You are on page 1of 14

International Journal of Pressure Vessels and Piping 76 (1999) 421–434

On the shakedown analysis of nozzles using elasto-plastic FEA


R. Preiss*
Institute for Pressure Vessel and Plant Technology, Vienna University of Technology, Gusshausstraße 30, A-1040 Vienna, Austria
Received 19 January 1999; accepted 4 February 1999

Abstract
The performed shakedown calculations of a dished end with a nozzle in the knuckle region with varying internal pressure load, and two
cylinder–cylinder intersections with constant moment load and varying internal pressure load are example cases for the application of the
check against progressive plastic deformation as stated in the new European UFPV standard, Annex 5.B: “Direct route for design by
analysis” (DBA). To calculate the shakedown limits, Melan’s (lower bound) shakedown theorem is used. In this context, the usage of the
deviatoric maps of stress states to obtain proper self-equilibrating stress fields is shown. Furthermore, some problems and corresponding
possible solutions for performing the shakedown check using a finite element model with shell elements are stated and shown in the
examples. q 1999 Elsevier Science Ltd. All rights reserved.
Keywords: Progressive plastic deformation; Nozzle; Cylinder–cylinder intersection; Elasto-plastic FEA; Design by analysis; Direct route

1. Introduction repeated application of an action cycle progressive plastic


deformation shall not occur for:
The performed shakedown calculations which are
• first order theory;
described in this article are example cases for the applica-
• a linear-elastic ideal-plastic material;
tion of the new European UFPV standard, Annex 5.B:
• Mises’ yield criterion and associated flow rule;
Direct route for design by analysis (DBA) [1], and some
• specified design strength parameters RM (determined
of them are part of the work of the European research
with the 0.2% proof strength for ferritic steels and 1%
project “Guidelines for the use of the results of elasto-plastic
proof strength for austenitic steels at the relevant
FEM” [2].
temperatures, respectively).
The direct route DBA section of the new standard deals
more clearly with the critical failure modes of pressure The corresponding application rules for action cycles
equipment than the stress categorisation approach, whereby with constant principal stress directions states that the
the following design checks are incorporated: global plastic diameter of the deviatoric mapping [4] of each path of the
deformation (GPD), progressive plastic deformation (PD), stress states, determined for every point of the stress-
instability (I), fatigue (F) and static equilibrium (SE). In the concentration-free structure using linear elasticity theory
following, the check against PD will be examined in detail. for the action cycle, shall not be greater than 2 RM. In
Like in the Eurocode for steel structures, the notions of cases where the stress path under investigation corresponds
principles and application rules [3] are used. Principles to a cyclic action superposed on a non-negligible permanent
comprise general statements, definitions and requirements action, the assessment shall also be made using the zero
for which there is no alternative, and requirements and stress point (0, 0, 0) of the deviatoric map.
analytical models for which no alternative is permitted In this context, the stress-concentration-free structure
(unless specifically stated). Application rules are generally is an idealised structure without local structural discon-
recognised rules which follow the principles and satisfy tinuities; a local structural discontinuity is a geometric
their requirements; alternatives are allowed provided it is or material discontinuity which affects the stress or
shown that they accord with the relevant principle. strain distribution through a fractional part of the wall
For the check against PD, the principle states that on thickness. The stress distribution associated with a local
discontinuity causes only very localised types of defor-
* Tel.: 1 43-1-58801-4060; fax: 1 43-1-504-1588. mation or strain and has no significant effect on the
E-mail address: sekretariet 1 e329@tuwien.ac.at (R. Preiss) gross structural deformations.
0308-0161/99/$ - see front matter q 1999 Elsevier Science Ltd. All rights reserved.
PII: S0308-016 1(99)00019-8
422 R. Preiss / International Journal of Pressure Vessels and Piping 76 (1999) 421–434

with the yield condition—the equivalent stress nowhere and


at no time exceeds the (yield) material parameter.
Using this theorem the principle is fulfilled, since PD and
Alternating Plasticity are the two possible inadaption modes
if a structure does not shake down under a cyclic load set.
A problem arises if the maximum allowable action given
by the shakedown limit is lower than the one resulting from
the check against GPD. In this case, detailed examination of
the structure’s behaviour under a cyclic loading with the
maximum allowable action according to the check against
GPD, i.e. determination of the structures inadaption mode,
would be necessary. Unfortunately, the possibility of such
examinations are restricted, on one hand because of the
hardware and time limits for performing cyclic calculations
and on the other hand due to lack of generally applicable
theorems. Determining generally applicable theorems in
this field is a present topic of research and, therefore, they
should be available in the future.
Performing linear-elastic FE calculations—as necessary
for a shakedown check—, stress singularities [7] can arise at
specific points of a weld modelled without a fillet. As it is
stated in the standard that the check against PD can be
performed for a stress-concentration-free structure, possibi-
lities of avoiding stress singularities can be:
• Modelling the weld with fillets corresponding to the weld
influence zone (see Fig. 1(a)).
• Modelling the weld with one fillet, which is completely
inside of the weld (see Fig. 1(b)).

