Sunteți pe pagina 1din 8

2nd International Conference on Engineering Optimization

September 6-9, 2010, Lisbon, Portugal

Optimization of Keyway Design


Niels L. Pedersen
Dept. of Mechanical Engineering, Solid Mechanics, Technical University of Denmark Nils Koppels All e, Building 404, DK-2800 Kgs. Lyngby, Denmark, email: nlp@mek.dtu.dk

Abstract Keys and Keyways are one of the most common shaft hub connections. Despite this fact very little numerical analysis have been reported. The design is often regulated by standards that are almost half a century old where most results reported in the literature are based on experimental photoelastic analysis. The present paper shows how numerical nite element (FE) analysis can improve the prediction of stress concentration in the keyway. Using shape optimization and the simple super elliptical shape it is shown that the fatigue life of a keyway can be greatly improved. The design changes are simple and therefore practical realizable with only two design parameters active. Keywords: Keyway, Parallel key, Stress concentration, Optimization 1. Introduction Keys and keyways commonly perform connection of shaft and hubs. The designs of these are controlled by different standards, e.g. [1]. Different design principles are possible, these include parallel keys, tapered keys or Woodruff keys, see e.g. [2, 3]. Among these, the most common is the parallel key that is the subject of the present paper. The key and keyway design is fully controlled by the standards based on only one parameter, the shaft diameter. It is remarkable that very little effort have been done to improve the design with respect to fatigue, i.e. by minimizing the stress concentrations. This was already pointed out in [4] and to the authors knowledge very little have been done since. Other designs are possible and have been proposed in the literature, e.g. [5, 6]. The rst paper addressing the torsional stiffness of shafts with a kind of keyway is probably [7]. In this paper, the shafts were modeled with elliptical cross section and the keyways were modeled as hyperbolas. Following this analytical paper, there have been a number of experimental papers dealing with the stress concentrations of key and keyway connections. Many of these papers have used photoelastic analysis, see e.g. [8, 9, 10, 11, 12, 13]. Other papers have used electroplating of copper to the surface , see e.g. [14, 15]. Between the two paper [7] and [8] other papers have been dealing with experimental stress concentration verication, see the references in [4]. The most commonly used reference with respect to stress concentration factors is [16] which is reproduced and extended in [17]. The keyway results reported here are taken from the papers [8, 9, 15]. The use of FE modeling and computational power makes it possible to improve these results, but it seems that this has not yet been done. The purpose of the present paper is therefore to improve/optimize the keyway design by lowering the stress concentration. The keyway related stress is indeed fully 3 dimensional as also stated in [18]. A number of different factors will have an inuence on the needed FE analysis complexity and on the resulting maximum stresses found by the analyses. 1. Loading: tension, bending or torsion 2. Key: loaded with or without the key inserted in the keyway 3. Stress: at the keyway end or in the prismatic part Restricting the numerical analysis the present paper deals only with torsion; with respect to the other loads or any load combinations the reader is referred to [9]. To make easy comparison to the numerical and experimental work in [8] possible, the keyway is loaded in torsion without the key. This means that there is no need for contact analysis, which would complicate the numerical analysis considerably. The reported results in [15] state that there is a difference in the maximum stress for pure torsional loading without the key relative to torsion applied through the key. The experiments presented in [15] where grouped in two (group A and B) for different relative size of keyway to shaft diameter. The reported experimental result is that in the prismatic keyway part the maximum stress is 8 12% for group A and 4 7% for group B greater with a key relative to no key, while the difference is 16 24% for group A and 12 14% for group B at the key end. These values were rather unaffected by different ratios of llet radius to shaft diameter. This leads to the conclusion that the true stress concentrations can be found 1

from a study without the torsion coming from the key by adding at maximum 12% to the stresses in the prismatic part. The end of a keyway has two standard designs, shown in Figure 1.

a) PSfrag replacements

     

     

                                    

b)

Figure 1: The two standard keyway ends for parallel keys. a) End-milled or prole keyway b) Sled-runner keyway. The prole keyway is cut by an end-mile while the sled-runner keyway is cut by an ordinary milling cutter. The stress concentrations at the keyway end are most severe for the prole keyway so with respect to fatigue the sled-runner is the best design. [4] suggested a design change to the sled-runner keyway end that further improves the fatigue properties. The stress concentration factor for pure torsion for a prole keyway end was in [8] found to be Kt = 3.4 for a width of keyway to diameter ratio equal to b/d = 1/4. This value was unaffected by the keyway bottom llet radius. If we are to improve the prole keyway end design we should move away from the circular design, this would most probably increase the machining cost and is not discussed further in this paper. For the sled-runner keyway in pure torsion the stress concentration factor is higher in the keyways prismatic part relative to the keyway end if the same milling cutter is used for the hole cutting operation. With the simplication made the analyzed stress concentration factor is in the present paper fully controlled by the keyway llet in the bottom of the prismatic part. The design domain is two dimensional and shown in Figure 2.

