You are on page 1of 19

Mechanism and Machine Theory 35 (2000) 609±627

www.elsevier.com/locate/mechmt

Tradeo€ analysis in minimum volume design of multi-stage


spur gear reduction units
David F. Thompson*, Shubhagm Gupta, Amit Shukla
Department of Mechanical, Industrial and Nuclear Engineering, University of Cincinnati, P.O. Box 210072 Cincinnati
OH 45221-0072, USA
Received 19 January 1999; accepted 14 July 1999

Abstract

The problem of minimum volume preliminary design of simple and multiple-stage spur gear reduction
units has been a subject of considerable interest, since many high-performance power transmission
applications (e.g., automotive and aerospace) require low weight. One of the principal steps in
traditional design procedures relates to the determination of an ``optimal'' value of diametral pitch at
which tooth bending fatigue failure and surface fatigue failure are equally likely. The purpose of this
paper is to present a generalized optimal design formulation with multiple objectives which is, in
principle, applicable to a gear train of arbitrary complexity. The methodology is applied to the design of
two-stage and three-stage spur gear reduction units, subject to identical loading conditions and design
criteria. The approach serves to extend traditional design procedures by demonstrating the tradeo€
between surface fatigue life and minimum volume using a basic multiobjective optimization procedure.
This information allows the designer to judge overall trends and, for example, to assess the penalty in
surface fatigue lifetime which would occur for a given weight reduction. # 1999 Elsevier Science Ltd.
All rights reserved.

Keywords: Optimal gear design; Multiobjective optimization; Fatigue design-gears

1. Introduction

Techniques for minimum volume (i.e., minimum weight) preliminary design of simple and

* Corresponding author. Tel.: +1-513-556-3693; fax: +1-513-556-3390.


E-mail address: david.thompson@uc.edu (D.F. Thompson).

0094-114X/00/$ - see front matter # 1999 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 9 4 - 1 1 4 X ( 9 9 ) 0 0 0 3 6 - 1
610 D.F. Thompson et al. / Mechanism and Machine Theory 35 (2000) 609±627

multiple-stage spur gear reduction units have been a subject of considerable interest. In the
earlier literature, perhaps one of the ®rst notable presentations of an optimal gear design
formulation is that of Johnson [1]; recent work related to the minimum-volume design problem
includes that of Savage, Coy, and Townsend [2], Carroll and Johnson [3], and in addition an
AGMA Information Sheet, ``A Rational Procedure for the Preliminary Design of Minimum
Volume Gears,'' [4]. In the latter reference, a closed-form procedure is presented which, under
speci®c assumptions, is applicable to single-stage and two-stage gear sets.
The purpose of this paper is to present an alternative, generalized approach to the minimum
volume design of spur gear units based upon a traditional nonlinear programming, optimal
design formulation with multiple objectives. One of the immediate bene®ts of this approach is
that it is, in principle, applicable to a gear train of arbitrary complexity. In this paper, the
methodology is applied to the design of two-stage and three-stage spur gear reduction units,
subject to identical loading conditions and design criteria. An apparent disadvantage of this
approach is that a closed-form solution, or solution by manual iteration, is not generally
possible although this problem is mitigated by the relative ease of use of present-day
optimization software, such as the MATLAB Optimization Toolbox [5].
The primary motivation for this work stems from the lead author's recent experience in
developing design projects for undergraduate courses in machine design, and for a graduate
course in optimal design, at the University of Cincinnati. The principal textbook used in the
undergraduate machine design sequence is Juvinall and Marshek [6]. Although this reference
does not contain the full detail of the AGMA standards, it is more than adequate to illustrate
the key concepts. Hence, for ease of presentation, the basic assumptions and notation used in
[6] will be followed in this paper, with the understanding that the formulation could be
extended and design constraints added or modi®ed in a straightforward manner.
Perhaps the key contribution of this paper can be outlined as follows. One of the principal
steps in the design procedure presented in [6] relates to the determination of an ``optimal''
value of diametral pitch P at which tooth bending fatigue failure and surface fatigue (i.e.,
pitting) failure are equally likely. For cyclic loading in simple bending, ferrous metals exhibit
the well-known endurance limit phenomenon, and hence in®nite fatigue life with respect to the
bending load is theoretically possible. However, the Hertz contact stress occurring on the tooth
surface represents a state of tri-axial stress which, in cyclic loading, does not exhibit the same
endurance limit characteristic. Hence, a speci®ed lifetime in surface fatigue must be assumed a
priori which a€ects all subsequent calculations. Although the procedure outlined in [4] is
generalized, this basic assumption remains. The approach in this paper is to extend this
procedure by studying the tradeo€ between surface fatigue life and minimum volume using a
basic multiobjective optimization procedure. It is asserted here that consideration of surface
fatigue lifetime within the context of a tradeo€ analysis, rather than as a hard constraint, also
makes sense from the standpoint of practical design considerations, surface fatigue failure is
generally not a well-de®ned, catastrophic failure (as is tooth breakage in bending fatigue).
The ``answer'' obtained from a multiobjective optimization analysis is not a single optimum
design but rather a tradeo€ curve (i.e., a Pareto optimal set ) which represents a collection of
optimal designs, each of which is optimal for a given surface fatigue lifetime. This information
allows the designer to judge overall trends and, for example, to assess the penalty in surface
fatigue lifetime which would occur for a given weight reduction. Similar analyses have
D.F. Thompson et al. / Mechanism and Machine Theory 35 (2000) 609±627 611

undoubtedly been conducted over the years by experienced gear designers, perhaps involving
signi®cant trial and error, but to the authors' knowledge, the proposed methodology has not
been expressed formally in the literature.

