Downloaded from orbit.dtu.

dk on: ,

Optimal key design for shaft hub connections

Pedersen, Niels Leergaard

Published in:
Proceedings of the Institution of Mechanical Engineers, Part C: Journal of Mechanical Engineering Science

Link to article, DOI:

10.1177/09544062231174125

Publication date:
2024

Document Version
Peer reviewed version

Link back to DTU Orbit

Citation (APA):
Pedersen, N. L. (2024). Optimal key design for shaft hub connections. Proceedings of the Institution of
Mechanical Engineers, Part C: Journal of Mechanical Engineering Science, 238(3), 811-821.
https://doi.org/10.1177/09544062231174125

General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright
owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.

 Users may download and print one copy of any publication from the public portal for the purpose of private study or research.
 You may not further distribute the material or use it for any profit-making activity or commercial gain
 You may freely distribute the URL identifying the publication in the public portal

If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately
and investigate your claim.
 Optimal key design for shaft hub connections
Niels Leergaard Pedersen
Dept. Civil and Mechanical Engineering, Solid Mechanics
Technical University of Denmark
Koppels Allé, Building 404, DK-2800 Kgs. Lyngby, Denmark
email: [email protected]

Abstract
Key connections in shaft hub assemblies are typically used due to the price of pro-
duction but also the possibility for using it as an overload protection and safety device is
important. The design of keyways and keys is specified in standards, different designs are
available but typically the contact area is flat. The connection fatigue strength is improved
if the stress concentration is reduced; important here is both the key shape and also the
tolerances between key and keyway. The present designs described in standards are to
some extend the result of previous possibilities for production. Today larger variations are
possible without increasing the production price significantly. New designs with improved
strength are presented and the design evaluation is done using both 2D and 3D finite ele-
ment models including contact modelling. The design is primarily described by the super
elliptical shape, i.e. the design parameterization is simple. Improvement in the stress level
found is significant, a maximum reduction of 78% is found, indicating that the fatigue
strength can be significantly increased.

Key words: Machine elements, key connection, torsion, stress concentration, FEA, 2D/3D.

1 Introduction
Key connections are one of the most simple shaft hub connections with design specifications
available in standards see e.g. [1]. For other standard designs see general machine element
books e.g. [2] or [3]. The points favouring a key connection is; low production cost, easy
assembly/disassembly and the possibility for overload avoidance. The negative points are no
axial fixation, low strength and uneven stress distribution in the contact. The typical solution
to the problem of strength is to increase the key length or the number of keys used. With more
keys there is a tolerance problem related to multiple keys engaging at the same time. In the limit
a spline connection is used. For multiple keys or a spline connection the normal design rule is
that 75% of the teeth are engaged (carry the load).
For all designs in standards it is such that the torsional moment can be transmitted equally
well in reversed direction, i.e. the designs are symmetric. If a connection is to operate only in
one direction we can make an unsymmetrical design that can improve the strength further.
The key connection strength can be evaluated in two ways depending on the load situation.
For non-fatigue loading the strength is related to plastic bearing failure of the shaft and hub
or the key shearing (ultimate strength), the latter is the case when the key connection is used