2. Problems in performing a shakedown analysis using


shell elements

Using generalised stresses to verify that PD does not


Fig. 1. (a) Fillets according to the weld influence zone. (b) Fillet completely occur is often not an adequate possibility, because:
inside the weld.
• the validity of the corresponding theorem [8] seems to be
restricted to passive (unloading) processes [6,9–11];
Principally, this application rule is equal to the well-
• (local) interaction surfaces [12] are available for rota-
known criterion for the sum of the primary and secondary
tional-symmetric shells under rotational-symmetric load-
stresses in stress categorisation—the (often) so-called “3
ing only.
f—criterion” (where f stands for the allowable stress).
This criterion is an upper bound criterion for shakedown, Therefore, a shakedown check using stresses has to be
and therefore, the requirement given in this application is carried out frequently. But if shell elements are used, the
only a necessary and not a sufficient condition for the fulfil- only stress results available via the postprocessor are surface
ment of the principle [5,6]. The use of this application rule is stresses (top/bottom of the shell, i.e. outer/inner surface).
often the easiest way of applying the check against PD, but Using Melan’s shakedown theorem and corresponding
because of its uncertainty another possibility of fulfilling the self-equilibrating stress states to ensure that a structure
principle—Melan’s lower bound shakedown theorem—is shakes down under a given cyclic load path could be non-
employed. conservative, because the fulfilment of the theorem can only
Melan’s lower bound shakedown theorem states [5,6]: for be proven at the surface but not in the interior of the shell
a given cyclic action the structure will shake down if a time- model.
invariant self-equilibrating stress field can be found such In fact, when the absolute maximum of the self-equili-
that the sum of this stress field and the cyclically varying brating stress distribution in a cross-section is located in the
stress field determined with the (unbounded) linear-elastic interior, the above stated non-conservatism is possible.
constitutive law for the given cyclic action is compatible In the elasto-plastic simulation of the structures
R. Preiss / International Journal of Pressure Vessels and Piping 76 (1999) 421–434 423

Fig. 2. Geometry of dished end with nozzle.

behaviour during the loading or unloading half cycle, three Alternatively, if the FE-software confirms directly that no
phenomena can occur: further plastic strains occur during a half cycle (after some
initial load cycles)—requiring a suitable parameter in the
1. if the surface equivalent stress in at least one point of the
computation output—, it is proven that the structure has
structure remains, or becomes equal to the material
shaken down, independently of possible numerical errors
strength parameter, yielding occurs during the half
in the postprocessor.
cycle, yielding on the surface and/or in the interior;
A possibility to show that Melan’s theorem is fulfilled by
2. if the surface equivalent stress in all points of the struc-
performing the shakedown check with the shell model, i.e.
ture is less than the material strength parameter, no yield-
only at the surface but not at the interior of the structure, is
ing occurs on the surfaces during the half cycle, but
to use submodeling [13]. Submodeling is a finite element
yielding inside of cross-sections cannot be excluded;
technique to get more accurate results in a region, i.e. in the
3. only if the stress state on the surface after the half cycle is
case under consideration by performing an analysis of a
identical to the one resulting from purely elastic loading
coarse model (shell model), interpolating the results to the
or unloading of the structure, no stress redistribution due
(cut -) boundaries of the fine model (solid element model),
to plastic deformation occurs, neither at the beginning
which only represents the critical part of the structure, and
nor at the end of the half cycle. Therefore, it is assured,
compute the stress distribution in the fine model. If this
that the half cycle is purely elastic and the structure has
stress distribution confirms that the maximum stresses at
shaken down.
the critical parts of the structure are on the surface of the
There follows, to verify in a strict manner that a structure model, the shakedown check using the shell model fulfils
modelled by shell elements shakes down, one has to show the theorem.
that condition (3) is fulfilled.
If computation of the half cycle shows that condition (1)
is fulfilled, further cyclic calculations are necessary until 3. Exemplary shakedown analysis of nozzles
condition (2) is fulfilled—if actually possible. Afterwards,
to verify that the structure shakes down, one has to show that 3.1. Nozzle in dished end—single varying action (internal
condition (3) is fulfilled. However, this procedure could be pressure)
difficult if some considerable numerical errors, for example
due to extrapolations and averaging, are present in the stress In the following procedure, it is assumed that the only
plots. action acting at the structure under consideration is internal
Additionally, the practical relevance of the results in pressure, and that it varies between 0 and the maximum
points (nodes) on the intersection curve of two shells depend admissible pressure for shakedown PSmax SD.
on the kind of geometry of the structure, and, therefore care The problem in using Melan’s theorem is to find an opti-
must be taken by using these results. For example, if compu- mal self-equilibrating stress field. Often, the optimal, or a
tation of load cycles of a structure like a cylinder–cylinder near optimal, stress field can be found from the stress fields
intersection show that plastification occurs only in nodes of at the limit load:
the intersection curve, no conclusions about the structure’s The difference of the linear-elastic stress field at (or near)
behaviour can be made, since the considered nodes do not the limit pressure of the structure …sij †le;l and the elasto-
correspond to points of the real structure. plastic stress field at (or near) the limit pressure …sij †ep;l is
424 R. Preiss / International Journal of Pressure Vessels and Piping 76 (1999) 421–434

Fig. 3. Mises’ equivalent stress plots.