2. De®nitions and assumptions

As indicated in Section 1, the basic nomenclature for spur gear design adopted in [6] will be
used in this paper. Important de®nitions include the pitch diameter d, tooth number N,
diametral pitch P (1/in. in English units), addendum diameter da, and face width b, as
indicated in Fig. 1. For a mating gear and pinion, the corresponding variables will be denoted
as dg, dp, Ng, Np, etc., when necessary. Note that N ˆ Pd and, for standard, full-depth involute
teeth, the addendum diameter is given as da ˆ d ‡ 2=P: The base circle diameter is de®ned to be
db ˆ dcos f where f represents the pressure angle. (The value of f ˆ 208 is used in this paper.)
The maximum tooth bending stress s in the vicinity of the root ®llet is estimated as
Ft P
sˆ Kv Ko Km …2:1†
bJ
where Ft represents the tangential gear force, and J the Lewis spur gear geometry factor. The
geometry factor J will be a function of f and the tooth numbers of the pinion and mating gear
(Fig. 2). The velocity factor Kv is a function of pitch line velocity and degree of precision in the
tooth pro®les, re¯ecting the degree of impact which occurs as the teeth come into mesh. The
overload factor Ko is an empirical factor, chosen to re¯ect the degree of non-uniformity in the
driving and load torques, whereas the mounting factor Km is a factor chosen to re¯ect the
accuracy of gear alignment.
The e€ective endurance limit Sn of the gear material in simple bending is estimated from

Fig. 1. Illustration of the basic involute gear geometry. The actual location of the contact point between mating
teeth varies with rotation, but must remain outside the base circle for no interference to occur. For additional
details, see [6].
612 D.F. Thompson et al. / Mechanism and Machine Theory 35 (2000) 609±627

Fig. 2. Plot of third-order polynomial ®t representation of Lewis geometry factor J versus tooth number N
(assuming a 50±85 tooth mating gear) and of surface fatigue life factor CLi versus number of cycles, both from [6].

Sn ˆ S n0 Cs kr kms …2:2†

where Sn0 represents the standard R.R. Moore endurance limit in rotating bending (estimated
for steel as 250HB psi where HB represents the Brinnell hardness number), Cs represents the
surface factor, kr the reliability factor (equal to 1.0 for 50% reliability), and kms the mean stress
D.F. Thompson et al. / Mechanism and Machine Theory 35 (2000) 609±627 613

factor, equal to 1.0 for idler gears subject to completely reversed bending and equal to 1.4 for
input/output gears subject to one-way bending. The surface factor Cs is a function of both
surface ®nish and of material hardness HB, the latter, well-known relationship being one in
which the value of Cs diminishes with increasing hardness.
The Hertz contact stress sH at the tooth surface is estimated according to [6]:
r
Ft
sH ˆ Cp Kv Ko Km …2:3†
bdI
where Cp represents the elastic coecient of the material, d represents the diameter of the
element (i.e., pinion or gear) being analyzed, and I represents a geometry factor given by
1 dg
I ˆ sin f cos f …2:4†
2 dg ‡ dp

The tangential force Ft, face width b, and factors Kv, Ko and Km are identical to those in (2.1).
The e€ective surface fatigue strength SH for the material is estimated according to
SH ˆ Sfe CLi CR …2:5†
where Sfe represents the reference value of surface fatigue strength for a speci®c material with a
level of 99% reliability at 107 cycles. The factor CLi represents a surface fatigue life factor,
while CR represents a reliability adjustment factor; these empirically derived factors are chosen
so as to obtain values of SH for di€ering lifetimes and/or reliability. In a traditional gear
design approach, a speci®ed surface fatigue life requirement dictates the choice of CLi, which is
generally made by the use of a plot of CLi versus 99% reliability lifetime; an example of such a
plot (derived from [6]) is given in Fig. 2. Conversely, the approach used in this paper is, for a
speci®ed geometry and loading, to calculate (and ultimately minimize) the CLi, thus maximizing
the corresponding surface fatigue lifetime.
In addition to the constraints on bending fatigue and surface fatigue (2.1)±(2.5), a number of
constraints involving gear size and geometry come into play. For standard spur gears, a
practical limitation exists which governs the range of acceptable face width b. In [6] this
constraint is given in terms of the diametral pitch P as follows:
9 14
RbR …2:6†
P P
with P in units of 1/in. The other important geometric constraint is that of prohibiting
interference, i.e., requiring that the point of tangency between the pinion and gear tooth
pro®les remains on the involute portion of the pro®le, outside of the base circle. This
constraint, which is given in terms of a maximum allowable addendum diameter of the gear
dag, is derived in [6] as follows:
q
ÿ 2
dag R2 dbg =2 ‡c2 sin2 f …2:7†

where dbg denotes the base circle diameter of the gear and c ˆ …dp ‡ dg †=2 denotes the center
distance. Through a straightforward manipulation, it can be shown that it is sucient to
614 D.F. Thompson et al. / Mechanism and Machine Theory 35 (2000) 609±627