1
also as an overload safety device. Typically the stress is evaluated as a nominal stress when
compared to material limits. The second case is the fatigue loading of the connection, in this
case the maximum stress, i.e. the stress concentration, is controlling the strength, see [4] and
[5]. In the present paper the focus is on the second case.
As discussed in [6] also with a reference to [7] little efforts have been used for standard key
design modification. A proposal for simple modifications to improve the design is given in [7].
Evaluation of stress concentration can be performed experimentally but in relation to op-
timization the use of numerical methods and Finite Element Analysis (FEA) is the preferred
method applied in the present paper. In a key connection we transfer the torsional moment
through contact between shaft and key and key and hub, for a full stress evaluation we therefore
need to include contact modelling. In a keyway design there is several positions in which the
maximum stress might be found, e.g. in the keyway run-out where typically no contact with
the key takes place. To improve the design we must reduce the maximum stress wherever it is
found. In the present paper we do not investigate the stress state at the keyway run-out. Focus
is on the peak stress in the contact regions. Generally the maximum stress is controlled by two
factors; the geometric design and the size of load applied to the point. In a contact we also have
singularities related to the run-out of contact, see e.g. [8], with the possibility for high stress.
In [7] one simple modification suggested is to apply a circular hole close to the key ends for
avoiding the high stress related to the contact run-out in the length direction of the key.
The loading of the shaft hub connection can in principle include torsion, bending, shear and
tension, in the present paper we will assume pure torsion, see [9] in relation to other loads and
load combinations.
The numerical modelling of the power transmission between shaft and hub involves contact
modelling and the modelling of the load i.e. the transmitted torsional moment. In general it
is not possible to model the transmission of torsional moment fully correct using a 2D contact
model a 3D model is needed. As done in [6] we can model the pure torsion of a cross section
using a 2D model. With a spline connection we have the possibility to apply symmetry to reduce
the 3D modelling. This is however, not the case for a design with a single key and keyway. In
[10] the results reported show that there is a difference in the maximum stress found if the
torsion is performed with or without the key. The general findings is that the stress elevation
is restricted to be a maximum of 24%. This indicate that a 2D modelling using pure torsion is
sufficient. In the present paper it is shown that the difference in maximum stress found with or
without the key and contact modelling is significant. Similarities in the stress variation found in
full 3D contact modelling and a simplified 2D contact modelling show however that a simplified
2D contact model could be used for optimization purpose.
The standardized key and keyways are geometrically defined by straight lines and circular
arches, which from a point of low stress are not optimal shapes. The optimization in the present
paper is performed in order to reduce the stress concentration factor. The stress reduction is
achieved through shape optimization where the shape to design is also where we have the con-
tact. A simple analytical shape representation is applied using the super ellipse. This facilitates
an optimization problem with a low number of design parameters (less than 13). With the small
number of design parameters a simple parameter study to minimize the stress is possible. The
simple shape representation and the number of design variables also enables a simple result
communication making the designs in this regard more practical. With more design variables it
will typically be possible to optimize the designs further. We know that the optimal shape is the
one that result in a constant stress along the designed shape. If the optimization results in a con-
stant stress the possible further improvement by a more involved parameterization can therefore
be evaluated. As discussed in [11] we cannot use the FE nodes position as the design param-

2
eters. With stress minimization as the optimization objective it is also important that the FE
stress level is converged which can be checked by a mesh refinement. With an analytical shape
description that is not linked to the FE discretization this mesh refinement can be performed.
The paper is organized as follows. In Section 2 the super elliptic parameterization is pre-
sented mathematically, this design parameterization is used for the cross sectional key and key-
way design. Section 3 describes the numerical modelling of torsion and the applied assump-
tions. The comparison of maximum stress found in the full 3D contact model and the simple
2D contact model is also discussed. The optimization using symmetric key design in presented
in Section 4 where also a comparison to previous found optimized designs is made. In Section
5 the results assuming unsymmetrical design are presented. Conclusions are given in Section 6.

2 Super elliptic parameterization

One possibility for shape parameterization of the key is to use the super ellipse, see e.g. [12]
and [13]. The initial shape has only three design parameters, principle axes A and B and super
elliptical power η. The parametric form is

π
X =A cos(t)(2/η) , t ∈ [0 : ] (1)
2
π
Y =B sin(t)(2/η) , t ∈ [0 : ] (2)
2
Using this parameterization we can connect the two points (marked by a cross) in Figure
1a) and the shape will lie within the design domain indicated in grey. For η > 2 the curvature
is zero at points A and B and for η > 1 the curve slope at points A and B follows the design
domain boundaries.
In [14] the super ellipse is also used in a distorted form, with the parametric form

B π
X = A cos(t)(2/η) (1 − tan(β) sin(t)(2/η) ), t ∈ [0 : ] (3)
A 2
(2/η) π
Y = B sin(t) , t ∈ [0 : ] (4)
2
With this parameterization the design domain is changed to the one shown in Figure 1b).

