Journal of Materials Processing Technology 109 (2001) 3037

BE analysis of shafthub couplings with polygonal proles

R. Citarella*, S. Gerbino
Dipartimento di Progettazione e Gestione Industriale, Universita degli Studi di Napoli Federico II, P.le Tecchio 80, 80125 Napoli, Italy

Abstract
In the mechanical transmission eld, shafthub couplings with polygonal proles play an interesting role because of their characteristics
of self-alignment, lack of projecting elements (responsible for high stress concentration) and constructive compactness. Other
characteristics, like transmission of static/oscillating torque load, even with small overall dimensions, and easy hub interchangeability,
make such couplings competitive with the traditional ones based on keys and splined shafts.
This work concerns a study on steel made polygonal couplings, with trochoidal three-lobe prole, and is aimed to highlight the contact
stress and strain state of shafthub interface, with reference to particular prole geometric parameters.
From Mechnik's and Kollmann's works, in which the analysis was performed by the Finite Element Method, this work develops a CAD/
CAE methodology for coupling design, oriented to an efcient integration between CAD systems and BEM solvers. The stress analysis is
carried out with a Boundary Element code (BEASY) well suited for this kind of contact problems while coupling geometric model is made
by Pro/Engineer, a solid parametric modeller. # 2001 Elsevier Science B.V. All rights reserved.
Keywords: CAD; BEM; Polygonal prole

1. Introduction
The interest in polygonal prole shafthub couplings is
due to common characteristics as: constructive compactness; lack of projecting elements responsible for high stress
concentration; selfalignment; transmission of static-oscillating torque load (even with overall small dimensions);
axial sliding capability with applied loads, for some polygonal shapes, such as four-lobe couplings; easy hub interchangeability (useful in case of failure) [1].
Nevertheless, these joints are very complicated and costly
to manufacture because they require dedicated grinding
machines and the stress analysis is very difcult due to
triaxial stress state and lack of rotational symmetry. For a
long time, these drawbacks have limited their use, discouraging design and production engineers to leave traditional
connections based on keys and splined shafts.
In recent years there has been an increasing interest in
polygonal couplings (Fig. 1), mainly with reference to threelobe couplings (Fig. 1c and e).
Mechanical literature presents only a few works on these
joints. The rst studies date back to the 1950s in Germany
and proposed the so-called K prole (Fig. 1c). In the 1960s
epicycloidal proles with three- and four-lobes [2] were
made and, in 1979, included in the standard DIN (P3G and
P4C) [3,4]. Fig. 2 shows a standard three-lobe prole (P3G)
*

Corresponding author.

and its parametric equation (previously dened by Filemon

[5] in 1959), valid for n-lobe proles.
During that period, some analytical procedures of stress
analysis have also been developed, like that by Orlov [1],
Leroy and Viseur [6], Manhurim [7], based on very strong
approximations that are not able to clarify the real stress and
strain state of polygonal couplings (see also [8]).
Only recently, a sufciently detailed analysis of the
biaxial stress and strain state in shafthub joints has been
proposed by Mechnik, using the Finite Element Method
(FEM) [9]. By improving the connection model, Mechnik
obtained very reliable results.
After this work, a deeper understanding of this topic has
been reached [10,11]. Despite these recent results, the
employment of polygonal couplings is still very nite.
Traditional and already consolidated couplings based on
keys and splined shafts are still strongly preferred in the
mechanical transmission eld, because they are easy to
manufacture and size.
Nowadays, the development of numerical control grinding machines together with the improvement of hardware
and software resources has allowed to reduce the above
disadvantages, making the future of the polygonal couplings
very interesting.
From a production point of view, many drawbacks could
be circumvented using powder metallurgy (PM), as proposed in [12]. Lately, this technology has been more and
more appreciated in the automotive eld. It enables to obtain

0924-0136/01/$ see front matter # 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 9 2 4 - 0 1 3 6 ( 0 0 ) 0 0 7 7 2 - X

