CHAPTER-4
Finite Element Model Selection
4.1 Introduction
Finite element simulation of the welding is computationally very expensive, partly
because of highly non-linear formulation and partly because of its transient solution. Due
to the non availability of sufficient computational power and data storage in the past,
most of the welding simulations were performed with certain level of simplifications
such as reducing a three dimensional problem to a two dimensional either using plain
stress/strain or axisymmetric formulations. These simplifications drastically reduce
computational time but some times make the results erroneous because of over
simplification. This is why several studies for example Karlsson et al. [1], Goldak et al.
[2], Runnemalm and Lin {3] and Lindgren [4] advocated the use of three dimensional
models for appropriate results
Goldak et al. [5] in 1999 analyzed the possibility of using real time computational weld
mechanics with reference to the rapidly growing computational power. Based on the
development pace in the past, he anticipated that computational weld mechanics will
soon be faster than the real world welds. Computational time for thermal stress analysis
depends on several factors such as part geometry, level of discritization, number of load
steps, number of substeps in each load step and nature of non-linearities involved etc. To
analyze complex real world problems using more detailed FE models, the demand for
increased computational power is continuously growing with time. However, the most
recent developments in the computer hardware/sofiware had made it possible to use three
dimensional FE models for welding simulation problems (at least for simple geometries
and single pass welds) with fair accuracy and affordable computational time.
The availability of computational power alone does not provide sufficient justification
for using three dimensional FE models in welding simulation. If a two dimensional
96model provides sufficient engineering judgment of residual stresses and deformations in
a welded structure then the use of three dimensional model will merely be a wastage of
time and resources. However, the decision of the use of three-dimensionat model is
strictly based on the scope of study and nature of target application. In the present
dissertation the target application is the pipe-flange joint. It may have two types of in-
service problems; joint sealing (concemed with gasket behavior which strongly depends
‘on the welding deformations) and joint strength (concerned with stress corrosion
cracking or brittle fracture etc. and has strong dependence on residual stress state in the
bulk).
In all the previous FE studies, the flange joint’s performance is analyzed without taking
into account the welding distortion and residual stresses [6-10]. This is highlighted as
one of the major causes of joint failure during the experimental studies by Abid et al.
[10,12}. Itis a well-known fact that circumferential welding produces residual stresses in
and around the weld bead resulting in a distorted final geometry. It is mainly attributed to
highly non-uniform temperature field around the weld pool which results in large plastic
strain. In addition, clastic bending stresses arc also found outside the weld zone.
‘The most of the simulation work for circumferential welding is performed for butt
welded pipes by using the assumption of rotational and lateral symmetry (symmetry
across the weld centerline) such as, Rybicki et al. [13-15]. However, experimental work
by Jonsson and Josefson [16] and some three-dimensional FE studies e.g. [17-19].
reported deviations from rotational symmetry, especially at the beginning and end of the
welding cycle for pipe-pipe welding with lateral symmetry. By using a full three-
dimensional model for multi-pass welding of pipes, Fricke et al. [20] concluded that
residual stresses are by no means axisymmetric. The FE model of the pipe—flange joint is
different from the pipe-pipe model, as it has asymmetric geometry across the weld
centerline. For the pipe-flange joint, Troive and Jonsson [21] and Troive et al. (22)
presented experimental and two-dimensional FE results for welding distortions. In the
present chapter, a comparative study of a two-dimensional axisymmetric and a three
dimensional FE analysis of welded pipe-flange joint is presented for residual stresses and
deformations to identify their relative merits or demerits. The main objective in this
chapter is to study the profile of residual stresses in the welded pipe flange joint and its
variation in the axial and circumferential directions. Similarly the estimation of true
extent of deformations in view of their impact on joint performance will be discussed.
o7Based on the results of this preliminary study the type of finite element modet is selected
for upcoming parametric studies.
4.2 Finite Element Models
In both two- and three-dimensional models the flange is considered to have no bolt holes
assuming that this geometrical simplification has no significant effect on distortion and
residual stress. In order to facilitate data mapping from thermal to structural analysis, the
same FE model is used in both thermal and structural analyses, except for a change in the
clement type.
‘The axisymmetric two-dimensional FE model in Figure 4.1 represents a section of the
pipe-flange joint, This model consists of 3336 nodes and 3137 elements. The flange is
represented by 1873 elements and 1264 elements are for the pipe. At the weld centerline,
the clement size is 0.5 mm in the radial and 0.3 mm in the axial direction. The element
type for the thermal analysis is PLANESS which is a inear 4 node quadrilateral elements
with single degree of freedom ie. temperature on each node. The element type in
structural analysis is PLANE42 which is also a linear 4 noded element with two
translation degree of freedoms at its each node. Approximately same size of V-groove as
used in the three-dimensional model, discussed below, is achieved by deactivating the
elements in this region.
4.2.2. ‘Three-dimensional FE Model
The three-dimensional FE model (Figure 4.2 and Figure 4.3) contains a total of 25488
nodes with associated 21456 elements. Among these, 8496 elements are on the pipe side,
whereas, the remaining 12960 clements are on the flange side. Higher temperature and
flux gradients are expected in and around the fusion zone (FZ) and HAZ; therefore, a
relatively fine mesh is used within a distance of 10 mm on both sides of the weld
centerline. Away from the HAZ, the element size increases with increase in distance
from the centerline. Being the area of major concern, the description of the clement size
is limited to the FZ and HAZ only. In the axial direction, within the anticipated HAZ, the
element size varies from 0.6 to 1.25 mm, whereas in the circumferential direction, the
average length of the element is 2.4 mm
98