Professional Documents
Culture Documents
Metal Deposition
Metal Deposition
a r t i c l e i n f o abstract
Article history: Modelling and simulation of metal deposition (MD) poses several challenges to the modeller in addition
Received 20 August 2010 to the usual challenges in modelling of welding. The aim of the work presented in this paper is to enable
Received in revised form simulation of metal deposition for large three-dimensional components. Weld paths that are created in
5 May 2011
an off-line programming system (OLP) can be used directly to prescribe the movement of the heat
Accepted 9 May 2011
Available online 7 June 2011
source in the model. The addition of filler material is done by activation of elements. Special care must
be taken to the positioning of the elements, due to large deformations. Nodes are moved to ensure that
Keywords: the added material has correct volume and shape. A physically based material model is also
FEM implemented. This material model is able to describe the material behaviour over a large strain, strain
Simulation
rate and temperature range. Temperature measurements and deformation measurements are done in
Welding
order to validate the model. The computed thermal history is in very good agreement with
Metal deposition
Physically based material model measurements. The computed and measured deformations also show quite good agreement. It has
Validation been shown that the approach yields correct results, providing that flow stress and heat input models
are calibrated with sufficient accuracy. The method reduces the modelling work considerably for metal
deposition and multipass welding. It can be used for detailed models but also lumping of welds is
possible and often necessary for industrial applications.
& 2011 Elsevier B.V. All rights reserved.
1. Introduction 2. Background
The metal deposition (MD) process is quite similar to multi- 2.1. The metal deposition process
pass welding but with the aim to build up features or even
complete components with wanted geometry. This can both The advantage with the metal deposition process is that material
reduce weight and increase the flexibility of the manufacturing is added where it is actually needed instead of making a large
process. Simulations can assist in developing this manufacturing component and machining the wanted features. The structural
process. The current work is part of a project where the objective components in an aero-engine are good examples where deploy-
is to simulate the manufacturing chain of aerospace components. ment of the MD process simplifies the production provided
Two issues have been in focus. One is the development of a the quality can be assured. They are generally slender with more
material model that is valid over a large strain, strain rate and bulky features such as flanges and bosses needed to connect the
temperature range. A physical based material model has been components that could be produced by the MD process.
used for this and is described briefly in the paper. The two most common processes for manufacturing of struc-
The other issue, which is the main focus of the current paper, is tural components today is forging and casting. Forged compo-
enabling simulation of metal deposition. Simulation of multipass nents typically need to be largely oversized due to the complexity
welding and particularly metal deposition poses several challenges of the final shape. In many cases the subsequent machining
to the modeller in addition to the usual challenges in modelling of removes more than half of the volume of the component. Since
welding. There are numerical aspects such as, element activation nickel based super alloys and titanium are used, the actual
methods and heat sources. There are also practical aspects, how to material cost is therefore significant. Machining in these high
simplify the description of the weld process for the user. strength materials is also a time consuming and costly process.
An approach for supporting this in a convenient way is presented Casting produces a component that is much closer to the final
in this paper. The method is implemented in a commercial finite shape. However, the material integrity of castings is not as good
element code and applied to a validation case. as for forgings or sheet metal. There are also limitations in the
design of the component, transitions of thicknesses must not be
too sharp and the ratio between the thickest and thinnest section
n
Corresponding author. must not be too large. Metal deposition in combination with
E-mail address: andreas.lundback@ltu.se (A. Lundbäck). e.g. fabrication offers a product with high structural integrity,
0168-874X/$ - see front matter & 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.finel.2011.05.005
1170 A. Lundbäck, L.-E. Lindgren / Finite Elements in Analysis and Design 47 (2011) 1169–1177
[30]. This was a purely thermal simulation, but with the cap- Lindgren and Hedblom [34]. They modelled multipass welding
ability to predict the shape of the clad geometry. In the work by in 2D in an elegant way, especially regarding the addition of
Kim and Peng [31] they presented a 2D model with the aim to material. By minimising an objective function that contained the
predict the melt pool shape and dilution of laser cladding with smoothness of the external boundary, Gew , in Fig. 3, an element
wire feeding. Later it was extended to 3D in the work by Zhao distortion metric and the volume, Ow , of the added material. They
et al. [32]. Both of these papers only consider the thermal part of ensured that each layer of filler material had a smooth surface,
the problem. correct volume and a high quality mesh. However, the weld paths
The aim of the current developments has been to enable the are generally more complex in MD and the modelling is done in
simulation of each weld pass separately as well as lumping passes 3D in this work. Also, the bulk material is more flexible. Complex
together in order to reduce the computer requirements for large, weld paths make it necessary to have an automated procedure for
complex geometries. It can be combined with the quiet or the generation and treatment of the weld paths during modelling and
inactive element approach. It extends the work in [33], which simulation. The tool presented in this paper has the capability
implemented a logic for versatile description of the weld path to use weld paths from off-line programming software or CAD
leading to heat input in the finite element model. It also advances software. The method for generating and using the weld paths is
the approach in [34] to the three-dimensional context. presented in Lundbäck [33] and in some more detail in Lundbäck
[35]. The relatively low stiffness of the bulk material renders in
large deformations that need to be taken into consideration.
