Accepted Manuscript

Application of finite element, phase-field, and CALPHAD-based methods to additive

manufacturing of Ni-based superalloys

Trevor Keller, Greta Lindwall, Supriyo Ghosh, Li Ma, Brandon M. Lane, Fan Zhang,
Ursula R. Kattner, Eric A. Lass, Jarred C. Heigel, Yaakov Idell, Maureen E. Williams,
Andrew J. Allen, Jonathan E. Guyer, Lyle E. Levine
PII: S1359-6454(17)30380-4
DOI: 10.1016/j.actamat.2017.05.003
Reference: AM 13761

To appear in: Acta Materialia

Received Date: 16 March 2017

Revised Date: 1 May 2017
Accepted Date: 2 May 2017

Please cite this article as: T. Keller, G. Lindwall, S. Ghosh, L. Ma, B.M. Lane, F. Zhang, U.R. Kattner,
E.A. Lass, J.C. Heigel, Y. Idell, M.E. Williams, A.J. Allen, J.E. Guyer, L.E. Levine, Application of finite
element, phase-field, and CALPHAD-based methods to additive manufacturing of Ni-based superalloys,
Acta Materialia (2017), doi: 10.1016/j.actamat.2017.05.003.

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and all
legal disclaimers that apply to the journal pertain.
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D
TE
EP
C
AC
 ACCEPTED MANUSCRIPT

1 Application of Finite Element, Phase-field, and CALPHAD-based Methods

2 to Additive Manufacturing of Ni-based Superalloys

3 Trevor Kellera,∗, Greta Lindwalla , Supriyo Ghosha , Li Maa,b , Brandon M. Lanec , Fan Zhangd ,
4 Ursula R. Kattnera , Eric A. Lassa , Jarred C. Heigelc , Yaakov Idella , Maureen E. Williamsa ,
5 Andrew J. Allend , Jonathan E. Guyera , Lyle E. Levinea
a
6 Materials Science and Engineering Division, Material Measurement Laboratory,
7 National Institute of Standards and Technology, 100 Bureau Drive, Gaithersburg, Maryland 20899, U. S. A.
b
8 Theiss Research, 7411 Eads Avenue, La Jolla, CA 92037
c
9 Intelligent Systems Division, Engineering Laboratory,
10 National Institute of Standards and Technology, 100 Bureau Drive, Gaithersburg, Maryland 20899, U. S. A.
d
11 Materials Measurement Science Division, Material Measurement Laboratory,
12 National Institute of Standards and Technology, 100 Bureau Drive, Gaithersburg, Maryland 20899, U. S. A.

13 Abstract

Numerical simulations are used in this work to investigate aspects of microstructure and microseg-

regation during rapid solidification of a Ni-based superalloy in a laser powder bed fusion additive

manufacturing process. Thermal modeling by finite element analysis simulates the laser melt pool,

with surface temperatures in agreement with in situ thermographic measurements on Inconel 625.

Geometric and thermal features of the simulated melt pools are extracted and used in subsequent

mesoscale simulations. Solidification in the melt pool is simulated on two length scales. For the

multicomponent alloy Inconel 625, microsegregation between dendrite arms is calculated using the

Scheil-Gulliver solidification model and DICTRA software.1 Phase-field simulations, using Ni–Nb

as a binary analogue to Inconel 625, produced microstructures with primary cellular/dendritic arm

spacings in agreement with measured spacings in experimentally observed microstructures and a

lesser extent of microsegregation than predicted by DICTRA simulations. The composition profiles

are used to compare thermodynamic driving forces for nucleation against experimentally observed

precipitates identified by electron and X-ray diffraction analyses. Our analysis lists the precipi-

∗
Corresponding author.
Email address: trevor.keller@nist.gov (Trevor Keller)
1
Certain commercial entities, equipment, or materials may be identified in this document in order to describe
an experimental procedure or concept adequately. Such identification is not intended to imply recommendation or
endorsement by the National Institute of Standards and Technology (NIST), nor is it intended to imply that the
entities, materials, or equipment are necessarily the best available for the purpose.

Preprint submitted to Elsevier May 2, 2017

 ACCEPTED MANUSCRIPT

tates that may form from FCC phase of enriched interdendritic compositions and compares these

against experimentally observed phases from 1 h heat treatments at two temperatures: stress relief

at 1143 K (870 ◦C) or homogenization at 1423 K (1150 ◦C).

14 Keywords: Additive manufacturing; Finite element analysis (FEA); CALPHAD; Phase-field

15 simulations; Microsegregation

16 1. Introduction

17 Inconel 625 (IN625) is a Ni-based superalloy used for turbine parts that is strengthened by

18 substitutional alloying elements such as Cr, Mo, and Nb. Laser powder bed fusion (L-PBF), an

19 additive manufacturing technique, presents opportunities to reduce the cost of making IN625 parts

20 with appropriate geometries and internal cooling channels for high-temperature applications. Heat

21 treatment is often necessary following additive manufacturing to relieve residual stress [1, 2], and to

22 homogenize the microstructure [3, 4]. Recent work shows that common heat treatments promote

23 precipitation of secondary phases [5], which degrade mechanical properties (such as indentation

24 hardness) in IN625 [6]. Heat treatment schedules for wrought IN625 were designed to avoid these

25 same precipitates [7]; however, there are substantial microstructural differences between wrought

26 and L-PBF material [8], with significant microsegregation of as-solidified material of particular

27 interest here. IN625 processed by welding, casting, or directional solidification exhibits less homo-

28 geneity than wrought material, and typically contains NbC and Laves phase precipitates [9, 10].

29 Finding suitable stress-relieving and homogenizing heat treatments without sacrificing strength is

30 an iterative process, but numerical modeling can help narrow the search.

31 3D finite element modeling has been crucial to understanding the L-PBF process. Models

32 describing single laser tracks across the powder bed surface are routinely used to simulate heat

33 dissipation through the solid substrate [11, 12]. Multi-track models allow for coupling residual

34 stress evolution to these thermal profiles [13–15]. Recent models improve the thermal modeling

35 using coupled Calculation of Phase Diagrams (CALPHAD) methods for accurate prediction of the

36 melt pool boundary and solidification microstructure [16]. Such advances in modeling the L-PBF

37 system configuration and relevant materials properties are critical to accurate finite element analysis

2
 ACCEPTED MANUSCRIPT

38 (FEA) predictions of the real process under conditions of interest.

39 Phase-field methods are commonly used to simulate microstructural features, between atomistic

40 and continuum length scales. A scalar-valued order parameter defined throughout a spatial domain

41 is used in these methods to label the presence or absence of some phase, hence “phase field.”

42 The best available phase-field models for solidification were performed in 2D for binary alloys

43 [17, 18]. The simplified geometry and approximation of multicomponent alloys as binaries allows

44 for efficient computation, and the models produce quantitatively correct mass redistribution across

45 the solidifying interface. The model has been applied successfully to tungsten arc welding of Al–Cu

46 [19], laser powder forming of Ti–Nb [20] and Ni–Nb [21], and electron beam additive manufacturing

47 of Ti–6 % Al–4 % V by combining Al with V as a virtual element with mass fraction 10 % [22].

48 A third approach to studying microsegregation during solidification uses DICTRA software [23],

49 which implements a one-dimensional model for diffusion-controlled phase transformation. Despite

50 its simplicity, this approach has the advantage of allowing for simulation of microsegregation and

51 back-diffusion in multicomponent materials by combining CALPHAD thermodynamic and kinetic

52 materials descriptions. DICTRA is routinely applied to multicomponent alloys, including Ni-based

53 superalloys [24].

54 These modeling techniques—FEA, phase-field, and CALPHAD-based—are already in use study-

55 ing various aspects of L-PBF, separately and increasingly in cooperation, for a variety of alloy

56 systems. It is our goal to integrate all three models together to achieve high fidelity simulations

57 of dendritic solidification in L-PBF IN625, with direct comparison against experimental results.

58 This effort will improve the fundamental understanding of solidification in this system and produce

59 input data for modeling solid state transformations in the future.