R. Preiss / International Journal of Pressure Vessels and Piping 76 (1999) 421–434 425

a self-equilibrating stress field …sij †res : displacements of the coarse (shell) model. At the coarse
model, symmetry boundary conditions were applied in the
…sij †res ˆ …sij †ep;l 2 …sij †le;l : …1†
nodes in the symmetry plane of the structure and in the
Since one endpoint of the considered linear-elastic load nodes in the vertical plane perpendicular to the symmetry
cycle is the point PS ˆ 0, the self-equilibrating stress field plane. Furthermore, the vertical displacement in the nodes at
used in Melan’s theorem must not violate the yield condi- the lower end of the structure was constrained to 0.
tion itself. Therefore, the self-equilibrating stress field In the elasto-plastic FE analysis Mises’ yield criterion and
…sij †res has to be scaled with a factor b such that it does associated flow rule, an linear-elastic ideal-plastic law with
not violate the yield condition: material strength parameters of 200 MPa for the shell, the
flat end and the nozzle, and first order theory was used.
…sij †res;co ˆ b…sij †res : …2†
Defining and using load cases in ansysw, the superposi-
If the uncorrected self-equilibrating field according to the tion of stress fields can be carried out easily; therefore the
limit load …sij †res does not violate the yield condition, the first load step of the analysis was defined at a very low load
shakedown load is not smaller than the limit load. level (0.1 MPa), so that there was linear-elastic response of
After this correction of the self-equilibrating stress field, the structure. All other linear-elastic stress fields can then be
the linear-elastic stress field with the possible greatest value found easily by multiplication with a suitable scale-up
of internal pressure has to be determined, such that the factor.
superposition with the corrected self-equilibrating stress The analysis was carried out using the Newton–Raphson
field does not violate the yield condition. The stress field method; a pressure close to the (unknown) limit pressure
at a lower bound shakedown limit …sij †SD is found as was found to be 1.13 MPa after a computation time of 3 h on
a compaqw Professional Workstation 5000 with two
…sij †SD ˆ …sij †res;co 1 a…sij †le;l : …3†
Pentium Prow processors and 256 MB RAM.
The scaling factors a and b can be determined easily Fig. 3(a) shows the elasto-plastic Mises’ equivalent stress
using the equivalent stress plots (or the listings). distribution for the pressure of 1.13 MPa.
The geometry of the considered dished end with a nozzle Fig. 3(b) shows the linear-elastic Mises’ equivalent stress
in the knuckle region is shown in Fig. 2. The material of the distribution for the pressure of 1.13 MPa—the stress maxi-
structure is X6CrNiMoTi 17-12-2 (1.4571) according to mum is located in the symmetry plane of the structure, on
prEN 10028-7, and the service temperature is specified the inside of the nozzle, slightly above the intersection
with 1808C. curve. The second region with large linear-elastic stresses
Since a flat end is used at the end of the nozzle, no warp- is a part of the intersection curve.
ing and ovalization of the nozzle occurs (linear-elastic Fig. 3(c) shows the Mises’ equivalent stress distribution
calculations showed that these phenomena occur due to of the self-equilibrating stress field, the stress maxima are
the internal pressure load if no flat end is used). located like in the linear-elastic solution. Fig. 3(d) shows the
Because of the symmetry of the structure and the locally Mises’ equivalent stress distribution of this self-equilibrat-
limited influence of the nozzle, only a quarter of the struc- ing stress field in the submodel—a check verified, that the
ture was modelled. maxima are on the surface of the structure. The values of the
Due to the large diameter-to-wall thickness ratio of the submodel differ by about 10% from the values of the shell
structure (a total number of 1187) 8-node shell elements model, which is an acceptable difference for the submodel
shell93 (in ansysw5.4) have been used for the FE interpolation procedures in an elasto-plastic calculation.
model. To model the whole structure with solid elements Therefore, the scaling factor of the self-equilibrating stress
using an appropriate number of elements in wall thickness field is calculated, using the shell model, to b ˆ 0.283.
direction would result in a FE model with too many The self-equilibrating stress state of the submodel was
elements and nodes, and, therefore, in an unacceptable generated by performing two calculations—an elastic–plas-
long computation time. Further, a submodel was used, to tic one and a corresponding linear-elastic one—, creating
ensure that the maxima of the used self-equilibrating stress two load cases, copying the linear-elastic load case in the
state were located on the surface of the structure. The working directory of the elastic–plastic load case and,
submodel consisted of 19526 10-node tetrahedral solid finally, superposing the load cases. The corresponding
elements solid87. To get proper stress results, the used computation times were 7 min for the linear-elastic and
elements had midside-nodes and a maximum global element 4 h and 5 min for the elastic–plastic calculation, respec-
size of 5 mm was used in the free meshing of the structure. tively, on the compaqw workstation.
The submodel was bounded by a cylinder with radius Fig. 3(e) shows the Mises’ equivalent stress distribution
186.5 mm and by a horizontal plane, which is located at the lower bound shakedown limit, the scaling factor a is
85 mm below the flat end. In the plane of symmetry a given by 0.446, and the shakedown limit pressure by
symmetry boundary condition was applied to all nodes,
PSmax SD ˆ 0:446 × 1:13 ˆ 0:504 MPa:
the boundary conditions of the cut-boundaries were
interpolated by the software from the corresponding Determining the maximum allowable pressure according
426 R. Preiss / International Journal of Pressure Vessels and Piping 76 (1999) 421–434