Y Y β Y γ β

B B B

X X X
a) b) c)
A A A

Figure 1: a) Design domain for super ellipse. b) Distorted super ellipse. c) Double distorted super ellipse.

A double distortion of the super ellipse as indicated by the design domain in Figure 1c)
where two rotation angles are used is applied in [15]. The analytical form as given in [15] is

3
 π
X = A − B sin(t)(2/η) tan(β) cos(t)(2/η) , t ∈ [0 : ]


(5)
2

Y = B + cos(t)(2/η) (A − B sin(t)(2/η) tan(β)) tan(γ) · (6)
π
sin(t)(2/η) , t ∈ [0 : ]
2
With the parameterization in (5) and (6) we have a simple parameterization with two ro-
tations but the shape might, depending on the sign of the angles γ and β violate the design
domain.
With a slight modification (mapping) from Figure 1a) to 1c) we have the following analytical
form instead

X =A cos(t)(2/η) + cos(t)(2/η) sin(t)(2/η) · (7)

 
A − B tan(β) π
− A , t ∈ [0 : ]
tan(γ) tan(β) + 1 2
Y =B sin(t)(2/η) + cos(t)(2/η) sin(t)(2/η) · (8)
 
B + A tan(γ) π
− B , t ∈ [0 : ]
tan(γ) tan(β) + 1 2
The parametric shape as described in (7) and (8) lies within the design domain 1c) and for
η > 2 the curve curvature is zero at points A and B and for η > 1 the curve slope at points A
and B follows the design domain boundaries.
In Figure 2 different curves are shown to illustrate this.

η=6
η=3
η=2
2 η=1
η = 0.5
Domain
boundary

0
0 3

Figure 2: Examples of distorted super ellipse. The design domain is specified by A = 3, B = 2, β = π/6 and
γ = π/5.

We now have the possibility to model each quarter of the key cross section in the axial
direction by this distorted super ellipse. Applying this to the complete key cross-section and
if we seek to have continuity also in slope and curvature we can make a model with 13 design
parameters. An example of this is illustrated in Figure 3. If we seek a symmetric design this
can easily be achieved with a reduced number of design parameters as the result. The 13 design
parameters are shown in the figure. Please note that B1 and B2 are the principle axes for
the distorted super ellipse in 1 quadrant and 3 quadrant respectively. The principle axes (Y
direction) is then b − B2 and b − B1 for the distorted super ellipse in 2 quadrant and 4 quadrant
respectively.

4
 α2
η2
η1
B1
A2
α3 α1
b
A1
B2

η3
η4
α4

Figure 3: Example of key cross section described by 4 distorted super ellipse, b = 28mm, A1 = A2 = 8mm,
B1 = B2 = 14mm, α1 = α3 = α4 = 0.2, α2 = −0.1, and the power η1 = 3, η2 = 4, η3 = 2 and η4 = 3.5.

3 Numerical modelling of torsion 2D and 3D

Modelling pure torsion in 2D can be achieved, if we neglect contact modelling, using the warp-
ing function Ψ introduced by Saint-Venant, also described in e.g. [6]. Assuming a Cartesian
coordinate system with the z-axis aligned with the axial direction the displacements are

φ φ φ
vx = −yz , vy = xz , vz = Ψ(x, y) (9)
l l l
resulting from the torsional moment

φ
T = GJ (10)
l
where G is shear modulus of elasticity, J is the cross sectional torsional stiffness factor, φ cross
sectional angular rotation and l the shaft length. The cross sectional stress components are then

Ψ φ Ψ φ
τxz = τzx = ( − y)G . τxy = τyx = ( + x)G (11)
dx l dy l
The force equilibrium results in the Laplace differential equation assuming no volume force

∆Ψ = 0 (12)

With the correct boundary conditions we can solve this equation numerical to find the displace-
ment vz , and the shear stress on the cross section. The stress is evaluated by the theoretical
shear stress concentration factor

τmax
Kts = (13)
τnom

5
where maximum shear stress τmax and nominal shear stress τnom are given by

q
τmax =( 2 + τ2 )
τzx (14)
zy max

16T
τnom = (15)
πd3
In Figure 4 the stress in a shaft, with diameter d = 100mm and using a keyway as specified
in DIN 6885, is shown. The stress concentration is Kts = 2.93 when using the maximum fillet
as specified by the standard.

q
2 + τ 2 ) in shaft scaled by the nominal stress. The keyway design is according to DIN
Figure 4: Shear stress ( τzx zy
6885 for a d = 100mm shaft with the maximum fillet radius of r = 0.6mm.