60 2. Numerical methods
61 2.1. Finite element thermal model

62 Using the commercial FEA code ABAQUS [25], a non-linear, transient, thermal model was

63 designed and executed to obtain the global temperature history generated during laser irradiation

64 of one layer of powder covering a solid substrate. The simulated powder layer thickness is 36 µm,

65 combining the nominal 20 µm layer height with 16 µm of underlayer densification. This matches the

3
 ACCEPTED MANUSCRIPT

Figure 1: 2D section through melt pool centerline in 3D FEA simulation of IN625 L-PBF using a single layer of
powder on a bulk substrate showing thermal profile into the substrate. Dividing line between red and orange contours
represents the solid-liquid interface, from which transformation angle α and solidification speed Vs = Vb cos α are
calculated as shown. Indicated temperatures are measured in K.

66 steady-state powder layer thickness observed in corresponding experimental builds. Both single-

67 track and multiple-track laser scans across the metal powder layer were modeled, in which one

68 “track” is a linear path to be followed by the laser separated by 100 µm from adjacent laser pathways

69 traversed in anti-parallel directions. The width of the molten pool in the transverse direction is

70 140 µm, so material at the midpoint melts on adjacent scans. To reduce computation time, the

71 elements that interact with the laser beam are finely meshed within the diameter of the laser, and

72 a coarse mesh was used for the surrounding loose powder and substrate, visible in Figure 1. Ma et

73 al. [26] described this model previously, with detailed discussion of appropriate parameter values.

74 Heat transfer in the L-PBF process was modeled using the energy balance equation with

75 Fourier’s law of heat conduction and internal sources of heat [27],

∂ (ρcp T )
= ∇ · (κ∇T ) + Q, (1)
∂t

76 in which the evolution of temperature T with time t depends on material density ρ, thermal

77 conductivity κ, specific heat capacity cp which depends on latent heat, and internal heat Q which

78 depends on radiative loss. The temperature-dependent bulk material density and specific heat

79 were calculated from a Scheil-Gulliver simulation for the nominal IN625 composition and using the

80 TCNI8 thermodynamic database [28] within the Thermo-Calc software [23]. The initial condition

81 assumed a uniform temperature of 353 K throughout the specimen at time t = 0. Adiabatic

82 conditions were applied to all boundaries except the top surface, on which the boundary condition

83 is

(−κ∇T ) · n̂ = qs + h (T − Te ) + εσ T 4 − Te4 .


(2)

4
 ACCEPTED MANUSCRIPT

84 The three terms on the right-hand side represent heat input from the laser, heat convection due to

85 flowing process gas, and radiation. Equation 2 depends on the surface normal n̂, laser input heat

86 qs , convective heat transfer coefficient h, thermal radiation coefficient ε, the Stefan-Boltzmann

87 constant σ, and ambient temperature Te . The laser input was modeled after the single-mode

88 continuous wave Yb fiber laser (λ = 1070 nm) used in our experimental L-PBF system (described

89 in Sec. 3). Interaction between the laser and material is modeled using a Gaussian expression for

90 surface heat flux [29],

−2r 2
 
2AP
qs = exp , (3)
πrb2 rb2
91 with power P = 195 W, powder bed absorption coefficient A = 0.50, laser beam radius rb = 50 µm,

92 and radial distance to the beam centerline r measured in µm. The simulated laser scanning speed

93 was 0.8 m s−1 .

94 Each element stores its temperature and a Boolean variable indicating whether it has ever

95 exceeded the liquidus temperature, Tℓ . Elements in the powder layer are initialized with this

96 variable set to “false,” indicating a powder state. Upon melting, the variable switches to “true,”

97 indicating a bulk state. Substrate elements are initialized with this variable set to “true.” There

98 is no mechanism for switching this melt-state variable from “true” to “false:” the fused material

99 can never revert to powder. The materials parameters ρ and κ for each element depend on both

100 variables, T and melt-state. In the powder state, κ depends on the packing fraction, particle size

101 distribution, particle morphology, and thermal conductivity of the bulk material and process gas

102 [30, 31]. In the simulations described in this work, κ was specified in the range from 1.0 W m−1 K−1

103 to 3.0 W m−1 K−1 after [32]. As T rises during the first melting event, ρ and κ linearly increase

104 from their powder to bulk values when T is above the solidus temperature Ts and below the

105 liquidus temperature Tℓ given by the IN625 phase diagram [23, 28]. If T exceeds Ts , then the

106 melt-state variable switches to “true;” ρ and κ are thereafter functions of T , only. Note that the

107 bulk state variable does not differentiate between solid and liquid phases: the FEA model uses T

108 to choose phase-dependent materials properties appropriately. Latent heat effects are captured in

109 the evaluation of cp [25].

5
 ACCEPTED MANUSCRIPT

110 2.2. CALPHAD-based solidification models

111 To estimate the extent of microsegregation during solidification of a material, the Scheil-Gulliver

112 model [33, 34] is often applied. The model can be used for multicomponent materials, provided

113 that a thermodynamic description for the multicomponent alloy is available. The model assumes

114 perfect mixing in the liquid and no diffusion in the solid phase. These conditions are not found

115 in nature, so the result is a theoretical limit: less segregation is expected during real solidification

116 processes than the Scheil-Gulliver model predicts, since finite diffusion in both the liquid and solid

117 phases will contribute to mass redistribution.

118 Microsegregation predictions may also be obtained using DICTRA software [23] that include

119 the effect of diffusion in liquid and solid during solidification. Local equilibrium is assumed at the

120 interface between phases, in this case liquid and γ, and flux-balance is maintained for each element.

121 Solutions to the diffusion equation and the assumption of local equilibrium at the phase interface

122 are used to determine the tie-line that satisfies flux-balance. Because the simulations are 1D, the

123 effects of dendrite tip diffusion are not included.

124 For the IN625 segregation simulation, we used the commercial thermodynamic database TCNI8

125 [28] and the NIST Ni Superalloy mobility database [35]. The DICTRA simulation domain was

126 150 nm, which is half of the secondary dendrite arm spacing measured from experimental mi-

127 crostructures (e.g., Figure 2). Temperature was specified as spatially uniform, but time-dependent,

128 with values taken directly from the FEA thermal model. To reduce the computational complex-

129 ity, metals contributing mass fractions below 0.5 % were excluded, producing a simplified system

130 composition Ni–0.1 % C–20.11 % Cr–0.72 % Fe–8.83 % Mo–3.75 % Nb.

131 2.3. Phase-field solidification model

We used a quantitative phase-field model in 2D to study the cellular/dendritic nature of the

solidification of the melt pool [18], with a simplified representation of IN625 as a Ni–Nb binary

system with only FCC γ phase and liquid phase. The bulk composition has a mass fraction of 4 %

Nb. The scalar phase parameter φ indicates whether a point in the 2D field is liquid (φ = −1),

solid (φ = 1), or within the solid-liquid interface (−1 < φ < 1). The phase-field is not conserved,

6
 ACCEPTED MANUSCRIPT

and evolves in time t and space (x, y) as


∂φ 1 2
h
2
i
= 2 W 0 ∇ a(n̂) ∇φ + φ − φ3 (4)
∂t τ0 a(n̂)
   
∂ ∂a(n̂) ∂φ ∂ ∂a(n̂) ∂φ
− a(n̂) + a(n̂)
∂x ∂ n̂ ∂y ∂y ∂ n̂ ∂x
 
λ 2 2 T − T0
− (1 − φ ) exp (u) − 1 + .
1−k mℓ c0 /k

132 The dimensionless interfacial energy a(n̂) = 1 + ǫ4 cos(4θ), with four-fold anisotropy of magnitude
 
∇φ
133 ǫ4 , interface normal vector n̂ = − |∇φ| , and orientation angle θ = arctan ∂φ ∂φ
∂y ∂x . The non-
/
 
2ck/c0
134 dimensional deviation of chemical potential, u = ln 1+k−(1−k)φ , is defined with respect to the

135 equilibrium chemical potential at a reference temperature T0 and system composition c0 . The frozen

136 temperature approximation is applied such that a linear temperature profile with constant gradient

137 G translates along the growth axis (x) with constant velocity Vs [36–40]: T (x, t) = T0 + G(x − Vs t).

138 Interface thickness W0 and relaxation time constant τ0 are related through the capillary length,

139 d0 = a1 W0 /λ. Asymptotic analysis, performed by enforcing local equilibrium at the interface as

140 its width vanishes, also links these quantities through a dimensionless coupling parameter λ and

141 diffusion constant in the liquid Dℓ [18, 41]: τ0 = a2 λW02 /Dℓ . The fitting parameters a1 = 0.8839 and

142 a2 = 0.6267 depend on the forms of the free energy functional and free energy density, respectively

143 [17]. W0 is therefore the only free parameter, chosen to be 10 nm.

