Sunteți pe pagina 1din 10

Comput Mech (2006) 37: 121–130

DOI 10.1007/s00466-005-0664-7

ORIGINAL ARTICLE

Pauli Pedersen

On Shrink Fit Analysis and Design

Received: 19 October 2004 / Accepted: 28 January 2005 / Published online: 6 September 2005
Ó Springer-Verlag 2005

Abstract Shrink fit or interference fit is an important parts in a reliable and non expensive manner. A thick
connection in a number of mechanical parts. In contrast cylinder that must sustain high internal pressure is an
to the importance, the number of three dimensional important example, and so are shaft disk connections
finite element analyses of this contact problem is not that must transmit high torsional and/or axial loads.
large, because contact between two or more bodies in Stress concentration is important for the fatigue of the
general is a particularly difficult non-linear behaviour to shafts as described in Norton (2000) with reference given
analyze. to the early research of Peterson and Wahl (1935). It is
In the present paper we show how relatively simple recognized that the design of the shrink fit components is
axisymmetric analysis is possible, and we describe two an important issue, and still no final answer is given to
different points-of-view. Classical plane analysis is eval- the problem.
uated and the design of shrink fit surfaces is included. For the two dimensional classical models, we have
The first method determines the shape of a shrink fit analytical results that follow from a formulation which
surface that will result in a prescribed distribution of gives the Euler differential equation. We first show
contact pressure due to the shrink fit. such a two dimensional specific solution and then
The second method uses a super element technique, evaluate these results relative to an axisymmetric finite
and determine without iteration and incrementation element solution. As is well known, the two dimen-
the distribution of contact pressure that will result sional model lacks information about the changes in
from a prescribed shrink fit interference. Simplicity is the axial direction, and this might return some
our goal, so specific linear elastic shrink fits without important modifications to the contact stresses, to the
friction are treated, but a number of extensions are circumferential stresses, and to the von Mises stresses.
possible. Furthermore it puts our attention toward the fact that
shrink fit interference is not just a value, but should
Keywords Shrink fit Æ Interference fit Æ Shape be treated as a function of the axial direction z.
optimization Æ Optimal design Æ Contact Immediately, the question of design of this interference
stresses Æ Parameterization function (shape of the axisymmetric contact surface) is
important.
In contrast to the importance of shrink fit, the
number of papers on three dimensional finite element
analysis of this contact problem is not large, because
1 Introduction contact between two or more bodies is a particularly
difficult non-linear behaviour to analyze. The literature
Shrink fit or interference fit is an important connection on the solution of contact problems in general is large,
in a number of mechanical parts, say to obtain a pre- with a number of different suggested solution proce-
stress (residual stress) state or simply to connect different dures. Most of these methods are based on incremental
simulation and iteration with application of tools from
optimization, see the extended reviews in Zhong and
Mackerle (1992) and Mijar and Arora (2000), and in the
P. Pedersen book by Wriggers (2002).
Department of Mechanical Engineering, Solid Mechanics
Technical University of Denmark Nils Koppels Allaè,
Contact problems are difficult to solve, being highly
Building 404, DK-2800 Kgs Lyngby, Denmark non-linear and often extremely sensitive. A number of
E-mail: pauli@mek.dtu.dk classifications are possible:
122