Melan’s theorem, the following conditions have to be met:


f‰b…sij †res;…M1p† 1 …sij †M;le 1 a…sij †p;le Š # 0; …5†

f‰b…sij †res;…M1p† 1 …sij †M;le Š # 0; …6†


where f the yield condition, …sij †M;le the linear-elastic stress
field corresponding to the moment M, …sij †p;le is the linear-
elastic stress field corresponding to an arbitrary value of
internal pressure p, and b and a are the factors which
have to be determined such that the conditions are fulfilled.
Thus, the maximum admissible internal pressure according
to shakedown is given by
PSmax SD ˆ ap: …7†
Defining and using load cases in ANSYSw, the super-
position of stress fields can be carried out easily. Unfortu-
Fig. 4. Geometry of thick-walled cylinder–cylinder intersection. nately, in these examples, often no combination of stress
states, i.e. no factors b and a could be found, such that
to PD with the corresponding application rule given in EN- the two conditions (5) and (6) are fulfilled. Often, the
UFPV Annex 5.B, leads to a limit pressure of 0.506 MPa— Mises’ equivalent stress of the combined load cases are
this value is an upper bound shakedown limit, which is—in too high either at the outer surface of the weld-fillet or at
the case under consideration—very close to the shakedown the inner edge of the nozzle shell intersection.
limit determined using Melan’s theorem. The limit pressure In this case, the conclusion is that the chosen equilibrating
according to GPD is larger than the limit pressure according stress field was not an appropriate one. Therefore, a linear
to the shakedown check and, unfortunately, no statements combination of self equilibrating stress fields—the one
about the inadaption mode of the structure at cycling with according to the limit state …sij †res;…M1p† and the one accord-
the limit pressure according to GPD seem to be applicable. ing to the moment load only …sij †res;M —is used to fulfil
If it could be shown, that the inadaption mode is alternating Melan’s theorem. This procedure is permissible because
plasticity and not progressive plastic deformation, cycling of the following attributes of self-equilibrating stress fields:
with this larger value of internal pressure would be admis-
• The sum of two self-equilibrating stress fields is a self-
sible.
equilibrating stress field.
• The multiple of a self-equilibrating stress field is a self-
3.2. Thick walled cylinder–cylinder intersection—single equilibrating stress field.
constant action (nozzle moment) and single varying action
(internal pressure) Using this procedure, the necessary conditions are given
by
In the following procedure, the structures under consid-
eration are cylinder–cylinder intersections, where a
f‰b1 …sij †res;…M1p† 1 b2 …sij †res;M 1 …sij †M;le 1 a…sij †p;le Š # 0;
constant moment load M is acting at the nozzle and the …8†
cyclic load is given by the internal pressure, which varies
between 0 and the maximum admissible pressure for shake- f‰b1 …sij †res;…M1p† 1 b2 …sij †res;M 1 …sij †M;le Š # 0; …9†
down PSmax SD. where the maximum admissible internal shakedown pres-
Again, the problem in using Melan’s theorem is the deter- sure is now given by
mination of an optimal self-equilibrating stress field.
The difference of the elasto-plastic stress field PSmax SD ˆ ap: …10†
…sij †ep;…M1p† , which corresponds to a loading state with the The self equilibrating stress field according to the
constant moment M and an internal pressure near the limit moment only, …sij †res;M , is given by the difference of the
state of the structure, and the linear-elastic stress field at this elasto-plastic stress field at this state …sij †ep;M and the
state …sij †le;…M1p† , the stress field linear-elastic stress field at this state …sij †le;M :
…sij †res;…M1p† ˆ …sij †ep;…M1p† 2 …sij †le;…M1p† …4† …sij †res;M ˆ …sij †ep;M 2 …sij †le;M : …11†
is a self-equilibrating stress field. Usually, for the determination of the factors b 1, b 2 and a
One endpoint of the considered linear-elastic load cycle is the deviatoric maps of the stress states, i.e. the coordinates
the point (M; p ˆ PSmax SD ), the other endpoint is given by of a stress point given by its principle stresses, at the critical
(M, p ˆ 0), where p ˆ PSmax SD is the maximum admissible locations of the structure are used, as a result of the
pressure for shakedown of the structure. Therefore, to fulfil increased number of factors and the different critical
R. Preiss / International Journal of Pressure Vessels and Piping 76 (1999) 421–434 427