For a hub or tube with an internal keyway corresponding to the design in Figure 4 and shown
in Figure 5 the maximum stress in the fillet depends on the hub outer diameter. Selecting the
stress at the internal diameter of the tube without the keyway as the nominal stress, the stress
concentration for the hub is Kts ≈ 3.9 for outer diameter 3 times greater than internal diameter,
as seen in Figure 5. The real stress level is however smaller than in the shaft for transmission of
the same torsional moment size due to the larger cross sectional torsional stiffness factor.
Compared to the 2D modelling of pure torsion a full 3D modelling that includes contact
modelling is much more involved. The number of elements needed in the Finite Element (FE)
mesh is increased considerably for a reasonable stress level convergence. In Figure 6 the stress
is shown as a stress concentration factor defined as

vM
σmax
KvM = vM
(16)
σnom
vM
where σmax is the maximum von Mises stress and the nominal von Mises stress is defined as

vM
√ 16T
σnom = 3 3 (17)
πd

6
 a)

b)

q
2 + τ 2 ) in hub scaled by the nominal stress defined as the shear stress on the internal
Figure 5: Shear stress ( τzx zy
boundary if there were no keyway. The keyway design is according to DIN 6885 for an internal diameter d =
100mm outer diameter is do = 300mm. a) Full design. b) Zoom of fillet.

The applied boundary conditions for the FE model is a symmetry condition on the hub and
shaft center plan (in the axial direction), further the hub is clamped on the outer surface. The
load/deformation is applied by fixing the outer end of the hub and forcing an axial rotation that
leads to a total torsional moment of T = 200Nm on the hub (i.e. 100Nm on the half model
shown). In Figure 6 we have used the minimum key length, 80mm, as specified by DIN 6885.
We see that in this case we find a stress concentration that is much higher than the one found
for the 2D case of pure (individual) shaft and hub torsion, here a value of KvM = 22.3 is found.
The contact modelling includes a static Coulomb friction, here we use a friction coefficient of
µ = 0.1. The contact is modelled using the penalty method and the computations are done in
COMSOL [16].
In the FE models we use for 3D models tetrahedral elements and for 2D models triangular
elements, in both cases with a quadratic displacement assumption. The stress convergence for
the model in Figure 6 is shown in Figure 7, the stress is shown as a function of N (the number of
degrees of freedom in the FE model). The mesh with N = 2.7 · 105 is shown in Figure 8a) and

7
 a)

b)

Figure 6: Von Mises stress in hub and shaft (key is hidden) scaled by the nominal stress. The keyway design is
according to DIN 6885 for an internal diameter d = 100mm outer diameter is do = 200mm and length of key is
80mm. a) Full FE design. b) Zoom of fillet.

the mesh with N = 20.4 · 106 is shown in Figure 8b) and c). The change in the maximum stress
when the discretization is increased from N = 4.4 · 106 to N = 20.4 · 106 is approx. 1% i.e. we
consider the stress to be converged for the last model with the stress level being KvM = 22.3.
As seen the model must be highly refined. The highest stress is found in the shaft in the
fillet in the keyway bottom. The stress along the keyway in the center plan is shown in Figure
9. The stress is shown from the contacting side towards the side without contact, i.e. both the
sides and the bottom of the keyway is shown. It is clear due to the stress peak that there is a
large potential for reducing the maximum stress.
For optimization the full size FE model as presented is not feasible to use due to the large
number of degrees of freedom and the associated computational time. However it is clear that
the modelling must include contact modelling.
One possibility for reducing the computer time is to reduce the number of elements. One
way to do this is to reduce the hub width, in Figure 10 the stress along the keyway in the center
plan is shown for different hub width, the minimum element size is kept fixed. In the analysis
the rotation angle is kept fixed meaning that the transferred torsional moment is reduced with
reduced hub width. In Figure 10 the stress is scaled for all cases such that the maximum stress is
unity. We see that that for the contacting part (approximately the first quarter of the arc length)
the variation in shape is limited if the width is reduced to W = 20mm, and for all cases the
peak stress shape is identical. This indicated that an optimization could be performed on the