3. Modelling of metal deposition Fig. 4 shows an illustration of how the activation of elements
and the subsequent reposition of nodes are done. During the
The FE-software used in this work, MSC.Marc and its pre- and initiation of the simulation, time ¼0, all elements that belong to
post-processor Mentat, has an integrated connection to the the material that later are to be deposited on the base plate are
programming langue Python. In the model presented in this made inactive. In Fig. 4 the elements representing the base plate
paper we have 40 weld passes and cooling time between each are light grey and the elements that are to be activated, material
weld. The splitting of each weld pass into a separate load case was to be deposited, is dark grey. As seen in Fig. 4(a), the inactive
chosen in order to increase the flexibility. The user can easily elements are generated as simple square blocks. No consideration
change between the sequences of the weld deposits. However, to the wanted final shape or volume of the deposited material is
this means that we make at least 80 load cases in the model. Each taken at this point. In the next sequence, Fig. 4(b), the active
load case consists of a unique weld boundary condition that is elements have been distorted due to, e.g. thermal strains. The
coupled to a specific weld path. Then we also have the other nodes that solely belong to inactive elements, let’s call them free
boundary conditions such as fixation and thermal losses that nodes, are still at their original position. We now have a quite
should be active in all load cases. Building this model manually distorted mesh compared to the original one.
would be both time consuming and error prone. The implemen- Lindgren and Hedblom [34] used a modified simplex optimi-
ted logic is described below. Details of the element activation are sation method to minimise a function of the smoothness, element
described first followed by how the weld path is introduced and distortion and volume. A simpler approach is used in the current
heat input is made. work, as the number of elements that represent the filler material
in each layer is low. However, it is straightforward to apply the
3.1. Addition of filler material optimisation approach here also.
The first thing is to determine if an element is to be activated
Modelling of metal deposition is quite similar to modelling of or not during a given time step. The thermal and mechanical
multipass welding. The basic ideas in this work are taken from activation are separated. The thermal activation occurs before the
mechanical activation. This is done because we want the heat
source to heat up the element. But since the deposited material
consists of molten droplets from the feed wire in reality, it should
not have any mechanical strength yet.
The thermal activation is as follows. In Fig. 4(c) some of the
inactive elements have been activated thermally. This is done by
sweeping a user defined cross-section along the weld path. The
sweep distance equals the distance that the heat source travels
during the current time step plus a tolerance to avoid gaps. If the
midpoint of an element lies within this volume, then it will be
activated. The position of this volume along the weld path is
Fig. 3. Notation used for definition of weld geometry according to Lindgren and decided by the user as a distance between the centre of the
Hedblom [34]. volume and the centre of the heat source. Since this is done in the
Fig. 4. Illustration of MD-logic (a) initial mesh with active elements (light grey) and inactive elements (dark grey), (b) displacement of active elements, (c) activation of
elements and adjustment of nodes. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
1172 A. Lundbäck, L.-E. Lindgren / Finite Elements in Analysis and Design 47 (2011) 1169–1177
undeformed space, we do not need to take the current deforma- function is integrated over the length that the source moves
tions into account while searching for elements. The elements to during the time step in the welding direction. When using this
be activated are treated in groups of sections. In Fig. 4(c) we can model it is assumed that the heat source moves along a straight
see that one cross-section has already been processed, light grey, line during a time step. A thorough description of this heat source
and another section is about to be activated, white. If more than model can be found in [6].
one section of elements is to be thermally activated during the
current time step, the time step will be divided into sub-steps.