PSfrag replacements b

                                                                 y                                                                                                           x                                                                      d                                                                                        t                         

Figure 2: Cross section of prismatic part of parallel keyway, the coordinate system is placed at the shaft axis. The relative dimensions corresponds to a d = 100mm shaft according to [1], (t = 10mm, b = 28mm, 0.4mm r 0.6mm). Obeying the standards the only possibility to improve the stress concentrations for the design in Figure 2 is to select the maximum llet radius r. Previous work on shape optimization in relation to machine elements, see [19, 20], has shown that changing from the circular shape to an elliptical shape has a large inuence on the stress concentrations. This is also pursued in the present paper.

2. Mathematical formulation and FE The torsional moment is given by l (1)

Mt = GJ

where G is the shear modulus of elasticity, J is the cross sectional torsional stiffness factor, angular rotation of torsional cross section and l it the shaft length. In the literature it is common to use = /l, i.e. an angular rotation per length. It is assumed that a prismatic shaft is align with a Cartesian coordinate system with the x-, y and z -directions such that the shaft axis is aligned with the z -direction. Saint-Venant have introduced the warping function (x, y ) by which the shaft displacement under torsion is given by l l l (2)

vx = yz

vy = xz

vz = (x, y )

Using this denition the cross section shear stresses (all other stresses are zero) are given by d y )G dx l d + x)G dy l (3)

zx = xz = (

zy = yz = (

With zero volume force the force equilibrium gives the Laplace differential equation that the warping function must fulll = 0 (4)

To solve this differential equation the boundary conditions are needed. For free boundaries we have no surface traction. If we dene the normal to the surface as {nx , ny }T then the condition of no surface traction is given by zx zy (5)

{nx , ny }

=0

This can be reformulated into a Neumann boundary condition for the warping function by using (3) d dx

{nx , ny }

It is possible to utilize symmetry, see Figure 3 where half the cross section of a shaft is shown. The boundary condition for a symmetry line is given by zx zy (7)

d dy

= {nx , ny }

y x

(6)

{ny , nx }

=0

If the symmetry line fulll that y = 0, as done in Figure 3, then the boundary condition for the symmetry line (7) can be simplied. Since nx = 0 and ny = 1 then the boundary condition become zx = 0 or by using (3) that d/dx = 0. This is identical to the Dirichlet boundary condition =C (8)

where C is an arbitrary constant. Since we are only interested in the rst derivative of the warping function we may select C = 0. By formulating the torsional problem as (4) with the boundary conditions (6) and (8) it is possible to use a standard PDE solver. In the present paper the program COMSOL is used ([21]). It should be noted that the displacements (2) are all dened relative to a coordinate system placed at the center of torsion. The calculation of the involved strains and stresses are however insensitive to any movement or rotation 3

of the coordinate system. 2.1 Stress concentration The stress concentration is most often dened as Kt = max nom (9)

where nom is the nominal stress, i.e. the maximum stress without the keyway and max is the maximum stress with the keyway. Both stresses are the greatest principal stress. The subscript t indicates that it is a theoretical stress concentration based only on geometry and loading/boundary condition, no material sensitivity is included. For torsional problems we may give the stress concentration as Kts = where we for the present torsional problem have that 16Mt d3
2 + 2 ) zx zy max

max nom

(10)

nom =

(11) (12)

max = (

The nominal stress and the maximum stress are found under the same external loading. By the assumption of linear elasticity the external load size does not inuence the stress concentration. The size of M t is selected such that = 1 N/m3 l (13)

this leads to the nominal stress and maximum stress given as d N/m3 2 Jc Jk ( d d y )2 + ( + x)2 dx dy N/m3
max

nom = max =

(14) (15)

where Jc is the cross sectional torsional stiffness factor for the circular shaft and Jk is the cross sectional torsional stiffness factor for the shaft with a keyway. d4 32 ((
A

Jc = Jk =

(16) d d y )y + ( + x)x)dA dx dy (17)

