693
4.5 4.5
4 4
3.5 3.5
E1 Actual
E1 Actual
3 3
2.5 2.5
2 2
1.5 1.5
1.5 2 2.5 3 3.5 4 4.5 1.5 2 2.5 3 3.5 4 4.5
E1 Predicted E1 Predicted
R2 = 0.96 RMSE = 0.1152 R2 = 0.96 RMSE = 0.050 3
4.5 4.5
4 4
3.5 3.5
E2 Actual
E2 Actual
3 3
2.5 2.5
2 2
1.5 1.5
1.5 2 2.5 3 3.5 4 4.5 1.5 2 2.5 3 3.5 4 4.5
E2 Predicted E2 Predicted
2
R = 0.95 RMSE = 0.1180 R2 = 0.92 RMSE = 0.0827
4.5 4.5
4 4
3.5 3.5
E3 Actual
E3 Actual
3 3
2.5 2.5
2 2
1.5 1.5
1.5 2 2.5 3 3.5 4 4.5 1.5 2 2.5 3 3.5 4 4.5
E3 Predicted E3 Predicted
R2 = 0.92 RMSE = 0.1165 R2 = 0.94 RMSE = 0.1582
(a) (b)
~1 , E
Fig. 7. Validation of the statistical model for responses E ~2 and E
~3 performed on 100 pores not included in the original model: (a) carbon/carbon composite dataset; (b) 3D
woven carbon/epoxy composite dataset.
correspondingly. These results indicate that the proposed models The mean errors from the validation studies for predictions of E ~1 ,
can be used to estimate with reasonable accuracy the pore compli- ~2 and E
E ~3 are correspondingly 0.1%, 0.2%, 0.4% for 3-factor
ance contribution parameters of the remaining 8251 pores in the model, and 3.5%, 3.0% 1.9% for 4-factor model. Similarly, the
carbon/carbon composite dataset and 23,941 pores in the carbon/ root-mean-square error values are 4.6%, 4.7%, 5.3% for 3-factor
epoxy dataset that were not included in the original experiments. model, and 8.0%, 7.7%, 11.1% for 4-factor model. It can be seen that
The accuracy of the 3-factor model (based on the projected both the bias (mean error) and the scatter (RMSE) are lower for the
areas) for the carbon/carbon composite dataset can be compared model constructed using the normalized projected areas. We con-
to the 4-factor model (based on the principal moments of inertia clude that for the considered carbon/carbon composite dataset, the
and surface-to-volume ratio) presented in Drach et al. (2013). choice of projected areas as the model factors provides more
2694 B. Drach et al. / International Journal of Solids and Structures 51 (2014) 2687–2695
Table 3
Summary of data on accuracy of the considered statistical models.
Model E1 E2 E3
2 2 2
R ME RMSE R ME RMSE R ME RMSE
4-factor model (Drach et al., 2013) 0.87 0.0000 0.0805 0.75 0.0000 0.1084 0.82 0.0000 0.2836
(0.0%) (5.0%) (0.0%) (5.5%) (0.0%) (10.0%)
4-factor model validation (Drach et al., 2013) 0.69 0.0607 0.1389 0.55 0.0614 0.1579 0.55 0.0536 0.3106
(3.5%) (8.0%) (3.0%) (7.7%) (1.9%) (11.1%)
Proposed 3-factor model (carbon/carbon dataset) 0.96 0.0000 0.1287 0.95 0.0000 0.1366 0.96 0.0000 0.084
(0.0%) (5.1%) (0.0%) (5.5%) (0.0%) (3.9%)
Proposed 3-factor model validation (carbon/carbon dataset) 0.96 0.0030 0.1152 0.95 0.0040 0.1180 0.92 0.0082 0.1165
(0.1%) (4.6%) (0.2%) (4.7%) (0.4%) (5.3%)
Proposed 3-factor model (3D woven carbon/epoxy dataset) 0.96 0.0000 0.0599 0.93 0.0000 0.0907 0.97 0.0000 0.1493
(0.0%) (3.2%) (0.0%) (4.3%) (0.0%) (4.4%)
Proposed 3-factor model validation (3D woven carbon/epoxy dataset) 0.96 0.0044 0.0503 0.92 0.0024 0.0827 0.94 0.0393 0.1582
(0.2%) (2.7%) (0.1%) (4.0%) (1.2%) (4.7%)
accurate predictions of effective Young’s moduli as compared to CMMI-1100409. The microcomputed tomography data was pro-
the model based on the principal moments of inertia and sur- vided by Karlsruhe Institute of Technology (carbon/carbon com-
face-to-volume ratio. Note that another drawback of the 4-factor posite) and Albany Engineered Composites, Inc. (3D woven
model is that orientation of the pore principal axes needs to be carbon/epoxy composite). The New Hampshire Innovation
in the direction of the global axes, otherwise we would need to Research Center is acknowledged for initial support of the collabo-
include orientation vectors in the model. In the model based on ration between the University of New Hampshire and Albany Engi-
‘‘shadows’’, no such adjustment is necessary. Table 3 summarizes neered Composites, Inc.
the data on accuracy of predicting the pore contributions to the The authors would like to thank Philip Ramsey for valuable dis-
overall Young’s moduli based on 4 parameters (principal moments cussions on statistical analysis and design of experiment method-
of inertia and the surface-to-volume ratio) and 3 parameters (nor- ologies. The authors also express their gratitude to Robert
malized projected areas). Zimmerman for his insightful comments on compressibility and
overall elastic behavior of irregular pores. Andrew Drach gratefully
5. Conclusions acknowledges the financial support provided by the ICES Postdoc-
toral Fellowship.