(a) only one flexible body (in contact with rigid bodies) involved in these linear elastic finite element analyses,
or multiple flexible bodies in contact that therefore are robust and inexpensive even with a
(b) non-friction contact for the analysis or a friction large number of degrees of freedom (dof.).
model with, say stick and slip The second method uses a super element technique,
(c) elastic material or including non-elastic behavior and determines without iteration and incrementation
(d) linear elastic or non-linear elastic the distribution of contact pressure that will result
(e) linear geometry (linear strain measure) or non-linear from a prescribed shape of the shrink fit surface equal
geometry to a prescribed interference function (vector). A
(f) known contact surface or unknown contact surface practical application of the super element finite ele-
ment method is described. The final expression for the
The shrink fit problem treated in the present paper
nodal contact forces involves the inverse of the sum of
is based on a number of assumptions and may hardly
two inverse matrices, but the order of these matrices is
be named a contact problem. We show that it is
only equal to the number of contact dof., and there-
possible to obtain a direct solution without iteration.
fore inexpensive.
The obtained simplicity may be applied to more ad-
Simplicity is our goal, so only linear elastic shrink fits
vanced models, including incrementation and/or iter-
without friction are treated, but a number of extensions
ation. So to classify our problem relative to (a)–(f) we
are possible.
treat two flexible bodies in contact without friction.
Linear elastic material behavior is assumed with iso-
tropic material (direct extension to anisotropic mate-
rials is possible). Engineering strains with equilibrium
in undeformed system, and to a large extend the 2 A simple plane test case
contact connections are known.
We show that it is possible to use relatively simple As a case for later reference we analyze a plane problem
axisymmetric analysis, and we describe two different of shrink fit, taken from Benham et al.(1996). The two
points-of-view. The first method determines the shape of rings (cylinders) to be fitted are shown in Fig. 1 with
a shrink fit surface which will result in a prescribed definition of the length parameters: a for the inner radius
distribution of contact pressure due to the shrink fit. For of the inner ring, b for the nominal outer radius of the
practical reasons a few parameter description of contact inner ring equal to the nominal inner radius of the outer
pressure is applied, and the corresponding equivalent ring, and c for the outer radius of the outer ring. The
nodal forces in the axisymmetric model are derived. shrink fit interference e ¼ bi  bo  0 is the actual design
Thus the user only has to change one to three parame- parameter. The two rings are here of the same material
ters in order to obtain the nodal force model for another with modulus of elasticity E, and the internal pressure in
contact pressure distribution. The two bodies in contact the inner ring is p.
are analyzed individually. Assuming a state of plane stress we get, as is well
From the resulting displacements for the nodes in known, the Euler differential equation for the radial
contact follows an interference function (vector) that displacement. We can solve this problem analytically
gives the shape of the shrink fit surface(s), necessary to and get the two principal stresses as a function of radial
obtain the prescribed contact pressure. Contact is not position r, inner pressure load p, scalar value of shrink

Fig. 1 Shrink fit model from


Benham et al. (1996) with
internal pressure p and shrink fit
interference e
123

fit interference e, and radii a, b and c for the cylinders,


see, Fig. 1.

For inner part, i.e., a  r  b


   
1 2 c 2
rr ¼  pð2a bÞ 1
2bðc2  a2 Þ r
 a2 
2 2
 eEðc  b Þ 1 
r
    
1 2 c 2
rh ¼  pð2a bÞ  1
2bðc2  a2 Þ r
 a2 
 eEðc2  b2 Þ 1 þ ð1Þ
r
Fig. 3 Resulting circumferential rh and radial stresses rr , alone from
For outer part, i.e., b  r  c shrink fit (no internal pressure)
   
1 2 c 2
rr ¼  pð2a bÞ 1 accounted for in the simple, analytical plane analysis.
2bðc2  a2 Þ r
   We therefore leave the internal pressure p out and
2 2 c 2
 eEðb  a Þ 1 concentrate on the stresses related only to the shrink fit
r interference e. The actual parameters with p ¼ 0 gives
    
1 2 c 2 the results shown in Fig. 3.
rh ¼  pð2a bÞ  1 Using a three dimensional, axisymmetric finite ele-
2bðc2  a2 Þ r
    ment model we validate the accuracy of this solution
2 2 c 2
 eEðb  a Þ  1 ð2Þ based on the plane stress assumption. Note, that the
r explicit solution in (1) and (2) is independent of Pois-
With the specific values from Benham et al. (1996): son’s ratio m. With m 6¼ 0 the inner part expands in the
a ¼ 0:10 m, b ¼ 0:15 m, c ¼ 0:20 m, e ¼ 0:0001 m, E = axial direction due to the shrink fit pressure, while the
207 G Pa, p = 200 MPa, Fig. 2 shows the actual solu- outer part shrinks in the axial direction, so m must have
tion. The maximum circumferential stress without some influence. Also the stress distribution will not be
shrink fit is 333 Pa and with the actual shrink fit as constant in the axial direction z, and especially at axial
shown in Fig. 2 only 265 Pa. The circumferential stress boundaries the modifications to the shown solutions
rh may be critical at the inner boundary r ¼ a or at the may be important.
common boundary in the outer part at r ¼ b. To get Even with the use of a finite element program, contact
minimum of maximum circumferential stress we may set analysis is a complicated non-linear problem and many
rh ðri ¼ aÞ ¼ rh ðro ¼ bÞ and get a solution in the value alternative solution algorithms are suggested in the lit-
for the shrink fit interference e or for the position b at erature. Reviews are cited in the introduction, and see
the common boundary. The shown solution is close to also the thesis by de la Cour (2003). The algorithm used
this optimal shrink fit design. in the present paper is derived from an inverse point of
However, the shrink fit designs of the present paper view. We aim at designing the contact surfaces to give a
are depending on the axial dimension which is not prescribed distribution of the contact pressure. First we
deal with the practical problem of describing a load
distribution and the transfer of this distribution to finite
element nodal point forces.