R. Citarella, S. Gerbino / Journal of Materials Processing Technology 109 (2001) 3037

Fig. 1. Examples of shafthub couplings with polygonal proles.

production time and cost savings and to manufacture parts

with complex shapes and high mechanical performances.
Some recent works [13] demonstrate the benets of sintered
steels in applications which, up to one decade ago, were
reserved to special steels. For parts subject to fatigue,
sintered steel offer the best performance [14] and further
development on polygonal proles implementation is likely
to involve such materials.
The authors point out a CAD/CAE methodology for
coupling design, oriented to an efcient integration between
CAD systems and BEM solvers. The aforesaid methodology
is based on a synergetic and strictly integrated usage of the
Pro/Engineer CAD system, for geometric model generation,
and of the BEASY code for stress analysis.
2. Analysis system choice
In structural analysis, the Boundary Element Methodology offers some advantages with respect to FEM, with
particular regard to the preliminary stage of new com-

Fig. 2. Polygonal prole with n 3 lobes and its parametric equation.

31

ponent design when different engineering solutions are to

be compared. In fact, FEM would require a tremendous
amount of data (because of the volume meshing) making it
difcult to quickly update the model subject to continuous
development.
Stress and strain analysis strategy of mechanical components, up to now traditionally based on FEM, has recently
involved BEM. An advantage related to this methodologies,
is that, on the contrary of FEM, BEM allows to operate
directly on the boundary model, neglecting the volume, and
to calculate surface stress with the same accuracy as surface
displacements. That is why, it is well suited for stress
analysis in problems with high stress gradient such as
contact problems. Moreover BEMCAD coupling and the
meshing process are straightforward. This coupling is essential in shape optimisation analysis where it is necessary to
iteratively update the model geometry after stress analysis,
in an integrated and automatic process. Such an automatic
process in an FEM environment would cause increasing
mesh distortion in areas susceptible to variable geometry as
successive re-meshing goes on.
The study of geometrically complex models, by numerical analysis codes, always needs a preliminary work of
model simplifying. In particular, before analysing a model, it
is convenient to delete those features (llets, chamfers)
which do not affect the nal results, but are responsible
for increasing analysis run time. Actually, by using FEM, in
order to populate the volume with elements, exact geometric
details must often be sacriced, even if this affect the
solution accuracy [15,16]. Moreover, element shapes are
often distorted, thus causing loss of solution accuracy, whilst
the proper choice of correct simplications requires special
skills on behalf of the designer, in such a way to correctly
point out the redundant features in non-critical areas. As a
matter of fact these features, even if not localised in remarkable zones, require high local mesh rening, because of high
stress concentration. Since the FEM meshing process
involves the whole domain, there is a ``propagation effect''
which starts from a local dense mesh and proceeds throughout the volume, causing a considerable growth in the overall
element number.
On the contrary in a BEM system, only the model
boundary is affected by the meshing process, so it is easier
to decrease the mesh density around the aforesaid features,
in such a way that their simplication (even if advisable) is
no longer strictly necessary.
Moreover the possibility to use discontinuous elements
allows better tting of the mesh elements to the model
features and the capability to independently mesh such
geometry from the rest of the model. Consequently the
presence of small features in the model is not a problem
for a BEM system because a local rened mesh does not
involve ``mesh propagation''.
Despite recent advances in nite element auto-meshing
techniques, it is still quite difcult to obtain a satisfactory 3D
element model for a complex structure. Today the single