Composition is modeled with a conserved field c, and evolves as


∂c 1−φ c0
=∇· Dℓ [1 + k − (1 − k)φ] ∇ exp (u) (5)
∂t 2 k

W0 c0 ∂φ
+ √ (1 − k) exp (u) n̂ .
2 2 k ∂t

144 This expression neglects the effect of thermal gradients on diffusion, or the Soret effect [42], which

145 contributes to macrosegregation under low solidification velocities [43], but not microsegregation

146 during rapid solidification. The phase diagram of the Ni–Nb system exhibits a practically linear

147 liquidus with constant slope mℓ = −10.5 K %−1 , measured with respect to mass percentage Nb,

148 and constant partition coefficient k = 0.48 in this dilute region. Equations 4 and 5 were solved on

149 a uniform rectilinear grid using a finite volume method and an explicit time marching scheme with

7
 ACCEPTED MANUSCRIPT

Table 1: Allowable and measured mass fractions of constituent elements for IN625 L-PBF feedstock used in this work
[47], as determined by ICP, FAAS (indicated by an asterisk, ∗ ), or XRF.
Element Standard Range Supplied (ICP) Supplied (XRF)
Ni 58.0 % minimum balance balance
Cr 20.0 % to 23.0 % 20.7 % 21.1 %
Mo 8.0 % to 10.0 % 8.83 % 6.69 %
Nb 3.15 % to 4.15 % 3.75 % 3.06 %
Fe 5.0 % maximum 0.72 % 0.01 % maximum
Ti 0.4 % maximum 0.35 % 0.01 % maximum
Al 0.4 % maximum 0.28 % 0.17 %
Co 1.0 % maximum 0.18 % 1.01 % maximum
Si 0.5 % maximum 0.13 % 0.13 %
Mn 0.5 % maximum 0.03 % 0.02 %
C∗ 0.1 % maximum 0.01 %
P∗ 0.015 % maximum 0.01 % maximum 0.01 % maximum
S∗ 0.015 % maximum 0.002 %

150 zero-flux boundary conditions. Model parameters for a dilute solution of Nb in Ni were used directly

151 from Nie et al. [21], summarized in their Table 1. Further analysis of solidification microstructures

152 in dilute Ni–Nb alloys using this model are reported elsewhere [44].

153 3. Experimental methods and results

154 Test cubes of IN625 were additively manufactured by the NIST Engineering Laboratory using

155 an EOSINT M270 (EOS GmbH, Krailling, Germany); for L-PBF system details, the interested

156 reader may refer to [45]. The EOS NickelAlloy IN625 powder (EOS GmbH, Krailling, Germany)

157 was supplied with compositions listed in Table 1, as measured by inductively coupled plasma

158 atomic emission spectroscopy (ICP) and flame atomic absorption spectroscopy (FAAS), or by X-ray

159 fluorescence spectroscopy (XRF, conforming to [46]). All measured values are within the standard

160 ranges for IN625 [47]. For calibration of the FEA thermal model, multiple-track laser scans were

161 made on 1.27 cm (0.5 in) thick solutionized IN625 plate (High Performance Alloys, Inc. Windfall,

162 IN, USA). For both IN625 media, the Yb fiber laser power was 195 W and scan speed was 0.8 m s−1 .

163 Scanning electron microscopy (SEM) specimens were polished to a 1 µm surface finish using

164 standard metallographic preparation techniques, then etched for 30 s in aqua regia.

165 Specimens for transmission electron microscopy (TEM) were prepared using electrical discharge

8
 ACCEPTED MANUSCRIPT

166 machining to cut disks 3 mm in diameter, which were then electropolished until electron-transparent

167 with a 6 % perchloric acid, 60 % methanol, and 34 % butanol solution. Figure 2 provides a rep-

168 resentative secondary electron micrograph from the as-built L-PBF IN625 specimen: the primary

169 dendrite arms are spaced 1 µm and the secondary arms are spaced 300 nm. Regions without sec-

170 ondary arms are also present: the wide range of localized solidification conditions produced by

171 L-PBF can produce regions of cellular growth, so we describe the experimental microstructures as

172 “cellular/dendritic.” Figure 3 provides representative bright field scanning transmission electron

173 micrographs (STEM) showing precipitates as light and dark circular spots, 50 nm to 200 nm in

174 diameter. Precipitates are found near cell/dendrite boundaries in as-built material, but are not

175 apparently localized to microstructural features after stress relief at 1143 K for 1 h. Electron diffrac-

176 tion patterns confirm the presence of MC, M6 C, and M23 C6 carbides in stress-relieved material,

177 whereas diffraction patterns of precipitates in as-built material could not be indexed due to residual

178 stress and high dislocation densities.

179 X-ray diffraction experiments were performed at the ultra-small angle X-ray scattering (USAXS)

180 facility at the Advanced Photon Source at Argonne National Laboratory [48, 49], with specimen

181 preparation and measurement conditions reported previously [5]. The relative distance and tilt

182 between the specimen and detector are calibrated using LaB6 powder. The sample was heated to

183 and held at 1143 K for 1 h, then cooled to 303 K, corresponding to the manufacturer’s recommended

184 stress-relieving heat treatment for IN625 [50]. The sample was then heated to and held at 1423 K

185 for 1 h, then cooled back to 303 K, a treatment suggested by previous work [5]. The heating rates

186 were 100 K min−1 , the cooling rates were 200 K min−1 , and the temperature stability was ±1 K.

187 From the room temperature XRD patterns, Figure 4, only the FCC cubic lattice (γ phase) with

188 a lattice parameter of (0.358 ± 0.001) nm could be identified. The first annealing step to 1143 K led

189 to the formation of carbides, most of which are M6 C with a cubic lattice of F d3̄m group and a lattice

190 parameter of (1.011 ± 0.003) nm. The second heating step to 1423 K preserved the carbides and

191 promoted the formation of an intermetallic Laves phase. The Laves phase has a hexagonal lattice

192 of P 63 /mmc group, with lattice parameters a = (0.481 ± 0.001) nm and c = (1.565 ± 0.004) nm.

193 The uncertainties in these values are reported with 95 % confidence. Hence, we conclude that from

9
 ACCEPTED MANUSCRIPT

Figure 2: Representative cross section through as-built L-PBF IN625 specimen, etched with aqua regia, showing
cellular/dendritic microstructure. Hitachi S-4700 secondary electron image with 20 kV accelerating potential, 11.9 mm
working distance. This image has been manipulated to increase its contrast.

10
 ACCEPTED MANUSCRIPT

Figure 3: Representative micrographs showing cellular/dendritic microstructure in L-PBF IN625 (a) as-built and
(b) following 1 h anneal at 1143 K. Circular spots between 50 nm to 200 nm in diameter represent precipitates,
with arrows indicating an exemplar in each image. Precipitates in annealed material were indexed as carbides, but
those in as-built material could not be uniquely identified. FEI Titan 80-300 bright field STEM images with 300 kV
accelerating potential and zone axes parallel to [1 1 0]. These images have been manipulated to increase contrast.

194 a statistical point of view, the dominant precipitates after this two-step heat treatment are M6 C

195 carbides and the Laves phase.