3 Pressure and force modeling

3.1 An analytical pressure distribution along a curve

The intention is to find for the contact pressure distri-


bution a parameterization with relatively few parameters
that gives an analytical smooth description, and still is
flexible in use. The resulting distributions are assumed
non-negative with zero value at the end of the contact-
surface (except when symmetry is utilized). Good expe-
Fig. 2 Resulting circumferential rh and radial stresses rr with riences with skew super ellipses in modeling for shape
combined internal pressure and shrink fit design is behind our choice.
124

With reference to the parameter presentation de- The two nodal points are denoted 1 and 2, and a
scribed in Pedersen (2004), we use p ¼ pðsÞ, where p is specific point on the pressure loaded side is denoted r
the pressure as a function of the length position s on the corresponding to the radial coordinate of this point.
contact curve. In addition to the total length 2S of the With linear displacement assumption the displacement
contact curve only three parameters are involved, i.e., vr at the actual point becomes
the magnitude p0 of the pressure at the middle (s ¼ S) of
the contact curve, the super-ellipse power g (shape of the r  r1
vr ¼ v1 þ  ðv2  v1 Þ ð4Þ
distribution), and a skewness parameter c (1 < c < 1). r2  r1
The algebraic expression for p ¼ pðsÞ is The pressure function (3) gives the pressure value at the
 
sS nodal points, and we assume a linear variation of the
p ¼ p0  1 þ c  sinðtÞð2=gÞ pressure along this short part of the total contact curve.
S
Then for the pressure at the actual point we get
with the parameter t from
! r  r1
pr ¼ p1 þ  ðp2  p1 Þ ð5Þ
js  Sjðg=2Þ r2  r1
t ¼ arccos for 0  s  2S ð3Þ
S where p1 ; p2 are the nodal pressure values, not to be
mistaken for the nodal forces Q1 ; Q2 . Now the work
A classical ellipse is obtained for g ¼ 2, an almost rect-
done by the nodal loads Q1 ; Q2 must equal the work
angular distribution for large g, say g ¼ 10, a symmetric
done by the distributed pressure for all possible nodal
triangular distribution for g ¼ 1. For c ¼ 0 the skewness
displacements v1 ; v2
parameter gives a symmetric distribution relative to the Z r2
middle of the contact curve at s ¼ S. Triangular skew
Q1  v1 þ Q2  v2 ¼ ðpr  vr  2prÞdr ð6Þ
distributions are obtained when c ! j1j. When symme- r1
try is utilized in the finite element modeling, half of the
function (3) is used. The resulting polynomial to be integrated has up to third
order terms in r, so we use a formula manipulation
program to find the result, and the simpler case of
3.2 Nodal loads from pressure on an axisymmetric r2 ¼ r1 is treated separately. The final formulas, directly
model suited for computer programming, are omitted here.
Note, that (6) is general and not restricted to the linear
Figure 4 shows the parameters for derivation of the function (5).
equivalent nodal loads Q1 ; Q2 in the axisymmetric finite
element model based on linear displacement triangular
ring-elements. We apply the standard procedure to 3.3 Examples of pressure distributions
determine equivalent consistent nodal forces.
Figure 5 shows four cases of pressure distributions with
corresponding parameters. The distributions are plotted
directly from the finite element program, and thus show
nodal loads. In case I an almost uniform pressure
distribution is illustrated, then in case II an almost