32

R. Citarella, S. Gerbino / Journal of Materials Processing Technology 109 (2001) 3037

biggest cost of performing a typical structural analysis task

is the engineering time required for the pre- and postprocessing phase, because run times in this eld are no
longer critical, at least for linear problems. That is why there
is an increasing interest in BE methodology, which is
characterised by very low pre-processing times, effectively
balancing longer run times with respect to FEM [16].
3. CADCAE integration
CAD and BEM systems work extremely well together
because each uses identical geometric entities, such as
points, lines, splines, circles and patches (3D only). This
is particularly relevant in an industrial environment characterised by numerous efforts aimed at the integration of
analysis systems with CAD systems.
Model transfer among different Computer Aided Everything (CAx) systems often involves a lot of problems, related
to the correct model data exchange because differences in
database structure of CAx systems do not allow the direct
data transfer to other engineering systems.
The most used approach for CAD data transfer is the one
based on neutral format. This approach is based on a
common neutral format for exchanging data, and on a pair
of translators (read and write) between each application and
the common neutral format.
Initial graphics exchange specication (IGES) appears to
be the most used method of transferring geometric entities,
although this approach has a drawback. IGES scrambles line
and patch direction, whilst, inside a BEM solver, consistent
entity direction is critical in determining the type of analysis,
nite or semi-innite, to be performed. Therefore the user
must spend time adjusting the data prior to attempting an
analysis [15].
In general, however, moving data among engineering
systems, by IGES or other translators, may cause loss or
alteration of data, independently from CAD user responsibility [17]. For example, accuracy errors may occur, because
intersection curves between non-planar surfaces are
approximated in most solid modellers. Approximations
are required when the precise intersection between two
curved surfaces is too complex to be exactly computed.
The best a programmer can hope to do is to limit the
maximum deviation between the curve approximating the
intersection between two surfaces and the actual intersection
itself. CAD modellers employ different tolerances when
computing the maximum deviation allowed between topological entities such as faces, curves of intersection, and the
vertex where intersection curves meet.
When deviations between faces, edges, and vertex
become too large, processes using solid geometry may fail.
Translation between systems may also fail if the maximum
allowable differences between surfaces and edges in the
receiving system are smaller than those of the sending
system. Thus moving from a system using double precision

data representation, to another using single precision, will,

result in a loss of accuracy.
Preserving surface accuracy in the translation between
different representations, such as Bezier, B-spline or
NURBS, will often require mapping a single surface into
a collection of different surfaces in another representation.
Certain approaches, such as topology mapping, can minimise the impact of this mapping, but the quantity of underlying data will increase too.
With regard to the aforesaid problems the following
remarks are due: it is necessary to set up the same geometric
accuracy in both engineering systems involved in the translation and to choose the proper parameters responsible for a
correct surface/line translation from CAD to IGES format.
In this work, for the IGES model generation, all surfaces and
curves have been converted in B-spline entities (respectively
128 and 126 IGES entities).
4. Problem analysis
A polygonal coupling, which undergoes a static torsional
torque, has been studied with reference to the induced stress
state.
The main part of this work concerns modelling the contact
surface between hub and shaft, because, from a structural
point of view, the interface is the critical area in this kind of
problem.
4.1. Contact analysis with BEM
BEM contact analysis [1821] preliminarily requires the
calculation of interface stresses and displacements, imposing compatibility and equilibrium conditions.
In zero friction contact analysis only normal stresses
are transferred through the interface boundary, in such a
way that compatibility conditions are satised in normal
direction.
In particular, for a conservative analysis the friction action
has been neglected because it reduces the amplitude of the
stress and strain on the contact surface.
To determine the stress and strain behaviour, special
contact elements are used. Contact modelling by means
of BEASY code is implemented through corresponding
pairs of elements along the interface of the two zones in
which the overall domain is subdivided. The pair of corresponding elements, called ``normal gap elements'', even if
coincident in the pre-processing model, can be separated by
a positive (interference) or negative (clearance) gap. By the
sign of the calculated normal stress it is possible to determine the contact state: negative indicates contact, positive
clearance.
Contact problems usually are non-linear because of the
contact area variation with increasing load and eventual
sliding, so the stress state in the interface area depends on
load, geometry, material and contact area extension.