196 4. Simulation results

197 4.1. Finite element thermal simulations

198 To validate our FEA thermal model, we compared its surface temperature prediction against in

199 situ thermographic measurements. Details of the thermographic measurement setup were published

200 in [45], and data used for the single-track comparison given here are described in [51]. The same

201 thermal camera settings used in [45, 51] were used here: 40 µs integration time, and 1350 nm to

202 1600 nm spectral range.

203 As described in [45], thermographic imaging of laser scans on metal powder produces highly

204 stochastic temperature fields with localized gradients due to the varying surface structure and

205 emissivity, which inhibit true temperature measurement. In contrast, scans on flat plates of bulk

206 metal result in smooth temperature gradients, and single-line scans create steady-state melt pools

11
 ACCEPTED MANUSCRIPT

Figure 4: Synchrotron XRD observations of phase evolution during heat treatments of AM IN625 with scattering
vector magnitude q = 4π sin(θ)/λ expressed in terms of scattering angle θ and X-ray wavelength λ. Top panel:
experimental XRD patterns of the AM IN625 sample acquired at 303 K following the specified heat treatments.
Bottom panels: simulated powder XRD patterns for dominant phases (γ, M6 C, and Laves) based on the lattice
symmetry and parameters described in the text.

12
 ACCEPTED MANUSCRIPT

207 that simplify comparisons to FEA simulations. Therefore, FEA simulation results are compared

208 against experimental laser scans on bare plates, without powder. A representative temperature

209 profile on IN625 plate was taken from one frame of the thermal video after the melt pool attained

210 nominal steady state. Figure 5 compares the measured and predicted thermal profiles for a single

211 track scan without powder using a scan speed of 0.8 m s−1 and laser power 195 W.

212 The emissivity of the surface of a real specimen—also known as emittivity—is highly dependent

213 on processing, and values for the IN625 melt pool, rolled plate, and powder surfaces are unknown.

214 Therefore, the thermal camera signal cannot be directly converted to temperature for comparisons

215 to FEA model predictions. Instead, we scale thermographic data based on an observed solidification

216 zone following a technique developed for analysis of Ti–6 % Al–4 % V powder [52]. A shoulder in the

217 thermographic profile behind the melt pool is attributed to the solidification zone, apparent between

218 −0.86 mm to −0.74 mm in Figure 5. The profile is scaled by an assumed emittivity and converted

219 to temperature units such that the temperature in this zone matches the solidus temperature,

220 Ts = 1587 K, which we calculated for IN625 using the TCNI8 thermodynamic database [28]. This

221 technique uses Ts as a reference temperature, and results in emittivity of ǫ = 0.13, which is a

222 reasonable value for a semi-specular metallic surface at the same spectral range and viewing angle

223 (43.7◦ ) as the thermal imaging setup [53]. The calibrated range of the thermal camera was 823 K

224 to 1298 K. Temperatures above this range saturate the camera, and temperatures below approach

225 the noise floor of the camera detector. Applying this emittivity maps the reportable temperature

226 range to 988 K to 1763 K, as shown by the solid blue line in Figure 5.

227 Figure 6 shows the predicted temperature as a function of time at a point on the surface located

228 midway between the centers of two of three anti-parallel laser scan tracks. Full melting and re-

229 melting occurs as the laser beam traverses the two nearest tracks, passing the same distance away

230 from the measurement point both times. Heating without melting is observed as the laser scans

231 material along the third track. This three-track temperature profile was used directly in DICTRA

232 simulations of IN625 solidification.

233 Figure 1 shows the thermal profile of the melt pool in cross-section. The rectangular meshing

234 elements near the surface measure 10 µm × 10 µm × 6 µm. Contours illustrate the temperature

13
 ACCEPTED MANUSCRIPT

Figure 5: Results of FEA thermal modeling (red dashed line) and in situ thermographic measurement (solid and
dotted blue lines) for single-line scan on bare IN625 plate. Shoulder in measurement data corresponds to melt pool
boundary. Dotted blue lines indicate thermographic data outside the calibrated range but still shown for clarity.

14
 ACCEPTED MANUSCRIPT

Figure 6: Temperature on the surface of a single powder layer as a function of time at a position midway between
two melt pool centers from a three track scanning FEA thermal model. The surface material fully melts during two
near passes, with the molten pool overlapping this surface midpoint by 20 µm; on the third pass, 80 µm removed from
the melt pool boundary, this material is reheated without melting.

15
 ACCEPTED MANUSCRIPT

235 field, with the solid-liquid interface coinciding with the border between red and orange bands. The

236 transformation angle α between the melt pool boundary and laser scanning direction indicates the

237 local solidification speed, Vs = Vb cos α [19].

238 FEA thermal model results were also used to determine solidification parameters to be employed

239 for the phase-field simulations described in Section 4.3. Average cooling rate Ṫ = 105 K s−1 was

240 calculated from Figure 6, measuring the slope of the line connecting the maximum and minimum

241 temperatures of the second peak to its subsequent valley. A constant value for thermal gradient

242 G = 107 K m−1 was chosen for the simulations. It is representative of the values along the solidus

243 contour in Figure 1, Ts = 1587 K: we computed G = |∇T | in the range from 0.57 × 107 K m−1 to

244 2.2 × 107 K m−1 , for mesh points to the left of the melt pool minimum. Values taken along the

245 solidus contour behind the melt pool minimum yielded a range of angles α from 78◦ to 89◦ for

246 Vb = 0.8 m s−1 , and Vs ranged between 0.01 m s−1 and 0.17 m s−1 Note that the solidifying interface

247 experiences localized thermal gradients and solidification speeds: the cooling rate Ṫ = GVs in

248 L-PBF processes is not constant.

249 4.2. CALPHAD-based solidification simulations

250 Figure 7a represents solidification of the interdendritic region as predicted by the Scheil-Gulliver

251 model and by DICTRA for the thermal profile in Figure 5. Based on the low fraction of secondary

252 phases observed in as-solidified experimental specimens (e.g., Figures 3 and 4), we made the simpli-

253 fying assumption that liquid solidifies into γ phase, only. Due to the high cooling rate, the DICTRA

254 results show solidification behavior very similar to the Scheil-Gulliver model (Figure 7a). Figure 7b

255 shows the Scheil-Gulliver model prediction of liquid composition. Microsegregation occurs for all

256 elements, with particularly high segregation of Mo and Nb into the liquid. In the last solidified

257 liquid, T = 1250 K, the Mo and Nb mass fractions are as high as 20 % and 29 %, respectively. The

258 last solidified liquid is also enriched in C whereas Cr and Fe are depleted.

259 Detailed DICTRA simulation results are shown in Figure 8, with an inset showing results near

260 the centerline of interdendritic liquid. The composition profiles show microsegregation from the

261 secondary dendrite core (x = 0 nm) to the interdendritic region (x = 150 nm). Since applying the

262 FEA thermal profile (Figure 6) for the DICTRA simulation leads to complete melting at the second

16
 ACCEPTED MANUSCRIPT

Figure 7: Effects of rapid cooling on phase transformation, using the thermal history presented in Figure 6. (a)
Transformation of liquid into γ, expressed as molar phase fraction. Low extent of solid-state diffusion under these
conditions is evident in the close proximity of curves from the DICTRA simulation (black line) and Scheil-Gulliver
model (black). (b) C, Mo, and Nb enrichment and Cr depletion in γ at the growth front during solidification as
predicted by the Scheil-Gulliver model, demonstrating significant microsegregation. Phases other than liquid or γ
are excluded from this analysis.

263 scan temperature peak, the resulting segregation in Figure 8 occurs during this cooling. The third

264 laser scan, centered 150 µm from the FEA measurement point, leads to minor homogenization of the

265 segregated profiles. This is, however, only notable over a distance less than 0.5 nm (inset, Figure 8)

266 and can have no significant effect on the microsegregation profile. Figure 8 shows most of the

267 microsegregation occurs within the 10 nm region near the last solidified liquid, in good agreement

268 with a recent synchrotron SAXS study of the homogenization kinetics of L-PBF IN625 [5], where

269 a novel analysis directly linked the length of X-ray streaks to the length scale of the segregation.