8
 KV M

22

18

14

N/106
10
0 5 10 15 20

Figure 7: Maximum Von Mises stress scaled by the nominal stress as a function of degree of freedom in the FE
model. For the keyway design seen in Figure 6.

model with the reduced hub width.

One further possibility for reducing the computational time is to increase the element size.
It is still important that the stress variation is described sufficiently accurate. In Figure 11 the
stress variation for a reduced model is compared to the highly refined model. The computational
time is reduced here with a factor of 50. The reduced model cannot estimate the maximum stress
with sufficient accuracy, however the reduced model describes the stress variation sufficiently
accurate for the optimization purpose.
A 2D model including contact is also possible. For a 2D model the problem is how the
boundary conditions should be applied. For the result in Figure 12 we have fixed the hub
outer rim and a forced rotation is applied to the shaft at a minor circle with a diameter d/10.
The rotation size is made such that the transferred torsional moment is T = 200Nm. In the
modelling an assumption of plain strain is applied. Due to the model being 2D we can increase
the number of elements along the contacting boundaries.
In comparing the results from the two models we see that that the overall stress distribution
is reasonably described by the 2D model when comparing with the full model. It is noted that
the stress peak for the 2D model is smaller in value and also there is a variation in the stress
over the circular fillet arc in the keyway corner.
For the optimization in the next sections the 2D contact model is used. All examples are
with a shaft diameter of 100mm and the transferred torsional moment is 200Nm. The final
result is then verified by the full 3D modelling. In all the modelling we assume a prefect initial
fit between the three parts. The reason being that this makes a direct comparison between the
different designs possible. This point was also discussed in [13].

4 Optimized symmetric key design

In [6] different optimized designs were shown. The optimization was performed without the
key and contact modelling, i.e. it was modelled as pure torsion using the Saint-Venant warping
function. Three different keyway/key designs were presented depending on the level of allow-
able design changes. The designs are all symmetric, and the only change relative to the standard
is that the circular arc is substituted by a super ellipse, as seen in the key quarter model in Figure

9
 a)

b)

c)

Figure 8: Illustration of mesh for FE model in Figure 6 (key is hidden). a) Mesh with N = 2.7 · 105 b) Mesh with
N = 20.4 · 106 . c) Zoom of fillet for mesh with N = 20.4 · 106 .

13.
The three optimized results are repeated here but now with the 2D contact modelling for
comparison. The design variables (constraint/unconstraint) for the three designs are
• Design I, Constraint variables: b = 28mm, h = 16mm, L1 = 7.4mm. Design variables
L2 = 13.19mm, η = 1.63
• Design II, Constraint variables: b = 28mm, L1 = 7.4mm. Design variables h =
23.02mm, L2 = 0mm, η = 1.99
• Design III, b = 28mm, h = 16mm. Design variables L1 = 4.56mm, L1 = 0mm,
η = 2.22
For Design I the constraint values are in accordance with the DIN 6885-1 standard. In
Design II a larger key is allowed while Design III as Design I fits within the standard key area.
The result from the 2D contact simulation is shown in Figure 14.

10
 KvM

20

10

s/mm
0
0 10 20 30 40

Figure 9: Stress along shaft keyway at the center plan. The stress is shown as von Mises stress scaled by the
nominal stress.

W = 80mm
W = 40mm
0.8
W = 20mm
W = 10mm
W = 5mm
0.4

s/mm
0
0 10 20 30 40

Figure 10: Illustration of stress distribution along shaft keyway at the center plan depending on the hub width.
The stress for the different length are all scaled so that the maximum stress is unity.