R. Citarella, S. Gerbino / Journal of Materials Processing Technology 109 (2001) 3037

In the present work non-linear behaviour of the model has

been studied by an iterative-incremental procedure. The
incremental step is aimed to follow the contact evolution
while the iterative step allows to obtain an equilibrium
conguration for each load increment. In order to follow
the contact evolution, the load is gradually increased by few
sub-steps and for each of them an equilibrium conguration
is iteratively searched by Newton Raphson algorithm, conditional upon compatibility normal displacement restraints.
The proper choice of the convergence tolerance (1.5% in
this case) and of the maximum number of load sub-steps (15)
and iterations (15), is critical in order to achieve stable results.
To this aim, a convergence study has been carried out,
with mesh rening mainly oriented to the high gradient
contact area.
In this case, it is necessary to rene the mesh in those
areas where the contact state changes, getting high stress
gradients. Actually, as mentioned before, the convergence
study should also involve load step size and normaltangential convergence tolerance values, but these parameters are
not accessible to the users in BEASY code, so default values
have been adopted.
4.2. Problem description
Three different polygonal couplings with three-lobe prole, made by steel, are studied with a parametric CAD 3D
system (Pro/ENGINEER R.17 by PTC), in order to make the
coupling model, whilst the structural analysis is carried out
by a BEM solver (BEASY 7.0).
Fig. 3 shows the polygonal coupling model with parametric characteristic dimensions. Three models (A, B and C)
are characterised by the following common dimensions:
Ds 80 mm, L1 20 mm, L2 40 mm, R 5 mm,
Dm 60 mm, and separately by:

These models will be used for an extensive benchmark of

the alternative FEMBEM procedures.
C-Model is shaped with the optimum prole, characterised by an improved stressstrain behaviour, as obtained
by Mechnik [9] with a shape optimisation analysis.
All the models will be compared each other with reference to some parameters, representative of the elastostatic
behaviour. Material properties are based on isotropy, homogeneity and linear elasticity assumptions.
By correctly imposing cyclic symmetry kinematic conditions on one third of the models it has been possible to
reduce considerably the computational efforts.
One cylindrical shaft end has been supposed to be
clamped and the torque (M t 1000 N m) is introduced
on hub contour by tangential forces. In Fig. 4, the BEM
A-model is shown.
The Young module and Poisson ratio for the steel considered are respectively E 206; 000 N/mm2 and n 0:3.
As already mentioned a BEM code (BEASY) has been
used for the non-linear stress analysis and the convergence is
obtained through an iterative-incremental technique.
In the nal A-, B- and C-models, the following total
numbers of ``reduced'' quadratic elements (eight nodes) are,
respectively, used: 526, 524 and 498.
For the B-model it has been attempted the hub subdivision
in two zones, in order to get enhancement in calculation
times, but without success (it is probably due to the nonlinearity of the problem). That is why, B- and C-models have
different number of elements, even if they are completely
similar.
High mesh grading has been adopted in the interface area
(Fig. 5), where contact between shaft and hub causes very
high stress gradients.
The pre- and post-processing phase are carried out in the
BEASY environment.

 A-model: e 2:25 mm, Dh 155 mm;

 B-model: e 2:25 mm, Dh 120 mm;
 C-model: e 3:50 mm, Dh 120 mm.

Fig. 3. Three-lobe polygonal coupling parametric dimensions.

33

Fig. 4. BE polygonal coupling model.

34

R. Citarella, S. Gerbino / Journal of Materials Processing Technology 109 (2001) 3037

Fig. 5. BE polygonal coupling interface.

4.3. Analysis results

Figs. 6 and 7 respectively show normal stresses and tangential stresses on the hub coupling interface area with reference to
different values of z (axial axis), for the A-model.

Fig. 6. Contact normal stress by BEM.

Fig. 7. Contact tangential stress by BEM.

Fig. 8. Contact normal stress by FEM [9].