270 It can further be concluded that the segregation is in the same range as in the Scheil-Gulliver

271 simulation. The main difference is that the DICTRA simulation results in a greater amount of Mo

272 and somewhat less Nb in the interdendritic region compared to the last solidified liquid composition

273 obtained from the Scheil-Gulliver model, as expected.

17
 ACCEPTED MANUSCRIPT

Figure 8: Composition profiles in γ as a function of distance from the secondary dendrite core, as predicted by DIC-
TRA simulation using the thermal history presented in Figure 6. Inset compares compositions after the second scan
(dashed lines), during which material remelts; and third scan (solid lines), which reheats without melting, producing
short-range diffusion (note x-axis scale). Nearby laser scans promote solid-state diffusion, but with negligible impact
on composition profiles due to the short residence time at elevated temperature.

18
 ACCEPTED MANUSCRIPT

274 4.3. Phase-field solidification simulations

275 Several phase-field simulations were performed with different values for Vs , but constant G, in

276 order to study microstructural evolution under various cooling conditions of interest. Simulations

277 were initialized with a planar solid-liquid interface with random perturbations up to one grid

278 spacing in location, and using a hyperbolic tangent profile along the growth axis to smooth the step

279 change over several grid spacings for numerical stability. The microstructures evolved according to

280 Equations 4 and 5, and simulations ran until steady state was achieved. Depending on solidification

281 conditions, interfacial instabilities can grow into dendrites (with secondary arms) or cells (without).

282 For the parameters chosen in this work, only cellular domains resulted, as represented in Figure 9

283 with Vs = 25 mm s−1 . In Figure 9, the steady-state mass fraction of Nb in the cell center line was

284 3.0 %. At the midpoint of the intercellular liquid, the mass fraction of Nb was 9.4 %. Intercellular

285 liquid is shown pinching off at the root of intercellular grooves with Nb mass fraction of 16 %.

286 This composition is below the eutectic composition of 22.5 % Nb. Thus secondary solid phases,

287 which are not available to these simulations, are not expected to form. The main point is that

288 the microsegregation found by this model is considerably less than is predicted by a Scheil-Gulliver

289 analysis of binary Ni–Nb: such an analysis for k = 0.48 predicts a mass fraction of 1.9 % Nb in the

290 cell centerline, 21 % in solid formed when only 1 % liquid remains, and likely formation of secondary

291 phases from the last liquid to solidify.

292 The primary dendrite arm spacing λ1 , a commonly reported metric for cellular/dendritic mi-

293 crostructures, was averaged from multiple simulations with the same Vs by dividing the sum of

294 simulation domain widths by the sum of the numbers of cells advancing at steady state. We mea-

295 sured λ1 in the range from 0.245 µm to 1.81 µm from our simulations, depending on the cooling rate

296 Ṫ which ranged from 104 K s−1 to 106 K s−1 , as shown in Figure 10. For reference, we compared

297 our results with the analytical models of Hunt [54],

λ1 = A(kΓ∆T0 Dℓ )0.25 G−0.5 Vs−0.25 (6)

19
 ACCEPTED MANUSCRIPT

Figure 9: Cellular microstructure in Ni–4 % Nb predicted by phase-field simulation after 1.3 µs of growth with
Vs = 25 mm s−1 , as illustrated by the scalar composition field c. Image represents 6 µm × 12 µm window near the
growth front. Note steady-state Nb enrichment of undercooled liquid near cell tips and pinch-off of liquid droplets at
the root of intercellular grooves. Nb diffuses out of these droplets into the surrounding solid, resulting in the linear
pattern of disks with fading intensity. Arrow indicates growth direction.

298 with A = 2.83, and Kurz and Fisher [55],

λ1 = A(Γ∆T0 Dℓ /k)0.25 G−0.5 Vs−0.25 (7)

299 with A = 4.3, under the simplifying assumption that undercooling ∆T ≈ ∆T0 = Tℓ − Ts , measured

300 from the equilibrium phase diagram at c0 . Γ = 3.65 × 10−7 K m is the Gibbs-Thomson coefficient.

301 The proportionality constants depend on the 3D geometry assumed for the dendrite arrays: Hunt

302 assumed spherical dendrite tips and derived A(kΓ∆T0 Dℓ )0.25 , while Kurz and Fisher assumed

303 ellipsoidal tips and derived A(Γ∆T0 Dℓ /k)0.25 . From the line of best fit through our simulation data

304 points, we calculate A = 6.8 from Equation 6, and A = 4.7 from Equation 7. Figure 10 shows

305 that neither model provides an objectively good fit. The deviation is likely due to the combination

306 of our reduced geometry (2D rather than 3D), rapid cooling (∆T 6= ∆T0 ), and more complex tip

307 geometry than either simple model. This result is supported by a body of experimental evidence

308 demonstrating large effects of c0 , G, and Vs on dendrite tip geometries and λ1 [56–58].

20
 ACCEPTED MANUSCRIPT

Cooling rate, GVs (K/s)

1e6 5e5 3e5 2e5 1e5 5e4
3
Hunt
Kurz and Fisher
2.5 Simulation

2
λ1 (µm)

1.5

0.5

0
0.0006 0.00075 0.0009 0.00105
-0.5 -0.25
G Vs

Figure 10: Comparison of primary dendrite arm spacing measured from simulated microstructures with two analytical
models for simple cellular/dendritic tip geometries: spherical (Hunt [54]) and ellipsoidal (Kurz and Fisher [55]).

21
 ACCEPTED MANUSCRIPT

309 5. Discussion

310 Experimental observations via TEM reveal significant precipitation near cell/dendrite bound-

311 aries in as-built L-PBF material (Figure 3). Stress relief at the manufacturer’s recommended

312 temperature (1143 K) was found to promote further growth of carbides, with M6 C precipitation

313 and growth from the γ matrix observed by in situ synchrotron XRD (Figure 4). Further treatment

314 at 1423 K promoted M6 C growth and Laves phase precipitation: after 1 h at temperature, M6 C

315 persisted and Laves phase grew to measurable levels. This growth is despite the fact that homo-

316 geneous IN625 of nominal composition is single-phase γ, as established experimentally and from

317 thermodynamics [7, 28].

318 Investigation into the cause of the observed precipitation during heat treatments requires knowl-

319 edge of the solidification rate, which can not be directly determined from the laser scan speed. In

320 situ thermographic measurements on bare IN625 plate provided the melt pool width and thermal

321 gradients on the surface. FEA simulations were calibrated against these surface data, then used

322 to extract details of the full 3D melt pool shape during three-track L-PBF simulations (Figure 1

323 and Section 4.1). For a Yb fiber laser operating at 195 W and scanning a single powder layer at

324 0.8 m s−1 , the solidification rate is between 0.01 m s−1 at the bottom of the melt pool and 0.17 m s−1

325 at its trailing edge. The cooling rate along this interface was on the order of 105 K s−1 , a rapid

326 solidification condition.

327 Experimentally, these conditions produce fine cellular dendrites. To assess the level of mi-

328 crosegregation, we performed modeling on three levels: Scheil-Gulliver, with no diffusion in γ and

329 perfectly mixed liquid; DICTRA, with multicomponent diffusion in 1D for both γ and liquid; and

330 phase-field, with 2D cells/dendrites of γ growing into liquid.