A statistical model based on the normalized projected area
parameters is proposed for correlation of geometry of irregularly References
shaped pores with their contributions to the effective Young’s
moduli of the porous material. The model is applied to large data Arns, C.H., Knackstedt, M.A., Pinczewski, W.V., Garboczi, E.J., 2002. Computation of
sets of pores in two different composite materials. The Young’s linear elastic properties from microtomographic images: methodology and
agreement between theory and experiment. J. Geophys. 67, 1396–1405.
moduli in three orthogonal directions coinciding with the global Benveniste, Y., 1987. A new approach to the application of Mori–Tanaka’s theory in
coordinate system of the specimens are considered. In each case, composite materials. Mech. Mater. 6, 147–157.
one hundred pores were selected to construct the corresponding David, E.C., Zimmerman, R.W., 2011. Elastic moduli of solids containing spheroidal
pores. Int. J. Eng. Sci. 49, 544–560.
models. Design of experiments techniques were utilized to select
Dewey, J.M., 1947. The elastic constants of materials loaded with non-rigid fillers. J.
the pores producing minimal variability of model predictions Appl. Phys. D9, 665–675.
across the entire factor space. The constructed models for three Drach, B., Tsukrov, I., Gross, T., Dietrich, S., Weidenmann, K., Piat, R., Böhlke, T., 2011.
~1 , E
~2 , E
~3 have high values of R2 in both Numerical modeling of carbon/carbon composites with nanotextured matrix
Young’s moduli parameters E
and 3D pores of irregular shapes. Int. J. Solids Struct. 48, 2447–2457.
material systems (0.96, 0.95, 0.96 in carbon/carbon composite Drach, B., Drach, A., Tsukrov, I., 2013. Characterization and statistical modeling of
model, and 0.96, 0.93, 0.97 in 3D woven carbon/epoxy composite irregular porosity in carbon/carbon composites based on X-ray
model), as well as relatively low values of RMSE (5.1%, 5.5% and microtomography data. J. Appl. Math. Mech. 93, 346–366.
Ekneligoda, T.C., Zimmerman, R.W., 2006. Compressibility of two-dimensional
3.9% in carbon/carbon composite model and 3.2%, 4.3% and 4.4% pores having n-fold axes of symmetry. Proc. R. Soc. A Math. Phys. Eng. Sci. 462,
in 3D woven carbon/epoxy composite model). These results indi- 1933–1947.
cate that the proposed models, created based on 100 pore shapes Ekneligoda, T.C., Zimmerman, R.W., 2008a. Boundary perturbation solution for
nearly circular holes and rigid inclusions in an infinite elastic medium. J. Appl.
in each case, can be used to estimate with reasonable accuracy Mech. 75.
the Young’s moduli contribution parameters E ~2 and E
~1 , E ~3 for large Ekneligoda, T.C., Zimmerman, R.W., 2008b. Shear compliance of two-dimensional
data sets of pores, in our case more than 8000 in a sample of car- pores possessing N-fold axis of rotational symmetry. Proc. R. Soc. A Math. Phys.
Eng. Sci. 464, 759–775.
bon/carbon composite and more than 24,000 in a sample of 3D Eroshkin, O., Tsukrov, I., 2005. On micromechanical modeling of particulate
woven carbon/epoxy composite. One of the advantages of the pro- composites with inclusions of various shapes. Int. J. Solids Struct. 42, 409–427.
posed method is that each pore is considered in the global coordi- Eshelby, J.D., 1957. The determination of the elastic field of an ellipsoidal inclusion,
and related problems. Proc. R. Soc. A Math. Phys. Eng. Sci. 241, 376–396.
nate system of the specimen. In contrast to most presently used
Eshelby, J.D., 1959. The elastic field outside an ellipsoidal inclusion. Proc. R. Soc.
micromechanical approaches, there is no need for solving the elas- London Ser. A Math. Phys. Sci. 252, 561–569.
ticity problem in the coordinate system associated with the pore González, C., Segurado, J., LLorca, J., 2004. Numerical simulation of elasto-plastic
deformation of composites: evolution of stress microfields and implications for
followed by the coordinate transformation to the global coordinate
homogenization models. J. Mech. Phys. Solids 52, 1573–1593.
system. Thus, the only information needed to evaluate the pore’s Hashemi, R., Avazmohammadi, R., Shodja, H.M., Weng, G.J., 2009. Composites with
contribution to effective Young’s moduli is its projected areas in superspherical inhomogeneities. Philos. Mag. Lett. 89, 439–451.
the global coordinate system. Hashin, Z., Shtrikman, S., 1962. On some variational principles in anisotropic and
nonhomogeneous elasticity. J. Mech. Phys. Solids 10, 78–96.