consider (2.7) only, and that a similar constraint involving the pinion diameters dap and dbp is
redundant.
Note that since the gear addendum diameter for the standard involute pro®le is given as
dag ˆ dg ‡ 2=P, one possible means of eliminating interference if it occurs is to use a larger
value of pitch P, thus reducing the addendum diameter (thereby increasing the tooth number,
since Ng ˆ Pdg ). Generally speaking, it follows that interference is avoided if the tooth
numbers are suciently large (i.e., if the teeth are suciently small relative to the pitch
diameter). A surprisingly e€ective rule of thumb which can be used instead of (in an
approximate sense), or in addition to (2.7), is a lower bound on pinion tooth number Np of at
least 16±18 teeth which is usually sucient to prevent interference.
Based upon the preceding discussion, the presence of con¯icting objectives and/or constraints
in the gear design problem should become apparent. The values of dp, dg, b, P, core and/or
surface hardness of the gear material, and any other undetermined parameters (e.g.,
distribution of the overall ratio among multiple gear sets) must be selected so that all
constraints are satis®ed simultaneously and that a minimum value of volume is achieved. The
problem of simultaneous satisfaction of a number of such constraints, irrespective of a speci®c
objective, is referred to as a feasibility or constraint satisfaction problem. Note that, in general,
such a problem may have many solutions, or no solution, depending upon the nature of the
constraint set. Together with an objective function (e.g., weight or volume), such a problem
becomes a constrained optimization problem. If it is sought to simultaneously extremize two or
more objective functions (e.g., weight and surface fatigue life) subject to other constraints, the
problem becomes one of constrained multiobjective optimization. Prior to proceeding with the
formulation of the gear design problem within this framework, some of the basic de®nitions
and nomenclature pertaining to optimal design are brie¯y reviewed here.

3. Optimal design theory and multiple objectives

A number of excellent, readily accessible texts which cover optimal design within a
mechanical engineering framework are available, including Papalambros and Wilde [7], Arora
[8], and Rao [9]. Other, more general texts on the subject include Gill, Murray, and Wright [10]
and Luenberger [11]. The interested reader should refer to one or more of these references, as
only the essential de®nitions will be reviewed here. A standard, single-objective constrained
optimization problem is posed as follows:

min f…x† …3:1†


x

subject to the following set of inequality and equality constraints,


g1 …x†R0

..
.
D.F. Thompson et al. / Mechanism and Machine Theory 35 (2000) 609±627 615

gm …x†R0

h1 …x† ˆ 0

..
.

hp …x† ˆ 0 …3:2†

where x 2 Rn is said to be the design vector, which comprises all of the adjustable parameters
of interest for the system. The m scalar-valued inequality constraint functions g1 , . . . , gm and
the p scalar-valued equality constraint functions h1 , . . . , hp together comprise the constraint set.
A point x is said to be feasible if and only if it satis®es all of the constraints functions of (3.2).
As indicated in the preceding section, a particular constraint set may have one, many, or no
feasible points. A point x is said to be a local minimum for this problem if it is feasible and if
f…x†rf…x † for all feasible x within some neighbourhood of x : A point x is said to be a global
minimum if f…x†rf…x † for all x within the feasible set.
A multiobjective optimization problem is posed as follows

min f1 …x†, . . . , fr …x† …3:3†
x

with the r scalar-valued objective functions f1 , . . . , fr : A constrained multiobjective problem


would also include a constraint set of the form in (3.2). In general, the objective functions
f1 , . . . , fr will be competing in nature and a ``Utopia'' point which simultaneously minimizes all
r objective functions will not exist. For this problem, a point x is said to be a Pareto optimal
or non-inferior point if it is feasible and if there exists no other (feasible) point x which
simultaneously reduces all of the objectives; i.e., there exists no feasible x for which
f1 …x† < f1 …x †

..
.

fr …x† < fr …x † …3:4†

(On a technical note, a weaker but slightly more general de®nition of Pareto optimality is often
cited in the literature; however, this somewhat more obvious de®nition will suce for our
purposes.) As indicated in Section 1, the multiobjective problem (3.3) will (if solutions exist at
all) in general yield a continuum of Pareto optimal points, in the form of a tradeo€ curve (for
r ˆ 2† or a tradeo€ surface …r > 2). The designer must then select a ®nal design from among
the solutions on the Pareto set, based upon any of various preference criteria. Although the
development of solution techniques for the multiobjective problem (3.3) remains an emerging
research area, a number of e€ective methodologies have been developed, including those due to
Das and Dennis [12] and Bras and Mistree [13]. It is possible to show, however, that the
616 D.F. Thompson et al. / Mechanism and Machine Theory 35 (2000) 609±627

solution to the scalarized, weighted-sum optimization problem,

min a1 f1 …x† ‡    ‡ ar fr …x† …3:5†


x

subject to the constraint set (3.2), for a speci®c set of weights a1 , . . . , ar , yields a single Pareto
point. A subset of the Pareto optimal set can thus be generated through systematic variation of
the weights a1 , . . . , ar and repeated solution of the scalarized problem (3.5). Furthermore, it
can be shown that solutions to (3.5) will cover the entire Pareto set if the individual objectives
f1 , . . . , fr are all strictly convex. Although such an approach can be inecient for certain large-
scale problems, it is advantageous from the standpoint of simplicity and intuitive appeal and,
henceforth, is the solution methodology employed in this paper.