Fig. 5 Four cases of pressure distributions with corresponding


Fig. 4 Definition of symbols used for the derivation of equivalent parameters. The distributions are plotted directly from the finite
finite element nodal loads element program, and thus show nodal loads
125

symmetric triangular distribution, and in case III an plane analysis illustrated in Fig. 3, we get a gap in
elliptic distribution. These three different distributions mean equal to 0.0001 m, as predicted by the plane
are obtained by only varying the super-ellipse power g. analysis. However the gap is varying through the
The fourth case illustrates the influence of the skewness height. Before discussing this variation, which in fact is
parameter c. our main point of interest, we show and discuss the
Although the loaded surface (curve in the radial-axial overall displacements in Fig. 6 and the three dimen-
plane) is shown in Fig. 5 as a straight line, this is by no sional (axisymmetric) distribution of circumferential
means an assumption. We here treat pressure load, i.e., stress and of von Mises’ stress in Fig. 7. In these
load perpendicular to the loaded surface, but alternative Figures, we for illustration separate the two contact
load directions may also be accounted for. Thus from surfaces (half of the two cylinders) in the radial
the finite element point-of-view, friction modeling by the direction, and we magnify the displacements in Fig. 6.
present load description is possible. The axial extension of the inner cylinder as well as
Finally, it should be pointed out that the used load the axial compression of the outer cylinder is seen in
description rely on the analytical function (3), and thus Fig. 6 (most clearly at the top zoom), and this implies
singularities are avoided in the procedure. that part of the surface for the inner cylinder can not
be loaded (here assumed no friction). The assumption
of plane stress behind the solution in Sect. 2 is vio-
4 Analysis of two axisymmetric cases lated, at the symmetry line (in axial direction) of the
cylinders to the extent that the axial stresses are at the
The test example of Sect. 2 is now analyzed with an same level as the radial stresses. This is seen by the
axisymmetric finite element program, using axial sym- axial nodal reactions in Figs. 6 and 7.
metry relative to the middle of the axial length 2 L, with Figure 7 shows at the top the isolines of the circum-
the value L = 0.1 m. In addition we have the dimensions ferential stresses, which are the dominating stresses. For
in Fig. 1. We use for each of the two parts 7200 ring illustration, the contact pressure nodal loads and
elements with triangular cross-sections as illustrated in the nodal reactions are repeated in these pictures with
Fig. 6, which corresponds to 7382 degrees of freedom in the inner cylinder to the left and the outer cylinder to the
each of the two models. right. The boundary effect at the free end of the cylinder
Applying a contact pressure as shown in case I of is clearly seen, and we show this more quantitatively at
Fig. 5 with the magnitude p0 = 22.4 MPa from the the corresponding pictures at the bottom for the

Fig. 6 Finite element models


with mutual contact pressure
loads, reactions at the axial
symmetry surfaces, and magni-
fied displacements. In the full
models the magnification factor
is 100, while the zooms at the top
and bottom have a magnification
factor of 10. The dotted lines
correspond to the surface of
symmetry in the axial direction
and indicate that the two contact
cylinders are shifted in the radial
direction
126

Fig. 7 Stresses in contact cylin-


ders, resulting from shrink fit.
Top: circumferential stresses and
bottom von Mises’ stresses. Left:
the inner ring (cylinder) and
right: the outer ring (cylinder).
Internal isolines in 32 linear
steps. Nodal loads fAg and
reactions fRg also shown at the
top, and distribution of von
Mises’ stress rvM ¼ rvM ðzÞ along
the axial lines shown by the
thickness of the distribution
added to the pictures at the
bottom. The dotted lines corre-
spond to the surface of symmetry
in the axial direction and indicate
that the two contact cylinders are
shifted in the radial direction

reference stresses, calculated as the von Mises’ stress. At


the inner and outer boundaries for the inner cylinder to
the left and for the outer cylinder to the right, we have
added the resulting distributions along these lines. The
larger von Mises’ stresses at the radial boundaries vary
for the inner cylinder from 71.8 to 81.1 MPa, and for the
outer cylinder from 80.5 to 91.8 MPa.

4.1 An example of a bronze bush shrunk onto


a steel shaft

An other test example from Benham et al. (1996) illus-


trates two different materials and the deformation of the
shaft close to but outside the bush. Figure 8 shows the
dimensions. The material data for the bush is modulus
of elasticity Eb = 100 GPa with Poisson’s ratio mb =
0.29. The material data for the shaft is modulus of
elasticity Es = 209 G Pa with Poisson’s ratio ms = 0.28.
The shrink fit interference corresponding to contact
pressure equal to 69 MPa is asked for and with analyses
similar to (1) and (2) and the actual material data we
find e = 0.000112 m. The results of our finite element
axisymmetric model confirms this (in the mean). Figs. 9,
10 show the results presented in the same (almost) way
as in Figs. 6 and 7.
As expected the bronze bush is heavily stressed, and
of special interest for the present paper are the details at Fig. 8 Dimensions for models of steel shaft and bronze bush
127