2.2 FE model A FE model example is shown in Figure 3. The shaft design is the DIN standard presented in Figure 2. Only half the shaft is necessary for the modeling. The bottom edge is a symmetry line so here the Dirichlet boundary condition (8) is applied. To the remaining edges the Neumann boundary condition (6) is applied. The number of elements in the shown mesh is limited (917 elements) for illustrative purpose. The numerical calculations performed in this paper have all been performed with a much higher number of elements (30000 to 60000). Convergence test have been made to conrm the FE results. The maximum stress is of primary interest, since this stress controls the stress concentration. The maximum stress is in all numerical calculations found at the keyway boundary. In Figure 4 the stress concentration is shown along the keyway boundary (s is the arc length), in the close up Figure 4b the stress concentration along the llet is shown. From an optimization point of view it is clear that this is not optimal because the stress is expected to be 4

Center of keyway External point

PSfrag replacements

0 5 PSfrag replacements PSfrag replacements 10 Center of keyway 15 External 20 point a) b) 0.0 0.5a) Example of a nite element mesh, the shown Figure 3: The gures are for half the shaft given in Figure 2. .0 is applied to the bottom edge while the Neumann mesh is with 917 elements. The Dirichlet boundary condition 1 (8) 1 .5lines of resulting stress level, indicating the stress boundary condition (6) is applied to the remaining edges. b) Iso 2 .0 concentration at the corner. PSfrag replacements 2.5 3.0 Kt Kt Kt s/ mm 3 .0 3.0 2.5 2.0 1.5 1.0 0.5 0.0 a) 0 5 10 s/mm 15 20 1.8 b) 7.5 7.7 7.9 s/mm 8.1 8.3 2.2 2.6

Figure 4: The gures are for half the shaft shown in Figure 2 and show the stress concentration as a function of the arc length. a) The stress concentration factor along the keyway boundary starting from the external point until the center point. b) Stress concentration close up, here only show along the r = 0.6mm llet at the corner. The maximum value is Kt = 2.93; with a llet radius of 0.4mm which is also allowed by the standard the value is Kt = 3.32.

constant along major parts of the surface in order for the design to be optimal, see e.g. [19]. The stress level is such that for a llet radius of r = 0.6mm we nd Kt = 2.93, with a llet radius of r = 0.4mm which is also allowed by the standard the value is Kt = 3.32. This is a rather large variation in the stress concentration for designs that fullls the standard geometry. 3. Keyway optimization In keyway design as in many other designs within machine elements the standard preferred shape is the circle or a semicircle. This is probably due to the simple parameterization and/or ease of manufacturing. For the sled-runner design or the prole keyway there are however no difculty in introducing a different llet shape. It is well known from shape optimization that the circular shape is seldom optimal with respect to stress concentrations, see e.g. [19]. From a practical point of view focus should be on simplicity, although the optimization result should still be near to the optimal design. That a given parameterization is sufciently exible, i.e. that it can return optimal designs, can only be checked or veried after an actual optimization procedure. If the stress is constant along major parts of the surface then the shape is assumed optimal, see [19]. The parameterization chosen here is to use the super ellipse due to the simple parameterization and due to previous results obtained with this shape in relation to stress concentrations for other problems. The design domain is shown in Figure 5, where the elliptical shape can be seen for the llet. The super ellipse (with super elliptical power ) is in parametric form given by

X = L1 + A cos()(2/) , Y = L2 + B sin()(2/) ,

] 2 [0 : ] 2 [0 :

(18) (19)

PSfrag replacements b/2

                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                       L 1                                                                                                                                                                                     B                                                                                                                                                                                                                                          A                                                                                                                                                                                     Y                                                    L 2                                                                                                                                                                                                                                                  X                                       
t

Figure 5: The design domain: half a keyway where the llet is a super ellipse with semi-major axes A and B . The keyway design is according to Figure 5 fully controlled by ve design parameters, width b, depth t, length L1 and L2 and super ellipse power . The optimization is performed for a 100mm shaft. This is of course a specic choice of shaft diameter but the results will indicate what level of stress improvements are possible more generally. In the example the other preselected values are b = 28mm, L1 = 7.4mm

i.e., the width complies with DIN 6885 and the shoulder length L1 complies with the largest allowable llet ratio r = 0.6mm. In principle there are now three design variable; the depth t, length L 2 and the super elliptical power . However from the preformed parameter study it is found that L2 = 0 and the length parameter L2 is therefore not an active design parameter. The parameter study results in the optimized values L2 = 0mm, t = 11.51, = 1.99