4. Design variables and constraint formulation

In this section, the design variables and constraint set for a three-stage spur gear reduction

Fig. 3. Schematic illustration of a three-stage spur gear reduction unit with the associated design variables indicated
including gear and pinion diameters, face width, diametral pitch, and core hardness values. The driving torque Tin
and overall speed ratio e are prescribed.
D.F. Thompson et al. / Mechanism and Machine Theory 35 (2000) 609±627 617

unit is developed in the standard form described in Eqs. (3.1) and (3.2). Similar formulations
for a two-stage unit, as well as for a greater number of stages, would follow in an obvious
fashion. A completely general formulation for an arbitrary number N of stages appears
straightforward, but would be considerably more awkward for presentation here. A schematic
illustration of the three-stage unit is shown in Fig. 3. The unit is driven by an input torque Tin
and an overall speed ratio e for the gear train is prescribed. For each stage the following
constraints, as outlined in Section 2, must be enforced:
Tooth bending fatigue failure (2.1) and (2.2):

Ft P
Kv Ko Km RS n0 Cs kr kms …4:1†
bJ

Tooth surface fatigue failure (2.3)±(2.5):


r
Ft
Cp Kv Ko Km RSfe CLi CR …4:2†
bdI

Gear face width (2.6):

9 14
RbR …4:3†
P P

Interference (2.7):

q
ÿ 2
dag R2 dbg =2 ‡c2 sin2 f …4:4†

Minimum pinion tooth number:

Np r16 …4:5†

As indicated in the preceding discussion (4.5) is optional and the choice of Np r16 is essentially
arbitrary. In addition, it is assumed that the diameters of the two shafts ds1 and ds2 are to be
determined and that the maximum torsional stress present in the shafts must be below an
acceptable value, i.e.,
16Tin dg1
Rtmax …4:6†
pd 3s1 dp1

16Tin dg1 dg1


Rtmax …4:7†
pd 3s2 dp1 dp2
618 D.F. Thompson et al. / Mechanism and Machine Theory 35 (2000) 609±627

A candidate set of design variables would include the pinion diameters dp1, dp2, and dp3, gear
diameters dg1, dg2, and dg3, face widths b1, b2, and b3, diametral pitch values P1, P2, and P3,
gear material (steel) core Brinnell hardness numbers H1, H2, and H3, and shaft diameters ds1
and ds2. However, since the speci®ed overall speed ratio of the gear train is given as
e ˆ dp1 dp2 dp3 =dg1 dg2 dg3 , it is preferable to eliminate this equality constraint - here it is chosen to
eliminate dg3 as
dg3 ˆ dp1 dp2 dp3 =dg1 dg2 e …4:8†

The design vector x in its ®nal form is then organized as follows.


 
x ˆ dp1 dg1 b1 P1 H1 dp2 dg2 b2 P2 H2 dp3 b3 P3 H3 ds1 ds2 …4:9†

Several other factors in Eqs. (4.1)±(4.5) must be expressed in terms of the design vector x and
other known quantities such as the driving torque Tin. For example, since the input shaft
drives the pinion of the ®rst gear set, the tangential force Ft present in the ®rst gear set must
be expressed as
2Tin
Ft1 ˆ …4:10†
dp1

Since each gear set provides torque multiplication, the tangential forces present in the second
and third stages must be expressed as follows:
2Tin dg1
Ft2 ˆ …4:11†
dp2 dp1

2Tin dg1 dg2


Ft3 ˆ …4:12†
dp3 dp1 dp2

With appropriate substitution and reduction, the constraint set for the three-stage design
problem may subsequently be expressed as follows:
ÿ 
g1 …x† ˆ kb P1 ÿ b1 J P1 dp1 dp1 ‰250 H1 ŠCs …H1 †R0 …4:13†

ÿ 
g2 …x† ˆ kb P2 dg1 ÿ b2 J P2 dp2 dp2 dp1 ‰250 H2 ŠCs …H2 †R0 …4:14†

ÿ 
g3 …x† ˆ kb P3 dg2 dg1 ÿ b3 J P3 dp3 dp3 dp2 dp1 ‰250 H3 ŠCs …H3 †R0 …4:15†

g4 …x† ˆ kt dg1 ÿ d 3s1 dp1 R0 …4:16†

g5 …x† ˆ kt dg1 dg1 ÿ d 3s1 dp1 dp1 R0 …4:17†

g6 …x† ˆ ÿb1 P1 ‡ 9R0 …4:18†


D.F. Thompson et al. / Mechanism and Machine Theory 35 (2000) 609±627 619

g7 …x† ˆ ÿb2 P2 ‡ 9R0 …4:19†

g8 …x† ˆ ÿb3 P3 ‡ 9R0 …4:20†

g9 …x† ˆ b1 P1 ÿ 14R0 …4:21†

g10 …x† ˆ b2 P2 ÿ 14R0 …4:22†

g11 …x† ˆ b3 P3 ÿ 14R0 …4:23†

sin2 f 2 ÿ 
g12 …x† ˆ P1 dg1 ÿ P 1 dp1 2dg1 ‡ dp1 ‡ 1R0 …4:24†
4

sin2 f 2 ÿ 
g13 …x† ˆ P2 dg2 ÿ P 2 dp2 2dg2 ‡ dp2 ‡ 1R0 …4:25†
4

sin2 f 2 2 ÿ 
g14 …x† ˆ P3 dp3 dp2 dp1 ÿ P 3 d p3 2dp2 dp1 ‡ edg2 dg1 ‡ edg2 dg1 R0 …4:26†
4