Hill, R., 1963. Elastic properties of reinforced solids: some theoretical principles. J.
Acknowledgements Mech. Phys. Solids 11, 357–372.
Hill, R., 1965. A self-consistent mechanics of composite materials. J. Mech. Phys.
Solids 13, 213–222.
The authors gratefully acknowledge the financial support of the Horii, H., Nemat-Nasser, S., 1983. Overall moduli of solids with microcracks:
National Science Foundation through the grants DMR-0806906 and load-induced anisotropy. J. Mech. Phys. Solids 31, 155–171.
B. Drach et al. / International Journal of Solids and Structures 51 (2014) 2687–2695 2695
Jasiuk, I., Chen, J., Thorpe, M., 1994. Elastic moduli of two dimensional materials Onaka, S., 2001. Averaged Eshelby tensor and elastic strain energy of a
with polygonal and elliptical holes. Appl. Mech. Rev. 47, 18–28. superspherical inclusion with uniform eigenstrains. Philos. Mag. Lett. 81,
Kachanov, M., Sevostianov, I., 2005. On quantitative characterization of 265–272.
microstructures and effective properties. Int. J. Solids Struct. 42, 309–336. Roberts, A.P., Garboczi, E.J., 2000. Elastic properties of model porous ceramics. J. Am.
Kachanov, M., Tsukrov, I., Shafiro, B., 1994. Effective moduli of solids with cavities of Ceram. Soc. 83, 3041–3048.
various shapes. Appl. Mech. Rev. 47, S151. Rodin, G., 1996. Eshelby’s inclusion problem for polygons and polyhedra. J. Mech.
Lee, J.K., Johnson, W.C., 1978. Calculation of the elastic strain field of a cuboidal Phys. Solids 44, 1977–1995.
precipitate in an anisotropic matrix. Phys. Status Solidi A 46, 267–272. Ryan, T.P., 2007. Modern Experimental Design. John Wiley & Sons, Hoboken, NJ.
Li, M., Ghosh, S., Richmond, O., 1999. An experimental–computational approach to SAS Institute Inc, 2010. JMP 9 Design of Experiments Guide. SAS Institute Inc, Cary,
the investigation of damage evolution in discontinuously reinforced aluminum NC. <www.jmp.com/support/downloads/pdf/jmp902/doe_guide.pdf> (retrieved
matrix composite. Acta Mater. 47, 3515–3532. 28.09.11).
Lubarda, V.A., Markenscoff, X., 1998. On the absence of Eshelby property for non- Sevostianov, I., Giraud, A., 2012. On the compliance contribution tensor for a
ellipsoidal inclusions. Int. J. Solids Struct. 35, 3405–3411. concave superspherical pore. Int. J. Fract. 177, 199–206.
Mackenzie, J.K., 1950. The elastic constants of a solid containing spherical holes. Tsukrov, I., Kachanov, M., 1993. Solids with holes of irregular shapes: effective
Proc. Phys. Soc. (London) 63B, 2–11. moduli and anisotropy. Int. J. Fract. 64, R9–R12.
Mori, T., Tanaka, K., 1973. Average stress in matrix and average elastic energy of Tsukrov, I., Novak, J., 2002. Effective elastic properties of solids with defects of
materials with misfitting inclusions. Acta Metall. 21, 571–574. irregular shapes. Int. J. Solids Struct. 39, 1539–1555.
Myers, R.H., Montgomery, D.C., Anderson-Cook, C.M., 2009. Response Surface Wu, T.Te., 1966. The effect of inclusion shape on the elastic moduli of a two-phase
Methodology: Process and Product Optimization using Designed Experiments, material. Int. J. Solids Struct. 2, 1–8.
third ed. John Wiley and Sons, Hoboken, NJ. Zimmerman, R.W., 1986. Compressibility of two-dimensional cavities of various
Nemat-Nasser, S., Hori, M., 1999. Micromechanics: Overall Properties of shapes. J. Appl. Mech. 53, 500–504.
Heterogeneous Solids, second ed. Elsevier Science Publishers. Zimmerman, R.W., 1991. Compressibility of Sandstones. Elsevier.
Nozaki, H., Taya, M., 1997. Elastic fields in a polygon-shaped inclusion with uniform Zou, W., He, Q., Huang, M., Zheng, Q.-S., 2010. Eshelby’s problem of non-elliptical
eigenstrains. J. Appl. Mech. 64, 495–502. inclusions. J. Mech. Phys. Solids 58, 346–372.