The contact surface is the most stressed, in particular with

reference to the hub part for which the results are presented.
It is well evident that the results, obtained through an hp
convergence study, are well in agreement with FEM results
(Figs. 8, 9) from Mechnik [9] even if there is some discrepancy (up to 10% in just a few control points).
Actually, further convergent studies should be done with
regard to reduced values of the ``normal convergence tolerance'' (this parameter has yet to be customised in the
BEASY code), in order to avoid few ``hot spots'' still
existing in the contact area.
In the disconnected area, BEM normal stresses are rigorously zero because BEM does not need to extrapolate
results from the interior towards the surface as in FEM.
With respect to the A-model, in Fig. 10 a comparison is
made between experimental [9] and BEM results for that
concerns the equivalent stress seq calculated with the ``total
strain energy'' criterion, on the external hub surface at
z 40 mm.
Figs. 11 and 12 show, for the same A-model, the comparison between experimental [9] and BEM results for that
concerns the normal displacements at z 0 and z 40 mm,
respectively.

Fig. 9. Contact tangential stress by FEM [9].

R. Citarella, S. Gerbino / Journal of Materials Processing Technology 109 (2001) 3037

Fig. 10. Equivalent stresses for the BEM A-model compared with
experimental ones on the external hub surface at z 40.

Fig. 11. Hub normal displacements on the external surface at z 0.

Fig. 12. Hub normal displacements on the external surface at z 40.

35

Fig. 13. Relative tangential displacements between hub and shaft for
B- and C-models.

One of the criteria adopted in order to decide the critical

state for the coupling is based on the maximum deformation.
In particular, comparing BEM and FEM results, relative
tangential displacements between hub and shaft are presented in Fig. 13 for B- and C-models.
It is well evident that increasing the eccentricity it is
possible to obtain reduced values of coupling deformation,
as reported in more detail in Figs. 14 and 15 for that concern
hub tangential displacements and in Figs. 16 and 17 for hub
normal displacements, with reference to B- and C-model
respectively.
Moreover, there is an evident improvement from the
elastostatic point of view, getting higher Von Mises stress
for B-model (Fig. 18) with respect to C-model (Fig. 19).
In Figs. 20 and 21, deformed congurations are shown for
B- and C-model respectively, in the same deformed scale so
as to be comparable.
For a correct evaluation of a certain degree of discrepancy
still existing between BEM and FEM results, it is necessary

Fig. 14. Hub tangential displacements on the interface contact area for
B-model.

36

R. Citarella, S. Gerbino / Journal of Materials Processing Technology 109 (2001) 3037

Fig. 15. Hub tangential displacements on the interface contact area for
C-model.

Fig. 18. Hub Von Mises stress on the interface area for B-model.

Fig. 16. Hub normal displacements on the interface contact area for the
B-model.

Fig. 19. Hub Von Mises stress on the interface area for C-model.

Fig. 17. Hub normal displacements on the interface contact area for the
C-model.

Fig. 20. Deformed polygonal coupling (B-model).

R. Citarella, S. Gerbino / Journal of Materials Processing Technology 109 (2001) 3037

37

References

Fig. 21. Deformed polygonal coupling (C-model).

to point out that the comparison is made between two

approximated numerical solutions and in some case with
experimental results (affected by a certain degree of approximation) but never with analytical solutions. Moreover
different kind of elements order, linear in FEM analysis
and quadratic in BEM analysis, are used.
5. Conclusion
Nowadays, there is an increasing interest in BE methodology, which is characterised by very low pre-processing
time with respect to FEM, effectively balancing longer run
times (about 6 h for the studied models, with a PC Pentium
Pro, 128 Mb of RAM).
The use of the boundary representation, like in BEM,
integrates well with other Computer Aided Design and
Manufacturing systems, which use a similar description
of the component geometry.
Future perspective of the present work is turned to the
study of sintered steel coupling undergoing fatigue load, and
to 3-D automatic crack propagation, for which BE methodology is well suited.
Acknowledgements
The authors want to express their warmest thanks to
Prof. R. Esposito for his precious help in BE analysis and
to Prof. F. Caputo for his assistance.