g15 …x† ˆ ÿP1 dp1 ‡ 16R0 …4:27†

g16 …x† ˆ ÿP2 dp2 ‡ 16R0 …4:28†

g17 …x† ˆ ÿP3 dp3 ‡ 16R0 …4:29†

Eqs. (4.13)±(4.15) represent the bending fatigue constraints, Eqs. (4.16) and (4.17) represent the
shaft torsional stress constraints, Eqs. (4.18)±(4.23) re¯ect the face width constraints, Eqs.
(4.24)±(4.26) represent the interference constraints, and Eqs. (4.27)±(4.29) represent minimum
pinion tooth number constraints. The lumped constants kb in the bending fatigue constraints
and kt in the shaft torsional stress constraints are given as
2Tin Kv Ko Km
kb ˆ …4:30†
kr kms

16Tin
kt ˆ …4:31†
ptmax
Note that the constraint expressions in Eqs. (4.13)±(4.29) are not normalized, it is suggested in
[8] that such normalization is often advantageous from the standpoint of numerical scaling.
The advantage of the present form lies in its greater algebraic simplicity, particularly from the
standpoint of calculating user-supplied gradient expressions, which also proves advantageous
from a numerical standpoint.
620 D.F. Thompson et al. / Mechanism and Machine Theory 35 (2000) 609±627

The functional expression J… † in Eqs. (4.13)±(4.15) indicates the Lewis geometry factor as a
function of pinion tooth number. This quantity is approximated via a third-order polynomial
®t (shown in Fig. 2) based upon the data in [6], assuming load sharing between teeth and a 50±
85 tooth mating gear. Similarly, the functional expression Cs … † denotes the fatigue strength
surface factor as a function of the Brinnell hardness number, also based upon data in [6]
assuming a ``machined'' ®nish; in this case, a straight-line approximation is sucient. The
quantities of the form 250 H represent an approximate value of Sn0 , in psi units, as a function
of the Brinnell hardness. Since, for the sake of simplicity, it was assumed that the pinion and
gear of each gear set would possess the same core hardness (e.g., H1 for gear set 1), it is
sucient to consider bending fatigue failure of the pinion only; this assumption could also be
generalized, in which case (for the three-stage design) six core hardness variables and six
bending fatigue constraints would be required. From the standpoint of surface fatigue, it is
generally advantageous to surface harden the gears to the maximum extent possible; in this
analysis, an arbitrary value of 500 Bhn is selected. Hence, additional constraints on the core
hardness variables H1, H2 and H3 are added, of the form 200RHR500, where the lower
bound of 200 Bhn is also arbitrary. Since the surface fatigue lifetime for each gear set is treated
as an objective function, and not as a constraint, surface fatigue criteria do not appear in the
constraint set (4.13)±(4.29). The composite objective function for the problem is developed in
Section 5, to follow.

5. Formulation of objective functions

As indicated in Section 1, a formulation with multiple objective functions will be developed


to facilitate analysis of the tradeo€ between volume and surface fatigue life for the gear sets.
As indicated in Section 3, a weighted-sum approach will be used to form a composite, scalar
objective function. The candidate objective function proposed is of the form
ÿ 
f…x† ˆ a0 f0 …x† ‡ aCLi f1 …x† ‡ f2 …x† ‡ f3 …x† …5:1†

where f0 …x† represents the total volume of material in the gear train and where f1 …x†, f2 …x†, and
f3 …x† are equal to C 2Li for the respective gear sets, where CLi is the surface fatigue life factor
(Fig. 2).
The volume function f0 …x† is expressed as
"  2 !
p  2    d d
p2 p1 ÿ
f0 …x† ˆ d p1 ‡ d 2g1 b1 ‡ d 2p2 ‡ d 2g2 b2 ‡ 1 ‡ d 2p3 b3 ‡ d 2s1
4 edg2 dg1
#

‡ d 2s2 Ls …5:2†

where Ls represents the ®xed shaft length. The surface fatigue objective functions are obtained
by solving Eq. (4.2), assuming equality, for C 2Li : With appropriate substitutions and reduction,
these objective functions are found to be
D.F. Thompson et al. / Mechanism and Machine Theory 35 (2000) 609±627 621
ÿ 
ks dp1 ‡ dg1
f1 …x† ˆ …5:3†
b1 d 2p1 dg1