W = 80mm
W = 20mm
0.8 W = 20mm, less elements

0.4

s/mm
0
0 10 20 30 40

Figure 11: Illustration of stress distribution along shaft keyway at the center plan. The stress for the different
models are all scaled so that the maximum stress is unity.

11
PSfrag
KvM

20 W = 80mm
2D model

10

s/mm
0
0 10 20 30 40

Figure 12: Stress along shaft keyway at the center plan for the full 3D model and for the 2D model. The stress is
shown as von Mises stress scaled by the nominal stress.

L1 A

b/2

L2

h/2

Figure 13: Key/keyway quarter design.

The stress concentrations are in all cases not identical to the reported results from pure
torsion. The relative design improvements when compared to the DIN standard is however still
significant for Designs II and III. Design II has the largest improvement with a stress reduction
of 63% while design III has a reduction of 41%.
The cross sectional key area is important for the possible improvement in the stress level,
with a larger area a larger reduction in the maximum stress can be found. This can also be ob-
served in comparing Design II and III. In the remaining part of the paper the key cross sectional
area is constrained to be smaller than the original DIN design.
Performing the parametric optimization instead, directly using the 2D contact model and
the design variable as presented in Figure 3 we obtain significant reductions in the stress con-
centration. We first enforce double symmetry i.e. the only design variable is η1 . The optimized
result is η1 = 3.1 and stress distribution is shown in Figure 15.
The stress concentration is here KvM = 5.9 which is a reduction of 67% compared to the
DIN design. We notice that the maximum stress is now found in both the shaft and hub.
We also have the possibility for a single symmetric design with the two design variables
η1 = η4 and η2 = η3 (see Figure 3). The optimization result is η1 = 2.4 and η2 = 5.6, the stress
distribution is shown in Figure 16.

12
 KvM
20
DIN
Design I
Design II
Design III
10

s/mm
a) 00 10 20 30 40
KvM

DIN
Design I
Design II
8
Design III

s/mm
b) 00 10 20 30 40

Figure 14: Stress along the keyway. The stress is shown as von Mises stress scaled by the nominal stress. a) Stress
in shaft keyway for the four designs. b) stress in hub keyway for the four designs.

KvM
6
Shaft
Hub

s/mm
0
0 10 20 30

Figure 15: Optimized double symmetric design. Stress along shaft and hub keyway. The stress is shown as von
Mises stress scaled by the nominal stress.

With this design we have a further reduction in the stress level, KvM = 5.0, which is a

13
PSfrag
KvM

5 Shaft
Hub

2.5

s/mm
0
0 10 20 30 40

Figure 16: Optimized single symmetric design. Stress along shaft and hub keyway. The stress is shown as von
Mises stress scaled by the nominal stress.

reduction of 72% compared to the DIN design.

With the two symmetric designs the torsional moment can still be applied in both direction
equally well. If we allow for unsymmetrical designs the stress can be reduced even further.

5 Optimized unsymmetrical key design

If the torsional moment in the shaft hub connection is only applied in one direction we can
make an unsymmetrical design. The key shape is in the present paper controlled by the de-
sign parameterization shown in Figure 3. This parameterization has 13 design parameters. In
order for the design to be feasible for production but also for application we constrain two of
the angles α2 ≥ 0 and α4 ≥ 0. From numerical experiments it is clear that the important
boundary to design is where the contact is applied to both the shaft and hub. Assuming that the
torsional moment is applied in clockwise direction (on the shaft) it is the design of the key in
the quadrant 2 and 4 that is important (see Figure 3). For this reason the following values are
selected/constraint

b = 28mm, B1 = B2 = 26mm η1 = η3 = 2

In the two quadrants where we do not have contact the design of the keyway is simplified to
a straight line instead of the super ellipse. The straight lines are extended to the surface of the
hub and shaft respectively. This leaves 8 design parameters for the optimization.
With the further reduction in the stress level the stress increases at the run-out of contact
between the key and hub becomes significant. The reason is that the corner is too stiff, the same
point was addressed in [8]. One possibility to overcome this problem is to reduce the stiffness
by removing material. The shape and size of this stiffness relief/material removal is not highly
important as long as this does not give rise to a higher stress than the one found in the contact
between key, hub and shaft. The simple design shown in Figure 17 is selected.
Using this design for the stiffness relief and the selected 8 design parameters the optimized
solution is

14
 0.25rad

3.5mm
Hub 12mm

Key
Shaft

Figure 17: Design of stiffness relief.