This sub-stepping procedure only applies to the processing of the 4. Flow stress model and its calibration
activation of elements and subsequent reposition of the nodes.
The next step is to decide how the position of the free nodes The flow stress is assumed to consist of two parts,
should be adjusted. The user defines a polyline, the thick red line sy ¼ sG þ sn ð2Þ
in Fig. 4(c), which gives the position of each free node. That is, the
free nodes will be moved to a corresponding point on the poly- where sG is the athermal stress contribution from the long-range
line. The height relative to the base material or previously laid interactions of the dislocation substructure. The second term s*,
welds and the shape of the polyline gives the wanted cross- is the friction stress needed to move dislocations through the
section area of the deposited material. The cross-section area of lattice and to pass short-range obstacles. Thermal vibrations can
the deposited material can be calculated on beforehand. It is given assist dislocations to overcome these obstacles.
by the diameter of the feed wire, fwire, the feed speed vwire and The long-range term from Eq. (2) is derived by Seeger [37] as,
the weld speed, vweld. pffiffiffiffiffi
sG ¼ maGb ri ð3Þ
Awire vwire pf2wire where m is the Taylor orientation factor translating the effect of
Across ¼ where Awire ¼ ð1Þ
vweld 4 the resolved shear stress in different slip systems into effective
This method requires that the pre-defined mesh for the stress and strain quantities. Furthermore, a is a proportionality
deposited material is structured, which means that the topology factor, b is Burger’s vector, ri is the immobile dislocation density
of the elements representing the filler material does not change. and G is the shear modulus.
The reposition of the nodes is not done by changing the coordi- The strain rate dependant part of the yield stress from Eq. (2)
nates of the nodes, instead they are moved by adding a displace- can be derived according to the Kocks–Mecking [38] formu-
ment to the nodes. In this way it is possible to maintain the shape lation as
of the elements in the undeformed space. This is important since 0 !!1=q 11=p
it simplifies the search algorithm considerably. @ kT e_ ref A
s ¼ sath 1
n
ln p ð4Þ
When activating an element we want it to be born without a DFb3 e_
history. That is, the stresses, elastic-, plastic- and thermal-strain
p
should be set to zero when an element is activated. The mechan- where e_ ref and e_ are the reference and plastic strain rate,
ical activation criterion is temperature based. If the mean tem- respectively, k is the Boltzmann constant and T is the temperature
perature in an element falls below a given value, then the element in Kelvin. Here, DF is the activation energy necessary to overcome
will be activated. The activation temperature is typically chosen lattice resistance and sath is the shear strength in the absence of
equal to or just below the melting point of the current material. thermal energy.
As discussed previously, the tool can use weld paths generated It is assumed that mobile dislocations move, on average, a
from an OLP system or CAD software. In this work, four paths in distance L (mean free path), before they are immobilised or
layer one is generated using the CAD software I-deas and annihilated. According to the Orowan equation, density of mobile
subsequent paths in the remaining layers are generated using a dislocations and their average velocity are proportional to the
Python script. An offset distance is prescribed and the start and plastic strain rate. It is reasonable to assume that increase in
stop position is indented, which is commonly done for practical immobile dislocation density also follow the same relation. This
reasons. Each weld pass is defined in a file with a unique name. leads to
The name of the file defines the position of the weld path,
m 1 _p
e.g. name_X_Y.txt. Where ‘‘name’’ is an arbitrary name for the r_ ði þ Þ ¼ e ð5Þ
bL
current job, X and Y defines the layer and the sequence, respectively.
where m is the Taylor orientation factor.
3.3. Heat input model Based on the formulation by Bergström [39], the immobile
dislocation density is proportional to the plastic strain rate.