ÿ 
ks dp2 ‡ dg2 dg1
f2 …x† ˆ …5:4†
b1 d 2p2 dg2 dp1

ÿ 
ks dp2 dp1 ‡ edg2 dg1 dg2 dg1
f3 …x† ˆ ÿ 2 …5:5†
b3 dp3 dp2 dp1

where the lumped constant ks is given as

4C 2p Kv Ko Km Tin
ks ˆ …5:6†
cos f sin fS 2fe C 2R

Since smaller values of CLi imply longer surface fatigue life (cf., Fig. 2), we also seek to
minimize these objectives. However, in a practical situation, values of CLi which di€er
signi®cantly from one gear set to another would lead to an inferior design, since a gear set with
a value of CLi signi®cantly larger than the others would fail in surface fatigue considerably
sooner. Hence, the practical problem which is implied is, collectively, to minimize the values of
f1 …x†, f2 …x†, and f3 …x† with the equality constraint(s) f1 …x† ˆ f2 …x† ˆ f3 …x†, thus implying that
whatever the resultant surface fatigue lifetime, it will be the same for each of the three gear
sets.
However, the use of equality constraints is in many cases overly restrictive (i.e., from a
numerical standpoint), creating an extremely thin feasible set within the design space. A less
restrictive and more ¯exible approach is to penalize the di€erence between f1 …x†, f2 …x†, and f3 …x†
by means of additional objectives. These additional objectives (essentially penalty functions),
denoted as f4 …x† and f5 …x†, are chosen as
 2  
f2 …x† f3 …x† 2
f4 …x† ˆ 1 ÿ , f5 …x† ˆ 1 ÿ …5:7†
f1 …x† f2 …x†

Thus the ®nal, composite objective function which is formed is subsequently


ÿ  ÿ 
f…x† ˆ a0 f0 …x† ‡ aCLi f1 …x† ‡ f2 …x† ‡ f3 …x† ‡ ap f4 …x† ‡ f5 …x† …5:8†

The Pareto optimal set of interest for this problem, which involves the tradeo€ between
volume …f0 …x†† and surface fatigue lifetime …f1 …x†, f2 …x†, and f3 …x†), will be generated by
systematic variation of a0 relative to aCLi : Appropriate levels for the penalty weight, ap , must
be determined by trial and error to obtain solutions for which the values of f1 …x†, f2 …x†, and
f3 …x† are acceptably close.
622 D.F. Thompson et al. / Mechanism and Machine Theory 35 (2000) 609±627

6. Results and discussion

The formulation and methodology described in this paper is now demonstrated for selected
problems in which two-stage and three-stage minimum volume designs are compared.
(Hereinafter N is used to denote the number of stages.) In the ®rst series of cases, an overall
speed ratio of e ˆ 0:1 (gear ratio of 10 : 1) is chosen with torque input values of Tin ˆ 80, 120,
180 and 270 lb in. In the second series of cases, a single value of Tin ˆ 120 lb in. is chosen and
speed ratios of e ˆ 0:05, 0.0667, 0.1, and 0.15 (gear ratios of 20 : 1, 15 : 1, 10 : 1, and 6.67 : 1)
are evaluated. These values were chosen to be illustrative and are essentially arbitrary. A
variety of other design parameters, including those speci®ed in Eqs. (4.6), (4.7), (4.30), (4.31),
and (5.6) and not cited previously are outlined in Table 1 below.
The numerical analysis is carried out in MATLAB Version 5.2 using the constrained
minimization routine constr which is a component of the MATLAB Optimization Toolbox [5].
This algorithm, which implements a constrained quasi-Newton minimization routine, exhibits
generally good performance on a wide range of problems. Numerical performance is enhanced
by use of analytic, user-supplied cost function and constraint gradient expressions which, as
indicated earlier, was a principal consideration in the current formulation of Eqs. (4.13)±(4.29)
and (5.8).
Results of the analysis are summarized via the Pareto plots of Figs. 4 and 5. Fig. 4 illustrates
the case of e ˆ 0:1 with Tin ˆ 80, 120, 180, and 270 lb in., while Fig. 5 shows the case of Tin ˆ
120 lb in. for speed ratios of e ˆ 0:05, 0.0667, 0.1, and 0.15. The plots provide only the values
of the objective functions (volume and CLi ); additional information of interest would include
values of the design variables (e.g., distribution of individual ratios, pitch values, and face
widths) as well as the active constraints at each point. This information for selected cases is
provided in Table 2. Based upon a review of this information, some conclusions are proposed
here:
. Considering a ®xed ratio (i.e., Fig. 4), the tradeo€ between surface fatigue life and volume

Table 1
Gear design parameters

Description Symbol Value Units

Bending reliability factor (99.0%) kr 0.814 none


p
Elastic coecient Cp 2300 psi
Mean stress factor kms 1.4 none
Mounting factor Km 1.6 none
Overload factor Ko 1.0 none
Pressure angle f 208 degree
Shaft length (N = 2) Ls 8.0 in.
Shaft length (N = 3) Ls 4.0 in.
Surface fatigue strength Sfe 190,000 psi
Surface reliability factor (99.0%) Cr 1.0 none
Torsional stress limit-shaft tmax 25,000 psi
Velocity factor Kv 2.0 none
D.F. Thompson et al. / Mechanism and Machine Theory 35 (2000) 609±627 623

Fig. 4. Pareto optimal sets (tradeo€ curves) for N = 2, 3, for torque inputs of 80, 120, 180 and 270 lb in. with a
prescribed overall speed ratio of e = 0.1.