331 To assess how microsegregation correlates to the experimentally identified precipitates, we look

332 at the thermodynamic contribution to the driving force for nucleation of various phases from the

333 γ matrix [59], ignoring interfacial energy contributions and kinetic obstacles. The phase with

334 largest driving force would be expected to nucleate first. Comparison of these driving force values

335 may provide additional insight regarding the nucleation of precipitates in dendritically segregated

336 material. In Table 2, the phases most likely to nucleate are listed in decreasing order of driving force

22
 ACCEPTED MANUSCRIPT

Table 2: Thermodynamic driving force for nucleation of secondary phases from γ for the enriched (interdendritic) com-
position, Ni–0.13 % C–13.6 % Cr–0.35 % Fe–13.9 % Mo–23.5 % Nb, at the stress relief and homogenization treatment
temperatures. Greater values indicate larger driving forces for nucleation. Equilibrium phases for each temperature
are underlined.
1143 K 1423 K
Phase −∆Gnuc Phase −∆Gnuc
MC 20.5 kJ mol−1 MC 17.0 kJ mol−1
M2 C 15.6 kJ mol−1 M2 C 13.1 kJ mol−1
µ 8.0 kJ mol−1 M6 C 5.8 kJ mol−1
M6 C 7.9 kJ mol−1 BCC 3.3 kJ mol−1
BCC 6.3 kJ mol−1 µ 3.0 kJ mol−1
σ 5.2 kJ mol−1 Laves 2.9 kJ mol−1
Laves 4.1 kJ mol−1 liquid 1.8 kJ mol−1
δ 3.5 kJ mol−1 σ 1.2 kJ mol−1
γ ′′ 3.5 kJ mol−1 γ ′′ 1.2 kJ mol−1
M23 C6 3.4 kJ mol−1 δ 1.1 kJ mol−1

337 for the composition of solid when 1 % liquid remains (mass fraction Ni–0.13 % C–13.6 % Cr–0.35 %

338 Fe–13.9 % Mo–23.5 % Nb). On this basis, we expect Nb-rich MC carbide to nucleate first; if it

339 cannot, we expect M2 C. If M2 C cannot form, a Mo-rich phase is expected, M6 C or BCC, followed

340 by Cr-rich σ, P , Laves, or M23 C6 . As Table 2 demonstrates, the specific nucleation order depends on

341 temperature. Note that this is a sequence of energetic favorability, not a predicted time-evolution:

342 phases toward the bottom of the table can only precipitate if local conditions preclude any of the

343 more-favorable phases from doing so. The favored precipitate is also not the same everywhere:

344 local variations in composition and other materials properties will change the tabulated sequence.

345 Competitive growth and coarsening may occur when time and diffusion are factored in, but these

346 effects are beyond the scope of this paper.

347 The nucleation sequence suggested by the driving force calculations can be compared to the

348 calculated equilibrium phases for these compositions at the stress relieving and homogenizing tem-

349 peratures. Thermodynamic calculations show MC, BCC, δ and σ phases to be stable at both tem-

350 peratures, while µ phase is only stable at the lower temperature. Therefore, as the highly segregated

351 profiles diffuse during long heat treatments, solid state transformations are to be expected—an ac-

352 tive topic of research beyond the scope of this paper.

353 It is important to note that the nucleation sequence and the ability for a phase to form are

23
 ACCEPTED MANUSCRIPT

354 dependent on several mechanisms and not only the nucleation driving force. A complete description

355 of nucleation must include the interfacial energy between the matrix and secondary phase, which

356 depends on the interfacial area and crystallographic details. For large or complex unit cells in

357 particular, kinetic effects play an important role: a thermodynamically stable phase may be un-

358 able to nucleate or grow if its constituent elements are not sufficiently mobile in the matrix at the

359 specified temperature and composition. This may explain, for example, the fact that BCC phase

360 has not been observed in our experiments despite its apparently high driving force for nucleation.

361 Furthermore, the thermodynamic database used for the driving force calculation may contain un-

362 certainties and is also limited to a certain energy resolution. This could, in particular, be the case

363 when different phases have similar driving forces, e.g. M6 C and BCC in our case.

364 Even if M6 C and Laves are not equilibrium phases at these conditions, their driving forces

365 for nucleation are high enough to put them early in the nucleation sequence: earlier than the

366 equilibrium phases δ and σ. This could explain the experimental observation of these phases. It

367 is also interesting to note that the Nb-rich MC carbide is an equilibrium phase at temperatures as

368 high as 1423 K, which could complicate the homogenization treatment.

369 The binary phase-field simulations clearly show microsegregation in the cellular growth pattern.

370 The spherical droplets in Figure 9 form at high solidification speeds as the advancing cells pinch

371 off liquid pockets to maintain the steady-state intercellular groove depth. These liquid droplets

372 are enriched to more than twice the initial Nb composition, depending on cooling rate, with some

373 loss to diffusion during solidification. Experiments involving rapid solidification of electron-beam

374 melted Al–Fe–Ni alloys produced morphologically similar droplets, which precipitated intermetallic

375 compounds during solidification [60]. While our models did not consider secondary solid phases,

376 and therefore could not simulate precipitation in the evolving microstructure, the microsegrega-

377 tion patterns are real and significant. It is interesting to note that even the highly concentrated

378 droplets are enriched less than half as much as the microsegregation predicted by the DICTRA

379 simulations. To some extent, this difference reflects model geometry: the phase-field simulations

380 were performed in 2D with realistically curved dendrite tips, while DICTRA simulations repre-

381 sent 1D planar solidification. Curvature effects significantly affect diffusion. In addition, under

24
 ACCEPTED MANUSCRIPT

382 rapid solidification conditions, local chemical equilibrium across the solid-liquid interface is not

383 expected to hold, and results in the natural phenomenon known as solute trapping [61]. DICTRA

384 enforces local equilibrium without accounting for solute trapping, and therefore over-estimates the

385 extent of microsegregation in these simulations. While the phase-field model does not enforce local

386 equilibrium, the diffuse interface artificially magnifies the effect. Our phase-field model includes

387 an anti-trapping flux in Equation 5, to correct the spurious contribution [17]. The correction is
W0
388 accurate to second-order in d0 , the dimensionless ratio of interface thickness to capillary length,

389 and is exact for low solidification velocities. Under the rapid solidification conditions investigated

390 here, this second-order expression does not cancel the spurious effect, which artificially decreases

391 the simulated microsegregation. Under L-PBF conditions, which produce both curved interfaces

392 and solute trapping in real material, the phase-field model produces a more accurate estimate of

393 microsegregation than DICTRA for the the cellular/dendritic microstructure.

394 There are also confounding factors that affect the exact compositions predicted by our phase-

395 field and DICTRA models. The real temperature-time cycle during L-PBF is more complex than

396 the FEA model employed in this study: for example, the effects of building additional layers

397 on the temperature profile are neglected here. This simplified thermal history may discount the

398 influence of solid state diffusion, producing artificially high compositions in enriched regions of the

399 microstructure. This work also models the laser input as a surface heat source, only, which neglects

400 the effects of particle ejection on heat and mass transfer in the melt pool [62, 63]. Furthermore,

401 thermophysical properties such as k and mℓ depend on the solidification speed, and may deviate

402 significantly from equilibrium values in the rapid solidification regime, a topic explored in greater

403 detail elsewhere [44]. Finally, powder grains of somewhat different sizes and composition lead

404 to compositional variations throughout the part, which may produce highly localized segregation

405 patterns not captured in our models.

406 6. Conclusion

407 From this work, we conclude:

408 • For additively manufactured Inconel 625, heat treatments of 1 h at 1143 K and 1423 K promote

25
 ACCEPTED MANUSCRIPT

409 precipitation of secondary phases from the γ matrix.

410 • FEA simulations of the moving 3D melt pool shape predict maximum solidification rates

411 only 0.2 times the laser scan speed, accompanied by a cooling rate of 105 K s−1 . This L-PBF

412 process occurs under rapid solidification conditions.

413 • The Scheil-Gulliver model was used to compute the maximum possible extent of microseg-

414 regation for the multicomponent alloy IN625. DICTRA simulations that include the role of

415 diffusion in γ reduce the degree of microsegregation near the end of solidification. Phase-field

416 simulations of cellular/dendritic microstructures further reduce microsegregation, finding Nb

417 compositions 0.4 times the Scheil-Gulliver prediction.

418 • Carbides (MC, M2 C, and M23 C6 ), topologically close-packed (δ, µ, σ), and Laves phases

419 have negative thermodynamic driving forces for nucleation from the FCC phase under the

420 conditions of microsegregation and temperature studied: these may spontaneously precipitate

421 under the conditions of temperature and microsegregation investigated.