The heat input model used in this work is the commonly used p
r_ i ðÞ ¼ Ori e_ ð6Þ
double ellipsoid with Gaussian distribution proposed by Goldak
et al. [36]. Due to the rather coarse mesh used for the discretisa- where O is a function dependant on temperature.
tion of the model, the integration of the heat input function can In addition to dislocations, vacancies are also created during
vary quite a lot. This is avoided by controlling of the heat input in plastic deformation. This has significant effect on diffusion con-
each iteration in every time step. The heat input is calculated in trolled processes such as climb and dynamic strain aging. Militzer
the first iteration and the efficiency factor, in the equation by [36], proposed the following model for recovery of dislocations based
is then scaled to get the correct net heat input in the succeeding on dislocation climb.
iterations of the current time step. The heat input model is by no
cv Gb3 2
mean constrained to the double ellipsoid. A segmented heat r_ i ðÞ ¼ 2cg Dl ðri r2eq Þ ð7Þ
cveq kT
source with Gaussian distribution has also been implemented.
This heat source allows longer time steps than with the conven- where, cv and cveq are equilibrium and current vacancy concentra-
tional source. The only difference with this heat source is that the tions, Dl is the self-diffusivity and cg, is the material coefficient.
A. Lundbäck, L.-E. Lindgren / Finite Elements in Analysis and Design 47 (2011) 1169–1177 1173
4.2. Evolution of vacancy concentration return method performs this task. The left superscript is the time
step counter. DT is increment in temperature and Dt is increment
Militzer et al. [40] proposed a model for evolution of excess in time. The lower box in Fig. 6 illustrates the role of the logic for
vacancy concentration with generation and annihilation compo- updating the flow stress. This is the additional logic for the
nents. Assuming that only long-range stress contributes to presented plasticity model. The plastic strain is an internal
vacancy formation and introducing mean free path it can be variable in most plasticity models. It is replaced in the current
written as model by two internal state variables, the immobile dislocation
density and the vacancy concentration. The vector q with these
maGb2 pffiffiffiffiffi c O0 _ 1 eq Qvf
c_ ex
n ¼ w Q ri þ B j2 eDvm 2 ðcv eq
v Þcv DT variables is introduced
vf 4b b L T2
ð8Þ qT ¼ ðri ,cv Þ ð9Þ
where Qvf is the activation energy for vacancy formation, w ¼ 0:1 The additional part to the stress-update algorithms is the
is the fraction of energy spent on vacancy generation, B is the solution of the non-linear coupled two equations for these
neutralization effect by vacancy emitting and absorbing jogs, cj is internal state variables. The yield stress and hardening modulus
the concentration of jogs, O0 is the atomic volume and Dvm is the can be calculated, upward arrow to stress–strain box, when q is
vacancy diffusivity. determined. The hardening modulus is defined as
Fig. 5 shows the calibrated model along with isothermal dsy ðqÞ
uniaxial compression tests performed at a strain rate of H0 ¼ ð10Þ
dep
0.001 s 1. The model used for simulations is calibrated with
strain rates up to 1 s 1. A more detailed derivation of the flow stress model and its
implementation can be found in Lindgren [41] and Babu [42].
Table 1
Process parameters.
Fig. 6. The computation of internal variables and flow stress in finite element 87 10.3 0.002 0.0125 0.00114
context [41].
1174 A. Lundbäck, L.-E. Lindgren / Finite Elements in Analysis and Design 47 (2011) 1169–1177
Table 2
Estimated parameters.
Fig. 10. FE-model used for validation, four weld sequences and 10 layers.
2000
Measurement Pyro
Measurement Top
Measurement Bot
1500 Simulation Pyro
Temperature [°C]
Simulation Top
Simulation Bot
1000
500
0
500 1000 1500 2000 2500
Time [s]
Fig. 11. Fixture and plate with deposited metal.
Fig. 9. Measured and computed temperature history, single wall.