α1 = 0.57rad, α2 = 0,
α3 = 0.57rad, α4 = 0.26rad
η2 = 2, η4 = 2.2,
A1 = (1 + 5 tan(α1 ))mm, A2 = (1 + 2 tan(α3 ))mm

The unsymmetrical key and keyway design and the stress level is shown in Figure 18.

Figure 18: Design of optimized unsymmetrical key. The scale indicated von Mises stress scaled by the nominal
stress.

The stress along the keyway in the shaft and hub together with the stress along the stiffness
relief is shown in Figure 19
As shown in Figure 19 the design is made such that we have almost equal maximum stress
at both hub and shaft. We also see that the stress is constant over a larger part of the contact
length. The maximum stress in the relief grove is kept at approximately the same level. The
maximum stress concentration is for the shown design KvM = 3.80 which is a reduction of
78% compared to the DIN design. The stress variation indicates that a further reduction in the

15
 KvM
4
Shaft
Hub
Relief

s/mm
0
0 10 20 30

Figure 19: Optimized unsymmetrical design. Stress along shaft and hub keyway and also along the stiffness relief.
The stress is shown as von Mises stress scaled by the nominal stress.

stress level might be possible with a more involved parameterization. The design is not highly
sensitive to variation in the design variables and when comparing to the best symmetric design
it can be discussed if this further reduction in stress level is needed.
For verification the optimized design is used in a 3D simulation of the same design. It is
clear that the result as shown in Figure 20 and Figure 21 is not identical to the one found in the
2D modelling but overall the results are confirmed. In the 3D simulation there is also the run-out
of contact in the axial direction that leads to an increased stress level. Here the maximum stress
concentration found is KvM = 7.08, whereas in the center plane the value is KvM = 4.88. The
reduction of the stress level due to the run-out in axial direction is not investigated further in the
present paper.
The stress level found in the center plane still corresponds to a stress reduction of 78%,
which confirms the findings from 2D and illustrates that it is possible to use the 2D model.
Doing the optimization directly on the 3D model is infeasible, the shown model has 14 · 106
degrees of freedom, with a computational time on a regular PC of 8h.
The stress level is throughout the paper found using numerical FE calculation, no experi-
mental verification is made. However the same level of mesh refinement is used for the different
designs so the relative change in stress level for the optimized design when compared to the
standard design is considered to be valid.

6 Conclusion
In the present paper we use shape optimization to improve the strength of key connections
between shaft and hub. The numerical modelling of applying torsional moment is investigated.
The necessary modelling refinement is important for the subsequent optimization procedure.
Including contact modelling is shown to be necessary for the fatigue strength evaluation. A full
3D contact modelling is possible but a 2D contact modelling, with a much higher computational
efficiency, is sufficient for design optimization.
Previous designs optimized using an assumption of pure torsion is re-evaluated using contact
modelling. The evaluation shows that the proposed designs still exhibits a large improvement
in the maximum stress level relative to the standard DIN design.

16
a) b)
Figure 20: 3D model of optimized non-symmetrical key. The Von Mises stress is scaled by the nominal stress. a)
Full FE design. b) Zoom of key.

KvM
5
Shaft
Hub
Relief

2.5

s/mm
0
0 10 20 30

Figure 21: 3D results for optimized unsymmetrical design. Stress along shaft and hub keyway and also along the
stiffness relief at the center plan. The stress is shown as von Mises stress scaled by the nominal stress.

Parameterization of the key shape can be performed in many ways, in the present paper a
design parameterization based on the super ellipse is presented. Using the proposed parameter-
ization and the 2D contact modelling it is found that the maximum stress level can be reduced
even further as compared to the previous proposed designs. The new key design is constrained
to have a cross sectional area not greater than the original DIN design, the reason for this is to
make a direct comparison possible. If the area is not fixed, even larger stress reductions can
be achieved. Prescribing the design to be symmetric the maximum stress is reduced by 72%.
Allowing for an unsymmetrical design the reduction in maximum stress is 78%.
It can be argued that the improvement achieved by the symmetric design and the associated
simplicity might make this the more practical design.