Fig. 9 Finite element models of


steel shaft and bronze bush with
mutual contact pressure loads,
reactions at the axial symmetry
surfaces, and magnified displace-
ments with a factor of 50. The
dotted line corresponds to the
surface of symmetry in the axial
direction and indicate that the
bronze bush is shifted in the
radial direction

the bush ends in the axial direction. In general the algorithm. With an effective and reliable tool of analysis,
comments of the first example may be repeated. optimal design is in most cases not too complicated.
With the two dimensional plane stress assumption we
solve in an effective and reliable manner the shrink fit
5 Design of shrink fit surfaces problem and obtain the results for a given shrink fit
interference. However, for the three dimensional prob-
We now turn to the design problem, where the initial lem (even with the axisymmetric assumption) we did not
contact surface may be curved, and ask for the shape, solve the problem for given shrink fit interference
which will return a prescribed distribution of contact (interference function). This problem in general must be
pressure. The step from analysis to design (synthesis), solved incrementally and iteratively, and with friction
and in the wider perspective optimal design, normally the problem is furthermore irreversible. Algorithms in
involves sensitivity analysis and an optimization combination with the finite element method are

Fig. 10 Von Mises’ stresses


resulting from shrink fit. Left: the
steel shaft and right: the bronze
bush. Internal isolines in 32
linear steps. The distribution of
von Mises’ stress rvM ¼ rvM ðzÞ
along the axial lines shown by
the thickness of the distribution.
The dotted line corresponds to
the surface of symmetry in the
axial direction and indicate that
the bronze bush is shifted in the
radial direction
128

Then the inverse problem is formulated: given a shrink


fit interference function equal to this gap eðzÞ ¼ gðzÞ, the
results will be a shrink fit pressure distribution as shown
in Figure 5, case I. That this holds is due to the fact that
an extremely small modification of the design with nodal
position modified by eðzÞ will not influence the direct
analysis.
The skewness in pressure distribution is illustrated in
case IV of Fig. 5 as an alternative to the just used case I.
In Fig. 12 we see how strong an influence this has on the
shape of the gap function, when the parameter c is
changed from the value 0.0 to )0.4. The change in
magnitude of gap is not so interesting, because the total
pressure force has not been scaled to the same value. The
Fig. 11 Gap function g ¼ gðzÞ from analysis with pressure distribu- change in shape for the gap function is to be noted.
tion parameters g = 6 and c = 0.0. The coordinate z is the axial
position from axial symmetry measured in mm, and the gap function The in-between solution of c = )0.2 gives an almost
g (equal to the shrink fit interference function e ¼ eðzÞ is measured in constant gap, except for a drop of 4 lm close to the axial
lm end, to be compared with the 12 lm shown in Fig. 11.
The analysis from given pressure distribution is lin-
ear, direct, robust, and fast. It gives a gap function
available, but often we encounter extremely sensitive
which we can use as the corresponding shrink fit inter-
and often unstable procedures.
ference function. If the pressure distribution corre-
From this point of view, we here suggest to postulate
sponding to a prescribed interference function is our
the shrink fit pressure and then derive the corresponding
goal, it may still be advantageously to find this by iter-
shrink fit interference; in the axisymmetric case , the
ation with the shown procedure, because sensitivity
interference function e ¼ eðzÞ. In reality this is the simple
analysis will also be simple. As an alternative to this, in
linear, reversible analysis we have shown in Sect. 4, and
the next section we present a direct analysis based on a
we therefore return to the detailed results from this
finite element super element formulation.
section. In Sect.6 we present a simplified procedure for
the direct problem.

6 A direct shrink fit analysis by super elements


5.1 The contact surface for two cylinders
With reference to the case in Fig. 6, for illustration, we
Figure 6 shows the magnified displacements of the two analyze the outer cylinder (here index o) with only the
cylinder contact surfaces. With the outer cylinder surface contact nodal degrees of freedom
of the inner cylinder equal to the inner cylinder surface
of the outer cylinder before the prescribed pressure is ½So fDo g ¼ fAg ) Do g ¼ ½So 1 fAg ð7Þ
applied, this will result in a gap, a gap function g ¼ gðzÞ where ½So  is the super element stiffness matrix (in the
which we show in Fig. 11. actual case of order 61 equal to the number of nodes on
the contact surface), fDo g contains the contact nodal
displacements in radial direction, and fAg is the corre-
sponding contact nodal forces (acting in the radial
direction). The determination of the super stiffness ma-
trix ½So  is described later. The solution in terms of the
nodal displacements fDo g for given nodal loads fAg will
give exactly the same result as in the earlier calculation
with the full stiffness matrix (for the case in Fig. 6 of
order 7382). Our goal here is to determine fAg, knowing
fDo g only relative to the displacement of the inner
contact surface.
For the inner cylinder (here index i), with only the
contact nodal degrees of freedom, we have a similar
super element description