locations load case operations using the FE software are not from the length of the shell and the nozzle. Therefore, to
feasable. save computation time, the total length of the model was
The geometry of the considered cylinder–cylinder inter- restricted to 200 mm. The length of the set-on nozzle in the
section is shown in Fig. 4. The specified materials are model was 100 mm (including the weld).
P265GH according to EN 10028-2 for the shell and To apply the moment load, a beam element (3D elastic
11CrMo9-10 according to prEN 10216-2 for the nozzle, beam BEAM4 in ansysw5.4) was attached to the structure
and the service temperature is specified with 508C. A at the centre of the nozzle’s upper end. To transfer the
constant moment load M ˆ 711.1 Nm is acting at the end moment load from the beam to the nozzle, the nodes in
of the nozzle, the moment vector is normal to the plane the upper cross-section of the nozzle (slave nodes) have
through both cylinder axis. been linked to the lower node of the beam (master node)
To model the structure, 3D 8-node brick elements— rigidly. The constraining equations of this “rigid region”
SOLID45 in ansysw5.4—have been used. A higher order associate the vertical displacements of the nodes in the
version of this element is the 20-node brick element solid95 upper cross-section of the nozzle with the rotation in the
with midside nodes. To show the differences using these two node of the beam caused by the moment load. As a conse-
types of elements, linear-elastic and elasto-plastic calcula- quence of this boundary condition the upper cross-section of
tions, leading to the following results, have been carried out: the nozzle remains plane, i.e. the cross-section can become
oval but cannot warp.
• the linear-elastic peak stresses differ by about 5–8% (for
Because of the symmetry of the structure and the load,
pressure as well as for moment load), the larger values
only half of the structure was modelled and, therefore, the
were obtained by the model using the 20-node elements;
applied moment load was half of the specified value. In the
• the elasto-plastic results near the limit load of the struc-
plane of symmetry a symmetry boundary condition was
ture are almost identical, i.e. nearly independent of the
applied to all the nodes.
kind of elements used;
As boundary condition in the end cross-sections of the
• the time for computing the results, using the compaqw
shell, the hoop displacement was set to 0. This affects the
workstation, applying an internal pressure load of
hoop strain and the change of curvature in hoop direction,
26 MPa at constant moment (which is very close to the
but seems to be the best boundary condition applicable, as
limit state of the structure) was 1 h 15 min using the 8-
warping and oval deformations are possible. Setting the
node elements, and 63 h 15 min using the 20-node
hoop displacements to 0 and creating rigid regions in the
elements.
end cross-sections of the shell leads to plane end cross-
Because of the major difference in computation time and section and avoids warping. As linear-elastic calculations
the fact that there is no need to compute linear-elastic peak showed that the maximum stresses using this boundary
stresses very exactly, because the check against PD can be condition are lower than the ones using the boundary condi-
carried out using the stress-concentration-free structure, the tion without rigid region, the latter one was used.
8-node elements were used for all the calculations. To get a stable model, the horizontal displacement in one
Because of the possibility of using a stress-concentration- node at one end cross-section of the shell has been
free structure in the check against PD, the weld was constrained to 0. The vertical displacement in the node at
modelled with a fillet (radius 6.5 mm) located entirely the upper (dummy) end of the elastic beam has also been
inside the weld. At the inner edge of the cylinder–cylinder constrained to 0.
intersection a fillet with 1 mm radius was used (modelling The used loading state for the shakedown analysis with
details of a structure for avoiding stress singularities or peak constant moment M ˆ 711.1/2 ˆ 355.5 Nm and internal
stresses influence linear-elastic calculations, but do not pressure of p ˆ 26 MPa is close to the limit state of the
affect elasto-plastic results near the limit load of the struc- structure, as the theoretical limit pressure of the undisturbed
ture). shell is 27.04 MPa for Mises’ yield condition and a material
The nozzle of the considered structure is of the set-on strength parameter of 234 MPa.
type. Therefore, using the material characteristics of the To obtain the required stress states and the elasto-plastic
shell for the welding zone too, leads to conservative results, stress state corresponding to the moment load only during
because the material of the nozzle is stronger than the mate- one analysis—which is advantageous to realise load case
rial of the shell. operations in ansysw5.4—the following load path was
Calculations with the moment load only showed that the used in the FE analysis:
region of the shell which is influenced by the moment is
rather small, and, therefore, FE models with different length state 1 p ˆ 0 MPa, M ˆ 35.5 Nm (linear-elastic path)
of the shell render almost identical results in the intersection state 2 p ˆ 2.6 MPa, M ˆ 35.5 Nm (linear-elastic path)
region. As the opening in the shell is small and the wall state 3 p ˆ 0 MPa, M ˆ 0 Nm (linear-elastic path)
thickness are considerable, the results obtained for internal state 4 p ˆ 0 MPa, M ˆ 355.5 Nm (elasto-plastic path)
pressure are independent from the kind of boundary condi- state 5 p ˆ 26 MPa, M ˆ 355.5 Nm (elasto-plastic
tion used at the end of the nozzle (rigid region or free) and path).
428 R. Preiss / International Journal of Pressure Vessels and Piping 76 (1999) 421–434

Fig. 5. Critical (high stressed) locations.

The required linear-elastic stress field corresponding to junction (path I in Fig. 5) with the adjacent part of the nozzle
the limit state is found by scaling the stress field of state 2, (path Z in Fig. 5). To make sure, that the critical part of the
the one corresponding to an internal pressure of p ˆ 26 MPa structure is not in a part of the undisturbed shell a corre-
by scaling the stress field of state 1, and the one correspond- sponding point at the inner surface was also taken into
ing to the moment load of M ˆ 35.5 MPa by scaling the account.
stress field of the difference of the stress fields of the states 2 As the self-equilibrating stress field according to the
and 1. moment load …sij †res;M is equal to 0 in the critical point at
The computation time was 3 h and 10 min on the the inner surface (paths I and Z), i.e. the point of the linear-
compaqw workstation. The elasto-plastic FE analysis was elastic stress distribution for moment and pressure load with
carried out using Mises’ yield condition and associated flow greatest distance to the centre in deviatoric map (see Fig.
rule, a linear-elastic ideal-plastic law with material strength 6(a)), the necessary conditions in this point are now given
parameters of 234 MPa for the shell and 343 MPa for the by
nozzle (the boundary between the two materials was
f‰b1 …sij †res;…M1p† 1 …sij †M;le 1 a…sij †p;le Š # 0; …12†
assumed to be at the upper end of the weld), and first
order theory.
In the case under consideration, no factors a and b could f‰b1 …sij †res;…M1p† 1 …sij †M;le Š # 0: …13†
be found, so that the necessary conditions (5) and (6) were Thus, b1 and a can be determined from these two condi-
fulfilled. tions:
Therefore, a linear combination of self equilibrating
stress fields—the one according to the limit state b1 ˆ 0:567; a ˆ 0:679:
…sij †res;…M1p† and the one according to moment load only
Fig. 6(a) shows the stress distribution in the deviatoric map
…sij †res;M —is used to fulfil Melan’s theorem.
for the inner surface:
For the determination of the factors b1 ; b2 and a the
deviatoric maps of the stress states are used. • linear-elastic stress according to moment and pressure
The two critical locations are the outer surface of the weld load …sij †le;…M1p† —thin green line;
fillet (path A in Fig. 5) with the adjacent part of the nozzle • elasto-plastic stress according to moment and pressure
(path N in Fig. 5), and the inner surface of the shell-nozzle load …sij †ep;…M1p† —thick green line;
R. Preiss / International Journal of Pressure Vessels and Piping 76 (1999) 421–434 429