[1] P. Orlov, Fundamentals of machine design, Vol. 4, MIR Publishers,

Moscow, 1980, pp. 6973.
[2] R. Musyl, Die Polygon-Verbindungen und ihre Nabenberechnung,
Konstruktion, 14, Heft 6, 1962.
[3] DIN 32711, Antriebselemente Polygonprole P3G, 1979.
[4] DIN 32712, Antriebselemente Polygonprole P4C, 1979.
[5] E. Filemon, Production and Analysis of Polygon Proles, Periodica
Polytechnica M III/1, Budapest, 1959.
[6] A. Leroy, B. Viseur, Sollecitation, Resistance et Deformabilite des
Moyex 'Polygon', La Machine Moderne 61 (S), 1967.
[7] N.N. Manhurim, PolygonProle, Technischer Bericht, Unveroffentlicher Bericht der Fa. Festigkeitsberechnung, Mulhouse, 1953.
[8] F. Caputo, G. Giudice, Proli poligonali per trasmissione di coppie:
Analisi delle tensioni e verica sperimentale, XII Convegno AIAS,
Sorrento, Italy, 2427 September 1984.
[9] R.P. Mechnik, Beitrag zur Festigkeitsberechnung von PolygonWelle-Nabe-Verbindungen unter reiner Torsion, Konstruktion, 43,
1991.
[10] F.G. Kollmann, Chr. Gottlicher, Entwicklung einer verbesserten
Festigkeitsberechnung fur P3G-Polygon-Welle-Nabe-Verbindungen
beikombinierter Biege- und Torsionsbanspruchung, Zwischenber,
DFG-Forsch. vorhaben Ko 634/34-1, TH-Darmastadt, 1992.
[11] W. Beitz, R. Reinholz, Tragfahigkeit von P3G-Welle-Nabe-Verbindungen, Konstruktion, 46, 1994.
[12] G.F. Bocchini, F. Caputo, Componenti sinterizzati per trasmissione di
coppie: accoppiamenti con proli poligonali, XIX Congresso ATA,
Modena, Italy, 57 October 1983.
[13] G.F. Bocchini, Scuola di metallurgia delle polveri, Hoganas Italy,
Milano, Italy, 1995.
[14] F.J. Esper, C.M. Sonsino, Fatigue Design for PM Components,
EPMA, Old Bank Building, Bellstone, UK, 1994.
[15] A. Wanderlingh, Impact of the Boundary Element Method on
engineering quality and productivity, Industrial Application of the
BEM, 1994.
[16] Sugita, T. Itoh, T. Abe, R.D. Honda, Engine component design
system using Boundary Element Method, in: Proceedings of the
Society of Automotive Engineers Conference, Detroit, MI, 1991.
[17] S. Gerbino, S. Crocetta, C. di Martino, Data exchange in CAD
systems: limits, solutions, perspectives, in: Proceedings of the X
ADM Conference, Florence, Italy, 1719 September, 1997.
[18] A. Adey, S.M. Niku, Boundary element stress analysis involving
contact using BEASY, in: Proceedings of the Boundary Element
Conference, Southampton, UK.
[19] F. Cappello, Inuenza di geometria e carico sulla tensione in elementi
di lunghezza diversa in contatto piano, Il Progettista Industriale, N.7,
September 1996.
[20] F. Cappello, Calcolo delle tensioni di contatto con i metodi degli
elementi niti e di contorno, Il Progettista Industriale, N.9,
November 1996.
[21] K.W. Man, M.H. Aliabadi, D.P. Rooke, Analysis of contact friction
using the Boundary Element Method, in: M.H. Aliabadi C.A. Brebbia
(Eds), Computational Methods in Contact Mechanics, Computational
Mechanics Publications, Southampton, Elsevier, London, 1993,
pp.160.