Acknowledgment
For discussions and suggestions I wish to thank Prof. Peder Klit.

17
References
[1] DIN 6885-1. Paßfedern nuten (in German), 1968.

[2] Norton RL. Machine design: An integrated Approach. Pearson, 2020.

[3] Budynas R and Nisbett K. Shigley’s Mechanical Engineering Design. McGraw Hill, 2020.

[4] Pilkey WD, Pilkey DF and Zhumning B. Peterson’s stress concentration factors, 4th
Edition. John Wiley & Sons, inc., 2020. 640 pages.

[5] Peterson RE. Fatigue of shafts having keyways. American Society for Testing Materials -
Proceedings 1932; 32(part 2): 413–419.

[6] Pedersen NL. Stress concentrations in keyways and optimization of keyway design. Jour-
nal of Strain Analysis for Engineering Design 2010; 45(8): 593–604.

[7] Orthwein WC. A new key and keyway design. J Mech Des Trans ASME 1979; 101(2):
338–341.

[8] Pedersen NL. On optimization of interference fit assembly. Structural and Multidisci-
plinary Optimization 2016; 54(2): 349–359.

[9] Fessler H, Rogers CC and Stanley P. Stresses at end-milled keyways in plain shafts sub-
jected to tension, bending, and torsion. Journal of Strain Analysis 1969; 4(3): 180–189.

[10] Okubo H, Hosono K and Sakaki K. The stress concentration in keyways when torque is
transmitted through keys. Experimental Mechanics 1968; 8(8): 375–380.

[11] Ding Y. Shape optimization of structures: A literature survey. Computers and Structures
1986; 24(6): 985–1004.

[12] Pedersen NL. Stress concentration and optimal design of pinned connections.
Journal of Strain Analysis for Engineering Design 2019; 54(2): 95–104. DOI:
10.1177/0309324719842766.

[13] Pedersen NL. On optimal stress for shaft-hub connections (polygon connections).
Journal of Strain Analysis for Engineering Design 2021; 56(4): 195–205. DOI:
10.1177/0309324720969530.

[14] Pedersen NL. Improving bending stress in spur gears using asymmetric gears and shape
optimization. Mechanism and Machine Theory 2010; 45(11): 1707–1720.

[15] Pedersen NL. Minimizing tooth bending stress in spur gears with simplified shapes of
fillet and tool shape determination. Engineering optimization 2015; 47(6): 805–824.

[16] COMSOL AB. www.comsol.com: Stockholm, Sweden, 1986 -.

18
Appendix
Notation
b Width of key (Design variable)
d Diameter of shaft
h Height of key (Design variable)
l Lenght of shaft
vx Displacement in x direction
vy Displacement in y direction
vz Displacement in z direction
A Super elliptic principle axes (Design variable)
A1 Super elliptic principle axes (Design variable)
A2 Super elliptic principle axes (Design variable)
B Super elliptic principle axes (Design variable)
B1 Super elliptic principle axes (Design variable)
B2 Super elliptic principle axes (Design variable)
G Shear modulus
J Cross sectional torsional stiffness factor
Kts Theoretical shear concentration factor
KvM Theoretical von Mises stress concentration factor
L1 Length (design parameter)
L2 Length (design parameter)
T Torsional moment
W Width of hub
α1 Angle (Design variable)
α2 Angle (Design variable)
α3 Angle (Design variable)
α4 Angle (Design variable)
β Angle
η Super elliptic power (Design variable)
η1 Super elliptic power (Design variable)
η2 Super elliptic power (Design variable)
η3 Super elliptic power (Design variable)
η4 Super elliptic power (Design variable)
γ Angle
φ Cross sectional angular rotation
µ Coefficient of static friction
σ Normal stress
τ Shear stress
Ψ Warping function

19