2000 0
Measurement Experiment, P2
Simulation Simulation
−0.2
1500
Displacement [mm]
Temperature [°C]
−0.4
1000 −0.6
−0.8
500
−1
0
500 1000 1500 2000 2500 0 500 1000 1500 2000 2500
Time [s] Time [s]
Fig. 12. Measured and computed temperature history. Fig. 15. Out-of-plane displacement at point P2.
0
Experiment, P3
Simulation
−0.2
Displacement [mm]
−0.4
−0.6
−0.8
−1
0 Experiment, P1
Simulation the process the difference in the result is relatively constant. Looking
−0.2 at the individual weld passes one can see that they correlate very
well. At least up to layer six is laid, after that the model over predicts
Displacement [mm]
−0.4 the response in each weld pass. The deformation after finalisation of
each layer is still very low though.
−0.6
The behaviour of the deformations in point P2 and P3 is quite
−0.8
similar and is shown in Figs. 15 and 16. A comparison between the
measured and computed deformation after finalisation of each layer
−1 shows very good agreement. But here we can also see that the
model over predicts the response with respect to the deformations.
0 500 1000 1500 2000 2500
Time [s]
[28] L.-E. Lindgren, H. Runnemalm, M.O. Nasstrom, Simulation of multipass [36] J. Goldak, A. Chakravarti, M. Bibby, A. New, Finite Element Model for Welding
welding of a thick plate, International Journal of Numerical Methods and Heat Sources, Metallurgical Transactions B 15B (1984) 299–305.
Engineering 44 (1999) 1301–1316. [37] Alfred Seeger, The mechanism of glide and work hardening in FCC and HCP
[29] M. Mughal, H. Fawad, R. Mufti, Finite element prediction of thermal stresses metals, dislocations and mechanical properties of crystals, in: J.C. Fisher,
and deformations in layered manufacturing, Acta Mechanica 183 (2006) W.G. Johnston, R. Thomson, T. Vreeland Jr., (Eds.), John Wiely & Sons, Lake
61–79. Placid, USA, 1956, pp. 243–329.
[30] E. Toyserkani, A. Khajepour, S. Corbin, 3-D finite element modelling of laser [38] H. Mecking, U.F. Kocks, Kinetics of flow and strain-hardening, Acta Metal-
cladding by powder injection: effects of laser pulse shaping on the process, lurgica 29 (1981) 1865–1875.
Optics and Lasers Engineering 41 (2004) 849–867. [39] Y. Bergström, The plastic deformation of metals—a dislocation model and its
[31] J.-D. Kim, Y. Peng, Melt pool shape and dilution of laser cladding with wire
applicability, Reviews on Powder Metallurgy and Physical Ceramics (1983)
feeding, Journal of Materials Processing Technology 104 (2000) 284–293.
79–265.
[32] G. Zhao, C. Cho, J.-D. Kim, Application of 3-D finite element method using
[40] M. Militzer, W.P. Sun, J.J. Jonas, Modelling the effect of deformation-induced
Lagrangian formulation to dilution control in laser cladding process, Inter-
national Journal of Mechanical Sciences 45 (2003) 777–796. vacancies on segregation and precipitation, Acta Metallurgica et Materialia
[33] A. Lundbäck, CAD-support for heat input in FE-model, in: H. Cerjak (Ed.), 42 (1994).
Mathematical Modelling of Weld Phenomena, vol. 6, The Institute of [41] L.-E. Lindgren, K. Domkin, H. Sofia, Dislocations, vacancies and solute
Materials, Graz, Austria, 2002, pp. 491–518. diffusion in physical based plasticity model for AISI 316L, Mechanics of
[34] L.-E. Lindgren, E. Hedblom, Modelling of addition of filler material in large Materials 40 (2008) 907–919.
deformation analysis of multipass welding, Communications in Numerical [42] B. Babu, Physically based model for plasticity and creep of Ti–6Al–4V,
Methods in Engineering 17 (2001) 647–657. Department of Applied Physics and Mechanical Engineering, Luleå, Sweden:
[35] A. Lundbäck, Modelling of weld path for use in simulations, Department of Luleå University of Technology, 2008.
Mechanical Engineering, Luleå, Sweden: Luleå University of Technology, 2000. [43] GOM, Optical Measuring Techniques, /http://www.gom.comS.