for the two-stage and three-stage designs produces essentially similar results, particularly at
lower torque levels. At a torque level of 270 lb in., the tradeo€ curve for N = 3 lies
appreciably below the corresponding curve for N = 2, implying that the three-stage design
o€ers at least a modest gain in weight reduction at higher torque (0±10% depending upon
the level of CLi ).
. Considering a ®xed torque input (Fig. 5), the smaller reduction ratios of 6.67 : 1 …e ˆ 0:15†
and 10 : 1 …e ˆ 0:1† show little di€erence between the two-stage and three-stage designs.
However, at the higher ratios of 15 : 1 …e ˆ 0:0667† and 20 : 1 …e ˆ 0:05), the three-stage
design is clearly superior, o€ering a potential weight reduction of as much as 20% over the
two-stage design at the higher ratio. This gain is achieved over a wide range of surface
fatigue lifetimes.
. At larger values of CLi (i.e., shorter surface fatigue lifetime), a greater number of constraints
tend to be active including the bending fatigue constraints, face width constraints, and
minimum pinion tooth number (and/or interference). However, as surface fatigue lifetime is
weighted more heavily, the gear set must generally become larger and, for a value of CLi
suciently small, the bending fatigue constraints become inactive. (Note that the core
624 D.F. Thompson et al. / Mechanism and Machine Theory 35 (2000) 609±627

Fig. 5. Pareto optimal sets (tradeo€ curves) for N = 2, 3, for torque input of 120 lb in. with prescribed overall
speed ratios e = 0.05, 0.0667, 0.1 and 0.15.

hardness values at a particular solution become immaterial when the bending constraints are
inactive, in which case the minimum value of 200 Bhn may be assumed.) In most cases, the
bending constraints were found to become inactive for surface fatigue lifetimes between 105
and 106 cycles. Hence, for assumed surface fatigue lifetimes of 106 cycles or longer, design
based upon an a priori assumption that the bending constraints are active would lead to a
result which is sub-optimal.
. At extremely low values of volume and correspondingly short surface fatigue life, the three-
stage design is also superior, o€ering additional solutions beyond where the two-stage design
is limited by face width and/or interference constraints. A solution at the extreme minimum
volume end of a Pareto curve is, however, inadvisable since due to the slope at this point, a
steep price in shortening of the surface fatigue life must be paid for a small gain in volume
reduction.
It is anticipated that many, if not all of these conclusions would be recognized by experienced
practitioners, in the absence of a formal procedure, through signi®cant trial and error.
Although the methodology appears to have considerable potential, it is felt that more
D.F. Thompson et al. / Mechanism and Machine Theory 35 (2000) 609±627 625

Table 2

Volume (in.3) CLi dg1/dp1 dg2/dp2 dg3/dp3 Np1 Np2 Np3 P1 (in.)ÿ1 P2 P3 b1 (in.) b2 b3 Active constraint

N = 3, Tin ˆ 80 lb in., e = 0.1

11.55 1.45 2.02 2.17 2.29 16 16 16 22.36 17.70 11.80 0.63 0.79 0.76 2±5, 9, 15±17
12.34 1.32 1.47 2.47 2.76 16 17 16 18.54 19.89 13.37 0.60 0.70 0.81 1, 3, 4, 5, 17
14.50 1.20 1.45 2.44 2.83 17 16 16 17.07 15.72 12.34 0.53 0.60 0.79 4, 5
16.63 1.11 1.45 2.44 2.83 22 16 16 20.37 14.76 11.97 0.54 0.61 0.85 4, 5
20.19 1.00 1.45 2.43 2.85 20 16 17 17.50 13.85 12.96 0.60 0.66 1.05 4, 5
24.53 0.89 1.44 2.42 2.87 20 16 18 16.39 13.11 11.94 0.65 0.71 0.92 4, 5
28.17 0.82 1.44 2.42 2.88 19 16 21 16.22 13.45 13.03 0.82 0.87 0.91 4, 5
32.36 0.76 1.43 2.42 2.89 16 16 19 13.87 13.52 12.24 0.99 1.04 1.12 4, 5
35.16 0.72 1.42 2.42 2.90 17 16 26 13.86 12.94 12.94 1.00 1.03 0.73 4, 5
N = 3, Tin ˆ 120 lb in., e = 0.1

17.23 1.45 2.00 2.18 2.30 16 16 16 19.53 14.03 11.81 0.71 0.73 1.13 1±5, 9, 15±17
20.62 1.26 1.47 2.43 2.79 16 16 16 16.22 15.86 10.82 0.72 0.86 0.85 4, 5
25.05 1.13 1.48 2.42 2.79 20 16 16 18.48 14.50 11.49 0.75 0.88 1.20 4, 5
30.46 1.02 1.48 2.42 2.80 22 16 18 18.16 13.32 10.06 0.74 0.87 0.90 4, 5
35.00 0.94 1.47 2.41 2.81 19 16 22 15.37 13.24 10.94 0.90 1.05 0.83 4, 5
40.23 0.87 1.47 2.41 2.82 22 17 26 16.28 12.91 11.82 0.85 0.98 0.76 4, 5
43.72 0.83 1.47 2.41 2.83 19 16 29 14.52 12.02 12.38 0.96 1.10 0.73 4, 5
N = 3, Tin ˆ 180 lb in., e = 0.1