Fig. 6. Deviatoric maps.


430 R. Preiss / International Journal of Pressure Vessels and Piping 76 (1999) 421–434

Fig. 7. Geometry of thin-walled cylinder–cylinder intersection.

• linear-elastic stress according to moment load only • self-equilibrating stress according to moment load only
…sij †le;M —orange line; …sij †res;M —thick red line.
• self-equilibrating stress according to moment and pres-
Thus, the limit pressure according to the shakedown
sure load …sij †res;…M1p† —violet line.
condition is not smaller than
For the critical point at the outer surface (paths A and N),
PSmax SD ˆ ap ˆ 0:679 × 26 MPa ˆ 17:65 MPa:
i.e. the point of the linear-elastic stress distribution for
moment and pressure load with greatest distance to the
centre in deviatoric map (see Fig. 6(b)), the necessary condi- 3.3. Thin walled cylinder–cylinder intersection—single
tions are given by constant action (nozzle moment) and single varying action
f‰0:569…sij †res;…M1p† 1 b2 …sij †res;M 1 …sij †M;le 1 0:679…sij †p;le Š (internal pressure)

# 0; The geometry of the considered cylinder–cylinder inter-


section is shown in Fig. 7. The specified material is P295GH
…14†
according to EN 10028-2 for the shell and the nozzle, and
the service temperature is specified to 508C. A constant
f‰0:569…sij †res;…M1p† 1 b2 …sij †res;M 1 …sij †M;le Š # 0: …15†
moment load M ˆ 15644.4 Nm is acting at the end of the
It can be shown, that the first condition is fulfilled for all nozzle, the moment vector is normal to the plane through
values of b2 $ 0, and, therefore, the second condition both cylinder axis.
renders Principally, the model size and the boundary conditions
used for modelling a shell-nozzle intersection strongly influ-
b2 ˆ 2:4:
ence the calculation results with increasing ratio of nozzle
Fig. 6(b) shows the stress distribution in the deviatoric map diameter to shell diameter and decreasing wall thickness.
for the outer surface: The example under consideration was used to investigate
these influences; some results are given briefly in the
• linear-elastic stress according to moment and pressure
following:
load …sij †le;…M1p† —thin green line;
• elasto-plastic stress according to moment and pressure • Proceeding from a shell-nozzle intersection with an inner
load …sij †ep;…M1p† —thick green line; shell diameter of 496 mm, an inner nozzle diameter of
• linear-elastic stress according to moment load only 367 mm and wall thickness of 4 mm, the linear-elastic
…sij †le;M —thin orange line; behaviour of FE models with different shell length and
• self-equilibrating stress according to moment and pres- with boundary conditions as applied to the model in
sure load …sij †res;…M1p† —thin violet line; Section 3.2, loaded with internal pressure only, was
R. Preiss / International Journal of Pressure Vessels and Piping 76 (1999) 421–434 431