422 This effort demonstrates the viability of cooperatively modeling L-PBF processes using several

423 techniques to assess microstructural phenomena in IN625. Work in progress will extend this tool

424 chain to investigate the effects of interfacial energy and diffusion in microsegregated regions on

425 competitive precipitation and solid state transformations. With a complete microstructure model,

426 we plan to evaluate the whole process in order to find a better way to prepare this material for

427 service.

428 Acknowledgment

429 Use of the Advanced Photon Source, an Office of Science User Facility operated for the U.S.

430 Department of Energy (DOE) Office of Science by Argonne National Laboratory, was supported

431 by the U.S. DOE under Contract No. DE-AC02-06CH11357.

432 [1] E. R. Denlinger, J. C. Heigel, P. Michaleris, T. Palmer, Effect of inter-layer dwell time on

433 distortion and residual stress in additive manufacturing of titanium and nickel alloys, J. Mater.

434 Process. Technol. 215 (2015) 123–131.

26
 ACCEPTED MANUSCRIPT

435 [2] A. S. Tremsin, Y. Gao, L. C. Dial, F. Grazzi, T. Shinohara, Investigation of microstructure

436 in additive manufactured Inconel 625 by spatially resolved neutron transmission spectroscopy,

437 Sci. Technol. Adv. Mater. 17 (1) (2016) 324–336.

438 [3] F. Xu, Y. Lv, Y. Liu, B. Xu, P. He, Effect of heat treatment on microstructure and mechanical

439 properties of Inconel 625 alloy fabricated by pulsed plasma arc deposition, Phys. Procedia 50

440 (2013) 48–54.

441 [4] A. Mostafaei, Y. Behnamian, Y. L. Krimer, E. L. Stevens, J. L. Luo, M. Chmielus, Brief data

442 overview of differently heat treated binder jet printed samples made from argon atomized Alloy

443 625 powder, Data Brief 9 (2016) 556–562.

444 [5] F. Zhang, L. E. Levine, A. J. Allen, C. E. Campbell, E. A. Lass, S. Cheruvathur, M. R. Stoudt,

445 M. E. Williams, Y. Idell, Homogenization kinetics of a nickel-based superalloy produced by

446 powder bed fusion laser sintering, Scripta Mater. 131 (2017) 98–102.

447 [6] L. M. Suave, J. Cormier, P. Villechaise, A. Soula, Z. Hervier, D. Bertheau, J. Laigo, Microstruc-

448 tural evolutions during thermal aging of Alloy 625: Impact of temperature and forming process,

449 Metall. Mater. Trans. A 45 (7) (2014) 2963–2982.

450 [7] S. Floreen, G. E. Fuchs, W. J. Yang, The metallurgy of Alloy 625, in: E. A. Loria (Ed.),

451 Superalloys 718, 625, 706 Var. Deriv., The Minerals, Metals & Materials Society, Warrendale,

452 PA, 1994, pp. 13–37.

453 [8] Y. Idell, L. E. Levine, A. J. Allen, F. Zhang, C. E. Campbell, G. B. Olson, J. Gong, D. R.

454 Snyder, H. Z. Deutchman, Unexpected δ-phase formation in additive-manufactured Ni-based

455 superalloy, JOM 68 (3) (2016) 950–959.

456 [9] J. N. DuPont, Solidification of an alloy 625 weld overlay, Metall. Trans. A 27 (11) (1996)

457 3612–3620.

458 [10] A. Formenti, A. Eliasson, A. Mitchell, H. Fredriksson, Solidification sequence and carbide

27
 ACCEPTED MANUSCRIPT

459 precipitation in ni-base superalloys n718, In625 and In939, High Temp. Mater. Proc. 24 (4)

460 (2005) 239–258.

461 [11] B. Song, S. Dong, H. Liao, C. Coddet, Process parameter selection for selective laser melting

462 of Ti6Al4V based on temperature distribution simulation and experimental sintering, Int. J.

463 Adv. Manuf. Tech. 61 (9) (2012) 967–974.

464 [12] J. Yin, H. Zhu, L. Ke, W. Lei, C. Dai, D. Zuo, Simulation of temperature distribution in single

465 metallic powder layer for laser micro-sintering, Comput. Mater. Sci. 53 (1) (2012) 333–339.

466 [13] C. Li, C. Wang, H. Zhan, T. Han, B. Han, W. Zhao, Three-dimensional finite element analysis

467 of temperatures and stresses in wide-band laser surface melting processing, Mater. Des. 31 (7)

468 (2010) 3366–3373.

469 [14] A. Hussein, L. Hao, C. Yan, R. Everson, Finite element simulation of the temperature and

470 stress fields in single layers built without-support in selective laser melting, Mater. Des. 52

471 (2013) 638–647.

472 [15] N. E. Hodge, R. M. Ferencz, J. M. Solberg, Implementation of a thermomechanical model for

473 the simulation of selective laser melting, Comput. Mech. 54 (1) (2014) 33–51.

474 [16] J. Smith, W. Xiong, J. Cao, W. K. Liu, Thermodynamically consistent microstructure predic-

475 tion of additively manufactured materials, Comput. Mech. 57 (3) (2016) 359–370.

476 [17] A. Karma, W.-J. Rappel, Quantitative phase-field modeling of dendritic growth in two and

477 three dimensions, Phys. Rev. E 57 (1998) 4323–4349.

478 [18] B. Echebarria, R. Folch, A. Karma, M. Plapp, Quantitative phase-field model of alloy solidi-

479 fication, Phys. Rev. E 70 (6) (2004) 061604.

480 [19] A. Farzadi, M. Do-Quang, S. Serajzadeh, A. H. Kokabi, G. Amberg, Phase-field simulation of

481 weld solidification microstructure in an Al–Cu alloy, Modell. Simul. Mater. Sci. Eng. 16 (6)

482 (2008) 065005.

28
 ACCEPTED MANUSCRIPT

483 [20] V. Fallah, M. Amoorezaei, N. Provatas, S. F. Corbin, A. Khajepour, Phase-field simulation

484 of solidification morphology in laser powder deposition of Ti-Nb alloys, Acta Mater. 60 (4)

485 (2012) 1633–1646.

486 [21] P. Nie, O. A. Ojo, Z. Li, Numerical modeling of microstructure evolution during laser additive

487 manufacturing of a nickel-based superalloy, Acta Mater. 77 (2014) 85–95.

488 [22] X. Gong, K. Chou, Phase-field modeling of microstructure evolution in electron beam additive

489 manufacturing, JOM 67 (5) (2015) 1176–1182.

490 [23] J.-O. Andersson, T. Helander, L. Höglund, P. Shi, B. Sundman, Thermo-Calc & DICTRA,

491 computational tools for materials science, Calphad 26 (2) (2002) 273–312.

492 [24] P. D. Jablonski, C. J. Cowen, Homogenizing a nickel-based superalloy: Thermodynamic and

493 kinetic simulation and experimental results, Metall. Mater. Trans. B 40 (2) (2009) 182–186.

494 [25] Dassault Systémes Simulia Corp., Abaqus User’s Manual, version 6.13, Providence, RI USA

495 (2013).

496 [26] L. Ma, J. Fong, B. Lane, S. Moylan, J. Filliben, A. Hecker, L. Levine, Using design of experi-

497 ments in finite element modeling to identify critical variables for laser powder bed fusion, in:

498 Solid Freeform Fabr. Symp. Proc., Austin, TX USA, 2015, pp. 219–228.

499 [27] R. B. Bird, W. E. Stewart, E. N. Lightfoot, Transport Phenomena, Wiley, 2007, Ch. The

500 Equations of Change for Nonisothermal Systems, pp. 333–336.

501 [28] Thermo-Calc Software, Ni-based superalloys database, version 8, Online (2015).