An iso line plot of largest principal stress is presented in Figure 6. The iso lines in Figure 6 close to the llet are parallel to the llet indicating constant stress along the shape. This is visualize in Figure 7 that show the stress concentration factor along the keyway boundary. The dotted straight line indicates the maximum value, which in this case is Kt = 1.65, it is seen that the stress is close to being constant along the llet. The plot shows the stress concentration factor along half the keyway. It is known that at the starting corner the stress must be zero and then the stress must build up to the maximum value, which in this case is almost constant along the llet. Although the parameterization chosen is very simple with only two active design parameters the design is close to the optimum. A better parameterization with more design variables might lead to a more constant stress along the shape but from Figure 7 it is seen that the room for improvement is small. The maximum stress has for this design been reduced with 43.7% relative to the original design. The design improvement has been achieved using the same shoulder length as specied by the standard. The load carrying capacity is therefore identical. Acknowledgments For discussions and suggestions I wish to thank Prof. Pauli Pedersen and Prof. Peder Klit.

Figure 6: Iso lines of largest principal stress for the optimized design. Kt Kt = 1.65

PSfrag replacements 1.6 1.2 0.8 0.4 0.0

s/mm 0 5 10 15 20

Figure 7: The stress concentration as a function of the arc length along the keyway from the external point until the center point.

References [1] DIN 6885-1. Pafedern nuten (in german), 1968. [2] R. L. Norton. Machine design: An integrated Approach, third edition. Pearson education Inc., Upper Saddle River, N.J. 07458, 2006. 984 pages. [3] J. E. Shigley and C. R. Michke. Mechanical Engineering Design 6th ed. McGraw Hill, Singapore, 2003. 1230 pages. [4] W. C. Orthwein. A new key and keyway design. J Mech Des Trans ASME, 101(2):338341, 1979. [5] H. E. Merritt. The design of cylindrical keys. Machinery (London), 27(701):729732, 1926. [6] A. Kuske. Erh ohung der lebensdauer durch verbesserung der bauteilgestalt. Stahl und Eisen, 91(8):44651, 1971. [7] L. N. G. Filon. On the resistance to torsion of certain forms of shafting, with special reference to the effect of keyways. Philosophical Transactions of the Royal Society of London. Series A, Containing Papers of a Mathematical or Physical Character, 193:309352, 1900. 7

[8] M. M. Leven. Stresses in keyways by photoelastic methods and comparison with numerical solutions. Proceedings of the society for experimental stress analysis, 7(2):141154, 1949. [9] H. Fessler, C. C. Rogers, and P. Stanley. Stresses at end-milled keyways in plain shafts subjected to tension, bending, and torsion. Journal of Strain Analysis, 4(3):180189, 1969. [10] H. Fessler, C. C. Rogers, and P. Stanley. Stresses at keyway ends near shoulders. Journal of Strain Analysis, 4(4):267277, 1969. [11] W. C. Orthwein. Keyway stresses when torsional loading is applied by the keys. Experimental Mechanics, 15(6):245248, 1975. [12] M. Eissa and H. Fessler. Reduction of elastic stress concentrations in end-milled keyed connections. Experimental Mechanics, 23(4):401408, 1983. [13] H. Fessler and T. Appavoo. On the effect of key edge shape on keyway edge stresses in shafts in torsion. Journal of Strain Analysis, 24(3):121125, 1989. r keilnuten. Materialpr [14] K. Terada. Erneute untersuchung der formzahl fu ufung, 5(10):385387, 1963. [15] H. Okubo, K. Hosono, and K. Sakaki. The stress concentration in keyways when torque is transmitted through keys. Experimental Mechanics, 8(8):375380, 1968. [16] R. E. Peterson. Stress concentration design factors. John Wiley & Sons, inc., New York, 1953. 155 pages. [17] W. D. Pilkey. Petersons stress concentration factors. John Wiley & Sons, inc., 605 Third Avenue, New York, NY, 1997. 508 pages. [18] R. E. Peterson. Fatigue of shafts having keyways. American Society for Testing Materials - Proceedings, 32(part 2):413419, 1932. [19] N. L. Pedersen and P. Pedersen. Design of notches and grooves by means of elliptical shapes. Journal of Strain Analysis for Engineering Design, 43(1):114, 2008. [20] N. L. Pedersen. Reducing bending stress in external spur gears by redesign of the standard cutting tool. Structural and Multidisciplinary Optimization, 38(3):215227, 2009. [21] COMSOL AB. Stockholm, www.comsol.se, 1998-2009.

S-ar putea să vă placă și