studied. The maximum of the Mises’ equivalent stress Because of the symmetry of the structure and the load,
showed a strong dependence of the length of the shell— only the half structure was modelled and, therefore, the
the difference was about 12% for models with 1500 and moment load applied to the coarse model was half of the
2000 mm total length of shell, respectively. Therefore, a specified value. In the plane of symmetry a symmetry
restriction of the model size to minimise computation boundary condition was applied to all the nodes.
time is not admissible. The location of the maximum of To apply the moment load on the top of the nozzle
the Mises’ equivalent stress was the lowest point of the properly, the rotational degrees of freedom about the axis
intersection curve (on the symmetry plane normal to the normal to the plane of symmetry of all nodes of the flat end
shell axis). of the nozzle were coupled. Furthermore, the middle node of
• As the example was specified initially to compare results the left flat end of the shell was supported in vertical
with fatigue tests performed at WTCM-CRIF in and horizontal direction and the middle node of the right
Belgium, also a model with a geometry exactly the flat end of the shell was supported only in the vertical
same as used in the test was studied. This model had a direction.
greater shell thickness near the ends, and flat ends at the The loading state with a constant moment M ˆ 15644.4/
shell and the nozzle (see specification of examples). 2 ˆ 7822.2 Nm, as in the specification, and an internal
Again, the linear-elastic behaviour of this model under pressure of p ˆ 1.2 MPa was found to be relatively close
internal pressure load only was studied: the maximum to the limit state of the structure, and, therefore, will be
Mises’ equivalent stress was about half of the one of called the “limit state” in the following.
the model stated above (with the same total length of As there is no possibility within the submodel technique
the shell). Additionally, the location of the maximum to calculate different load steps during a particular analysis,
of the Mises’ equivalent stress was completely differ- separate calculations for the coarse model and the submodel
ent—it was the saddle point of the intersection curve. were necessary to obtain the required stress states—internal
• The limit load of the model corresponding to the fatigue pressure linear-elastic, moment only linear-elastic, moment
tests was almost twice of the limit load of the other model only elasto-plastic and the limit state elastic-plastic.
(the limit loads were determined with the very same The computation times were:
material strength parameter, Mises’ yield condition and
• 2 min for the coarse model and 5 min for the submodel
constant nozzle moment).
for the load case internal pressure linear-elastic;
• 2 min for the coarse model and 5 min for the submodel
As a straight-forward conclusion: the knowledge of the
for the load case moment only linear-elastic;
real “boundary conditions” of a real shell-nozzle intersec-
• 20 min for the coarse model and 55 min for the submodel
tion with such extreme diameter ratios and small wall thick-
for the load case moment only elastic-plastic;
ness is a basic requirement for FE calculations of practical
• 1 h and 3 min for the coarse model and 3 h and 12 min
significance. Therefore, the FE model used finally had
for the submodel for the load case limit state elastic-
exactly the same geometric features as the one tested.
plastic on the compaqw workstation.
To model the whole structure with solid elements using
an appropriate number of elements in wall thickness direc- The elasto-plastic FE analyses were carried out using
tion would result in an FE model with too many elements Mises’ yield condition and associated flow rule, a linear-
and nodes and, therefore, in an unacceptable long computa- elastic ideal-plastic law with a material strength parameters
tion time. Therefore, the shakedown check was performed of 272 MPa for the shell and the nozzle, and first order
by submodel technique and, for this purpose, and due to the theory.
large diameter-to-wall thickness ratio of the structure, 8- A linear combination of self equilibrating stress fields—
node shell elements shell93 have been used for the “coarse the one according to the limit state …sij †res;…M1p† and the one
model”. The submodel consisted of 13600 8-node brick according to the moment load only …sij †res;M —was used to
solid elements solid45. To get proper stress results, 4 fulfil Melan’s theorem.
elements were arranged in wall-thickness direction of the Again, for the determination of the factors b1 ; b2 and a
submodel. The submodel had a total length of 440 mm, the the deviatoric maps of the stress states are used.
highest cross-section of the nozzle had a distance of 280 mm The two critical locations of the structure are the inner
to the centre line of the shell, and the angle between the surface of the shell-nozzle junction (path I in Fig. 8, the fillet
lower cut-boundary of the shell and the x-axis was 308. To is excluded), where the maximum linear-elastic calculated
avoid stress concentrations, the weld surface was modelled stress arises in the case of internal pressure only (see Fig. 8),
with a fillet of 4 mm radius and the inner edge of the inter- and the intersection curve of the shell and the nozzle on the
section was modelled with a fillet of 2 mm radius. In the outer surface of the structure (path A in Fig. 9), where the
plane of symmetry a symmetry boundary condition was maximum linear-elastic calculated stress arises in the case
applied to all the nodes, the boundary conditions of the of moment load only (see Fig. 9). As can be seen from the
cut-boundaries were interpolated by the software from the following deviatoric maps, it is necessary to consider the
corresponding displacements of the coarse model. intersection curve from point X to point Y, a consideration
432 R. Preiss / International Journal of Pressure Vessels and Piping 76 (1999) 421–434

Fig. 8. Critical location/path I.

only of the point where the maximum stress arises would • linear-elastic stress according to moment load only
result in non-conservative results. …sij †le;M —orange line;
Again, the self-equilibrating stress field according to the • self-equilibrating stress according to moment and pres-
moment load …sij †res;M is equal to 0 at the inner surface sure load …sij †res;…M1p† —blue line.
(paths I). Therefore, the necessary conditions are given by
For the points at the outer surface (path A) the necessary
(12) and (13). Thus, b1 and a can be determined from these
conditions are now given by
two conditions
f‰b2 …sij †res;M 1 …sij †M;le 1 a…sij †p;le Š # 0; …16†
b1 ˆ 0; a ˆ 0:858:

I.e. the self-equilibrating stress field according to the limit f‰b2 …sij †res;M 1 …sij †M;le Š # 0: …17†
state …sij †res;…M1p† cannot be used because of the distribution
Fig. 6(d) shows the stress distribution in the deviatoric map
of the linear-elastic calculated stresses due to the moment
for path A:
load only at path I. Therefore the maximum admissible
internal pressure (at path I) is given by the condition that • linear-elastic stress according to pressure load
the sum of the two linear-elastic stress distributions are 0:858…sij †le;p —thin green line;
completely inside the limit circle in the deviatoric map, • linear-elastic stress according to moment load only
which renders the value of a for path I. …sij †le;M —thin orange line;
This behaviour can be visualised directly in Fig. 6(c), • elasto-plastic stress according to moment load
which shows the stress distribution in the deviatoric map …sij †ep;M —thick orange line;
for the part of path I located at the shell, where the maxi- • self-equilibrating stress according to moment load only
mum stresses arise: …sij †res;M —thin blue line.
• linear-elastic stress according to moment and pressure It can easily be seen, that the critical point at the outer
load …sij †le;…M1p† —thin green line; surface is point X, because of its large value of linear-elastic
• elasto-plastic stress according to moment and pressure stress due to the pressure load, and not the point where the
load …sij †ep;…M1p† —thick green line; maximum linear-elastic stress due to the moment load only
R. Preiss / International Journal of Pressure Vessels and Piping 76 (1999) 421–434 433