Fig. 12 Gap function g ¼ gðzÞ from analysis with pressure distribu- ½Si fDi g ¼ fAg ) Di g ¼ ½Si 1 fAg ð8Þ
tion parameters g = 6 and c = )0.4. The coordinate z is the axial
position from axial symmetry measured in mm, and the gap function We assume that the contact surfaces have the same
g (equal to the shrink fit interference function e ¼ eðzÞ) is measured number of (mutual corresponding) nodes, and then the
in lm equilibrium of the nodal actions is accounted for by
129

assuming the same nodal forces fAg in (8) and (7) except
for the sign.
In a shrink fit analysis the shrink fit interference
(interference function) may be described by a nodal
interference vector feg, that the nodal displacements
fDo g and fDi g should satisfy. Again with reference to
Fig. 6, for illustration, we have
feg ¼ fDo g  fDi g ð9Þ
and inserting the solutions (7) and (8) we get
 
feg ¼ ½So 1 þ ½Si 1 fAg )
 1 ð10Þ
fAg ¼ ½So 1 þ ½Si 1 feg
Fig. 14 Nodal interferences constant, i.e., eðzn Þ = 0.0001 m and
In general we can not always expect a meaningful solu- corresponding resulting nodal actions in radial direction An , with n as
nodal number giving the axial position
tion fAg  f0g corresponding to any prescribed inter-
ference vector feg. Testing for negative elements in fAg,
as obtained from (10), we know if the assumed interfer- with static condensation of the non contact degrees of
ence vector feg is possible from a physical point-of-view. freedom fDgn and thus getting of lower order system,
To test the programmed procedure we calculate fAg say of order 10 to 100, for the contact degrees of free-
from the vector feg corresponding to the interference dom fDgc . For our case fAgn ¼ f0g, thus no conden-
function shown in Fig. 11. Figure 13 shows that the sation of fAgc is involved and we get
result agree completely with the force distribution of
case I in Fig. 5. ½Sc fDc g ¼ fAc gwith ½Sc  ¼ ½Scc  ½Scn ½S1 T
nn ½Scn ð12Þ
Figure 14 shows the force distribution in the radial where fDc g; fAc g are the actual contact degrees of free-
direction (corresponding to radial stresses rr ) that re- dom and corresponding nodal loads, as exemplified in
sults from a constant interference function (feg ¼ ef1g), (8) and (7). This technique is well known, and we shall
and this force distribution is almost constant. only comment on the practical evaluation of the super
element matrix ½Sc . Note that this is just an alternative
solution technique and not a further approximation.
6.1 Calculation of a super element matrix The practical evaluation of the super element matrix
½Sc  as specified by (12) may seem complicated, but from
Solution by a super element algorithm as in (8) and (7) the basic definition of stiffness, as force (at a contact
corresponds to solving the overall finite element equi- dof.) per unit displacement (at the same or another
librium equation, say of order 10000 to 100000, contact dof.), follows that we can get a specific column
½SfDg ¼ fAg or of ½Sc , say column k, by a unit forced displacement at
" #    the corresponding dof. All the other contact dof. are
½Scc ½Scn fDgc fAgc fixed to zero, and the calculated reactions corresponding
¼ ð11Þ to the contact dof. is then the column k of ½Sc .
½STcn ½Snn fDgn fAgn
This implies that without performing the matrix
calculation in (12), we obtain the super element matrix
by solving a number of load cases, corresponding to
the number of contact dof., and doing so by proce-
dures already available in a finite element code. The
computational costs of solving for these additional
load cases correspond at most to one additional Gauss
factorization, and for larger systems far less computer
time is used.