502 URL http://www.thermocalc.com/products-services/databases/thermodynamic/

503 [29] I. A. Roberts, C. J. Wang, R. Esterlein, M. Stanford, D. J. Mynors, A three-dimensional finite

504 element analysis of the temperature field during laser melting of metal powders in additive

505 layer manufacturing, Int. J. Mach. Tool Manu. 49 (12–13) (2009) 916–923.

506 [30] M. Rombouts, L. Froyen, A. V. Gusarov, E. H. Bentefour, C. Glorieux, Photopyroelectric

507 measurement of thermal conductivity of metallic powders, J. Appl. Phys. 97 (2).

29
 ACCEPTED MANUSCRIPT

508 [31] M. R. Alkahari, F. Tatsuaki, T. Ueda, A. Hosokawa, R. Tanaka, M. S. Abdul Aziz, Thermal

509 conductivity of metal powder and consolidated material fabricated via selective laser melting,

510 in: Emerging Technology in Precision Engineering XIV, Vol. 523 of Key Engineering Materials,

511 Trans Tech Publications, 2012, pp. 244–249.

512 [32] T. H. C. Childs, C. Hauser, M. Badrossamay, Selective laser sintering (melting) of stainless

513 and tool steel powders: Experiments and modelling, Proc. Inst. Mech. Eng., Part B 219 (4)

514 (2005) 339–357.

515 [33] G. H. Gulliver, The quantitative effect of rapid cooling upon the constitution of binary alloys,

516 J. Inst. Met. 9 (1909) 120–157.

517 [34] E. Scheil, Bemerkungen zur schichtkristallbildung, Z. Metallkd. 34 (3) (1942) 70–72.

518 [35] C. E. Campbell, W. J. Boettinger, U. R. Kattner, Development of a diffusion mobility database

519 for Ni-base superalloys, Acta Mater. 50 (4) (2002) 775–792.

520 [36] V. G. Smith, W. A. Tiller, J. W. Rutter, A mathematical analysis of solute redistribution

521 during solidification, Can. J. Phys. 33 (1955) 723–745.

522 [37] W. W. Mullins, R. F. Sekerka, Stability of a planar interface during solidification of a dilute

523 binary alloy, J. Appl. Phys. 35 (2) (1964) 444–451.

524 [38] J. S. Langer, Instabilities and pattern formation in crystal growth, Rev. Mod. Phys. 52 (1)

525 (1980) 1–28.

526 [39] G. J. Merchant, S. H. Davis, Morphological instability in rapid directional solidification, Acta

527 Metall. Mater. 38 (12) (1990) 2683–2693.

528 [40] D. A. Huntley, S. H. Davis, Thermal effects in rapid directional solidification: linear theory,

529 Acta Metall. Mater. 41 (7) (1993) 2025–2043.

530 [41] A. Karma, W.-J. Rappel, Phase-field method for computationally efficient modeling of solidi-

531 fication with arbitrary interface kinetics, Phys. Rev. E 53 (1996) R3017–R3020.

30
 ACCEPTED MANUSCRIPT

532 [42] S. de Groot, P. Mazur, Non-Equilibrium Thermodynamics, Dover, 1984, Ch. 11.7 Thermal

533 Diffusion (Soret Effect) and Dufour Effect, pp. 273–284.

534 [43] L. Zheng, D. L. Jr., H. Zhang, Role of thermotransport (soret effect) in macrosegregation

535 during eutectic/off-eutectic directional solidification, J. Cryst. Growth 191 (12) (1998) 243 –

536 251.

537 [44] S. Ghosh, L. Ma, N. Ofori-Opoku, J. E. Guyer, On the primary spacing and microsegregation

538 of cellular dendrites in laser deposited Ni–Nb alloys, Modell. Simul. Mater. Sci. Eng. (2017)

539 In preparation.

540 [45] B. Lane, S. Moylan, E. P. Whitenton, L. Ma, Thermographic measurements of the commercial

541 laser powder bed fusion process at NIST, Rapid Prototyping J. 22 (5) (2016) 778–787.

542 [46] ASTM, Standard guide for elemental analysis by wavelength dispersive X-ray fluorescence

543 spectrometry, Tech. Rep. E1621 (2013).

544 [47] ASTM, Standard specification for additive manufacturing nickel alloy (UNS N06625) with

545 powder bed fusion, Tech. Rep. F3056-14e1 (2014).

546 [48] J. Ilavsky, P. R. Jemian, A. J. Allen, F. Zhang, L. E. Levine, G. G. Long, Ultra-small-angle

547 X-ray scattering at the Advanced Photon Source, J. Appl. Crystallogr. 42 (3) (2009) 469–479.

548 [49] J. Ilavsky, F. Zhang, A. J. Allen, L. E. Levine, P. R. Jemian, G. G. Long, Ultra-small-angle

549 X-ray scattering instrument at the Advanced Photon Source: History, recent development,

550 and current status, Metall. Mater. Trans. A 44 (1) (2013) 68–76.

551 [50] EOS GmbH, NickelAlloy IN625 material data sheet, Online (2011).

552 URL http://www.eos.info/material-m

553 [51] J. C. Heigel, B. Lane, Measurement of the melt pool length during single scan tracks in a

554 commercial laser powder bed fusion process, in: Proc. MSEC, Los Angeles, CA, 2017, pp.

555 575–591.

31
 ACCEPTED MANUSCRIPT

556 [52] I. Yadroitsev, P. Krakhmalev, I. Yadroitsava, Selective laser melting of Ti6Al4V alloy for

557 biomedical applications: Temperature monitoring and microstructural evolution, J. Alloys

558 and Compd. 583 (2014) 404–409.

559 [53] Y. Touloukian, D. DeWitt, Thermal Radiative Properties: Metallic Elements and Alloys,

560 Plenum Publishing Corporation, New York, 1970.

561 [54] J. D. Hunt, Cellular and primary dendrite spacings, in: Proc. Int. Conf. Solidif. Cast. Met.,

562 The Metal Society, London, 1979, pp. 3–9.

563 [55] W. Kurz, D. J. Fisher, Dendrite growth at the limit of stability: Tip radius and spacing, Acta

564 Metall. 29 (1) (1981) 11–20.

565 [56] L. H. Ungar, R. A. Brown, Cellular interface morphologies in directional solidification. IV. The

566 formation of deep cells, Phys. Rev. B 31 (1985) 5931–5940.

567 [57] L. X. Liu, J. S. Kirkaldy, Thin film forced velocity cells and cellular dendrites–I. Experiments,

568 Acta Metall. Mater. 43 (8) (1995) 2891–2904.

569 [58] J. S. Kirkaldy, L. X. Liu, A. Kroupa, Thin film forced velocity cells and cellular dendrites–II.

570 Analysis of data, Acta Metall. Mater. 43 (8) (1995) 2905–2915.

571 [59] M. Hillert, Phase Equilibria, Phase Diagrams and Phase Transformations: Their Thermody-

572 namic Basis, Cambridge University Press, 2007.

573 [60] W. J. Boettinger, L. A. Bendersky, R. J. Schaefer, F. S. Biancaniello, On the formation of

574 dispersoids during rapid solidification of an Al–Fe–Ni alloy, Metal. Trans. A 19 (4) (1988)

575 1101–1107.

576 [61] J. C. Baker, J. W. Cahn, Solute trapping by rapid solidification, Acta Metall. 17 (5) (1969)

577 575 – 578.

578 [62] S. A. Khairallah, A. T. Anderson, A. Rubenchik, W. E. King, Laser powder-bed fusion additive

579 manufacturing: Physics of complex melt flow and formation mechanisms of pores, spatter, and

580 denudation zones, Acta Mater. 108 (2016) 36–45.

32
 ACCEPTED MANUSCRIPT

581 [63] C. Teng, H. Gong, A. Szabo, J. J. S. Dilip, K. Ashby, S. Zhang, N. Patil, D. Pal, B. Stucker,

582 Simulating melt pool shape and lack of fusion porosity for selective laser melting of cobalt

583 chromium components, J. Manuf. Sci. Eng. 139 (1) (2016) 011009–011020.

33