Fig. 9. Critical location/path A.

is located. Further, as the self-equilibrating stress according the linear-elastic stress due to pressure load (which must be
to moment load only in X is 0, the admissible pressure to smaller than 0.858) is determined. This value is found to be
fulfil Melan’s theorem is very low—i.e. the admissible 0.425 and, therefore, a lower bound shakedown pressure is
value of a for path A is (independently from b2 ) close to 0. given by
The conclusion is that the used self-equilibrating stress
field, according to moment load only, is not an optimal one, PSmax SD ˆ ap ˆ 0:425 × 1:2 MPa ˆ 0:51 MPa:
and, therefore, another self-equilibrating stress field,
Fig. 6(e) shows some corresponding stress distributions in
according to a moment load only, is used—the one which
the deviatoric map for path A:
corresponds to a moment of M ˆ 14632 Nm. This moment
load is smaller than the limit moment of the structure. The • linear-elastic stress according to the large moment load
reason for the use of this moment is the plastification in …sij †le;M —thin orange line;
point X at this loading state and, therefore, a non-zero • elasto-plastic stress according to large moment load
self-equilibrating stress according to this moment load in …sij †ep;M —thick orange line;
point X. • self-equilibrating stress according to the large moment
As elasto-plastic results cannot be scaled up, an addi- load …sij †res;M —thin blue line;
tional FE calculation with this higher moment load had to • sum of the self-equilibrating stress according to the large
be carried out. The corresponding computation times were moment load …sij †res;M and the linear-elastic stress
1 h 12 min for the coarse model and 4 h and 7 min for the according to the specified moment load …sij †le;M —thin
submodel. yellow line;
The new self-equilibrating stress field according to the • sum of the self-equilibrating stress according to the large
larger moment load M is 0 at path I again, and, therefore, moment load …sij †res;M and the linear-elastic stress
does not change the admissibility conditions at this path. according to the specified moment load …sij †le;M and the
At path A now a value of b2 ˆ 1 for the self-equilibrating linear-elastic stress field according to the pressure load
stress field is used, and a new corresponding value of a for 0:51…sij †le;p —thin red line.
434 R. Preiss / International Journal of Pressure Vessels and Piping 76 (1999) 421–434

4. Conclusions Acknowledgements

From the examples discussed in this article the following Some of the research given in this article is a part of the
conclusions can be drawn. results of the European research project “Guide lines for the
If shell elements are used, the shakedown analysis is use of the results of elasto-plastic FEM” [2], and the soft-
difficult, as only surface stresses are available in the post- ware package ansysw was used under a University license.
processor of the FE software. If shell elements cannot be
avoided, e.g. for structures with large diameter to wall thick-
References
ness ratios, a check via submodeling of the distribution of
the self-equilibrating stress field which is used in Melan’s [1] prEN 13445-3 Annex 5.B: Direct route for design by analysis.
shakedown theorem may be suitable. [2] Guidelines for the use of the results of elasto-plastic FEM, based upon
The practical relevance of results in nodes of the inter- the Tender in response to call for Tender S 150/31 NL Petten (97/S
section curve of two shells depends on the kind of geometry 150-95149/EN).
of the structure, e.g. for cylinder–cylinder intersections the [3] Zeman JL. Some aspects of the work of the European working groups
relating to basic pressure vessel design. IJPVP 1997;70:3–10.
nodes on the intersection curve do not correspond to points [4] Zeman JL, Preiss R. The deviatoric map—a simple tool in DBA.
of the real structure. If cyclic plastification occurs in these IJPVP 1999;76:339–344.
nodes no statement of the behaviour of the structure can be [5] Zeman JL. Repititorium Apparatebau—Grundlagen der Festigkeits-
made according to these results. berechnung, Oldenbourg, 1992.
In general, for thin walled structures with nozzles, the [6] Gokhfeld DA, Cherniacsky OF. Limit analysis of structures at thermal
cycling, Sijtkoff & Noordkoff, 1980.
knowledge of the real “boundary conditions” of the struc- [7] Rammerstorfer FG. On the modelling of cracks by finite elements.
ture is a basic requirement for FE calculations of practical Scand J Metall 1983;12:293–298.
significance—the importance increases with the nozzle [8] König JA. Shakedown of elastic–plastic structures. Amsterdam: Else-
diameter. vier, 1987.
The determination of the shakedown action of structures [9] Burgreen D. Design methods for powerplant structures. C.P. Press,
1975.
in cases with more than one action by superposition of stress [10] Preiss R. Das Einspielverhalten des Balkens mit Rechteckquerschnitt
states in the postprocessor of the FE software is often not bei Biege- und Normalkraftbelastung, Bericht Nr. 13, Institut für
feasible. In these cases the use of the deviatoric map is Apparate- und Anlagenbau (A&AB), TU Wien, 1998.
suitable, but care must be taken in the selection of the points [11] Zyczkowski M. Combined loadings in the theory of plasticity; PWN.
(locations) which are considered—considering only the Polish Scientific Publishers, 1981.
[12] Chen WF, Han DJ. Plasticity for structural engineers. Berlin:
locations where the maximum stresses arise can lead to Springer, 1988.
non-conservative results. [13] ansysw5.4 Analysis Guides.

You might also like