7 Perspectives

Our primary goal for this project is to get additional


insight about the three dimensional effects of shrink fit
connections. For this purpose we used two, in principle,
Fig. 13 Nodal interferences as specified, i.e., eðzn Þ and corresponding
different approaches with a number of simplifying
resulting nodal actions in radial direction An , with n as nodal number assumptions. The two approaches may have a wider
giving the axial position range of applications, which are sketched in this section.
130

In the first approach, the complications of the but two direct procedures are described and documented
contact problem are eliminated by assuming the mu- by examples. This means that with assumptions of no
tual contact pressure to be given, and thus the contact friction and a known contact domain, we have robust
bodies can be analyzed separately. This analysis can and inexpensive 3D determination of axisymmetric
therefore, without too much trouble, be extended to shrink fits.
include non-linear elasticity and even plasticity. The The first method determine a shrink fit surface that
extension to include a large strain measure is also will result in a prescribed distribution of contact pres-
possible, but in shrink fit analysis this is not needed. sure due to the shrink fit. With separated models for the
With known pressure, a friction modeling should also two parts to be shrink fitted, traditional finite element
be possible but for stick slip modeling an incremental linear elastic analyses give the results. Practical aspects
procedure is needed. of parametrization of the contact pressure and its
For non contact problems we have rather simple pro- translation into nodal forces are described.
cedures for sensitivity analysis, so it should be possible to The second method uses a super element technique
iterate on the assumed pressure distribution, and in this and determine without iteration and incrementation the
manner solve the problem for a given shrink fit interfer- distribution of contact pressure that will result from a
ence. For such a sensitivity analysis with respect to change prescribed shrink fit surface. With super element stiffness
in the pressure distribution, the suggested simple param- matrices for both parts of the shrink fit we evaluate the
etrization of the nodal load distribution is valuable. inverse of the sum of the two inverted super element
However, the second approach is oriented toward stiffness matrices, and then directly get the nodal forces
this direct problem of calculating the pressure distribu- corresponding to the contact pressure. The super ele-
tion resulting from a prescribed shrink fit interference ment matrices are of the order 100 in contrast to the
function. With assumed active contact nodes it is shown total element matrices of order 10000–100000, thus the
that also this problem can be solved directly, without process of getting the inverse is computational non
incrementation and iteration. Adding iteration to the expensive.
approach may give a possibility to locate the active A number of extensions for the used procedures are
contact nodes, and the approach includes a test for indicated, and primarily the important connection
activeness. In this manner the assumption of known problem with a pin in a hole will be studied.
active contact nodes may be removed.
Super element technique was used in Pedersen et al.
(1998) to model hardness tests (indentation) and this
References
approach is in an extended version used to calculate the
mutual nodal contact forces. In principle the two bodies Benham PP, Crawford RJ, Armstrong CG (1996) Mechanics of
in contact are again analyzed separately, and only for engineering materials. Harlow, Essex, England: Longman, 2nd
the final calculation is the sum of two inverted super edition, p. 627
element matrices applied. For a number of contact de la Cour DD (2003) Identification of material and friction
parameters from deep drawing. Number S 90. Kgs. Lyngby,
problems where super element technique is valid this Denmark: DCAMM. Ph.D.-thesis, p. 102
approach may be valuable. A specific and important Mijar AR, Arora, JS (2000) Review of formulations for elasto-
case is the case of pin hole interaction for isotropic as static frictional contact problems. Struct Multidisc Optim
well as for anisotropic material behaviour. 20:167–189
Norton RL (2000) Machine design - an integrated approach. New
Jersey 07458, USA: Prentice-Hall, 2nd edition, p.1078
Pedersen P, Mortensen SA, Larsen DG (1998) A procedure for
8 Conclusions contact problems used in hardness test analysis. In Eriksson A,
Pacoste C (eds), NSCM11: Nordic Seminar on Computational
Shrink fit or interference fit is traditionally modeled in Mechanics, 31–34
Pedersen NL (2004) Optimization of holes in plates for control of
2D with a plane stress model for which analytical eigenfrequencies. Struct Multidisc Optim 28(1):1–10
solutions are known. With an axisymmetric finite ele- Peterson RE, Wahl AM (1935) Fatigue of shafts at fitted members,
ment model for the corresponding 3D problem we with a related photoelastic analysis. ASME J Appl Mech
illustrate the influence from the axial boundary condi- 57:A1–A11
Wriggers P (2002) Computational contact mechanics. Chichester,
tions. UK: Wiley. p. 441
Contact problems with two flexible bodies is even Zhong Z-H, Mackerle J (1992) Static contact problems - a review.
with the tools of finite element a complicated problem, Eng Comput 9:3–37

S-ar putea să vă placă și