25.70 1.44 1.98 2.18 2.31 16 16 16 17.06 13.21 9.11 0.82 0.97 1.01 1±5, 9, 15±17
27.66 1.29 1.50 2.48 2.69 16 16 16 13.50 13.44 9.50 0.74 0.86 0.94 3, 4, 5, 17
35.19 1.15 1.47 2.42 2.82 19 16 19 13.72 11.45 11.13 0.68 0.78 1.13 4, 5
39.18 1.08 1.46 2.42 2.82 20 17 16 13.77 11.86 9.24 0.72 0.83 1.20 4, 5
46.07 0.99 1.46 2.42 2.83 24 18 18 15.48 11.57 8.90 0.72 0.82 1.03 4, 5
52.02 0.92 1.46 2.41 2.85 17 16 21 12.67 11.57 9.62 1.08 1.20 0.94 4, 5
N = 3, Tin ˆ 270 lb in., e = 0.1

38.65 1.44 1.86 2.29 2.35 18 16 16 14.16 12.11 9.19 0.74 1.15 1.52 1±5, 16, 17
43.47 1.19 1.42 2.45 2.88 26 16 16 16.89 10.72 9.84 0.82 0.84 1.42 3, 4, 5, 17
46.97 1.13 1.39 2.44 2.95 32 22 16 18.87 12.80 8.40 0.72 0.70 1.07 4, 5
51.29 1.07 1.38 2.44 2.97 16 16 16 11.48 11.65 9.30 1.21 1.20 1.42 4, 5
N = 3, Tin ˆ 120 lb in., e = 0.05

35.68 1.40 2.90 2.71 2.54 16 16 16 16.85 12.68 9.82 0.53 0.87 1.42 1±5, 6, 11, 15±17
38.32 1.23 2.38 2.90 2.91 21 16 16 18.38 14.04 10.26 0.49 0.99 1.36 3, 4, 5, 17
42.10 1.13 2.23 2.87 3.12 22 18 17 17.50 14.26 9.36 0.51 0.90 0.96 4, 5
46.33 1.07 2.23 2.87 3.13 22 16 19 16.87 12.38 10.58 0.53 0.93 1.03 4, 5
50.20 1.02 2.22 2.87 3.15 26 17 21 18.51 11.89 11.92 0.48 0.84 1.17 4, 5
55.53 0.96 2.20 2.87 3.17 29 20 16 19.11 13.06 9.11 0.47 0.78 1.31 4, 5, 6, 7
60.26 0.91 2.19 2.87 3.18 27 18 16 18.40 12.17 8.02 0.58 0.98 1.12 4, 5
N = 3, Tin ˆ 120 lb in., e = 0.0667

26.27 1.43 2.46 2.48 2.45 16 16 16 19.53 14.07 9.25 0.71 0.91 0.98 1±5, 9, 15±17
30.34 1.18 1.85 2.69 3.00 19 22 17 16.25 17.47 11.17 0.55 0.69 1.08 1, 3, 4, 5, 6
(continued on next page)
626 D.F. Thompson et al. / Mechanism and Machine Theory 35 (2000) 609±627

Table 2 (continued )

Volume (in.3) CLi dg1/dp1 dg2/dp2 dg3/dp3 Np1 Np2 Np3 P1 (in.)ÿ1 P2 P3 b1 (in.) b2 b3 Active constraint

34.47 1.10 1.86 2.68 3.01 20 16 20 15.85 12.75 10.90 0.56 0.82 0.82 4, 5
43.57 0.95 1.77 2.74 3.10 23 16 16 15.74 11.54 10.04 0.58 0.82 1.31 4, 5
48.08 0.90 1.83 2.67 2.46 31 25 18 18.79 14.90 9.81 0.48 0.66 1.16 4, 5
52.23 0.86 1.82 2.67 3.07 35 20 17 22.56 13.17 9.92 0.60 0.82 1.34 4, 5
N = 3, Tin ˆ 120 lb in., e = 0.1

17.23 1.45 2.00 2.18 2.30 16 16 16 19.53 14.03 11.81 0.71 0.73 1.13 1±5, 9, 15±17
20.62 1.26 1.47 2.43 2.79 16 16 16 16.22 15.86 10.82 0.72 0.86 0.85 4, 5
25.05 1.13 1.48 2.42 2.79 20 16 16 18.48 14.50 11.49 0.75 0.88 1.20 4, 5
30.46 1.02 1.48 2.42 2.80 22 16 18 18.16 13.32 10.06 0.74 0.87 0.90 4, 5
35.00 0.94 1.47 2.41 2.81 19 16 22 15.37 13.24 10.94 0.90 1.05 0.83 4, 5
40.23 0.87 1.47 2.41 2.82 22 17 26 16.28 12.91 11.82 0.85 0.98 0.76 4, 5
43.72 0.83 1.47 2.41 2.83 19 16 29 14.52 12.02 12.38 0.96 1.10 0.73 4, 5
N = 3, Tin ˆ 120 lb in., e = 0.15

11.48 1.46 1.58 1.96 2.15 16 16 16 19.23 16.77 13.40 0.72 0.83 1.04 2±5, 9, 15±17
16.80 1.12 1.10 2.21 2.74 20 17 16 16.80 14.80 11.54 0.78 0.61 0.82 4, 5, 6, 9
20.39 1.00 1.10 2.20 2.75 16 19 17 13.90 16.55 12.80 1.00 0.78 1.09 4, 5, 9, 11, 15
24.76 0.90 1.10 2.19 2.76 16 16 23 12.57 12.73 13.13 1.02 0.78 0.73 4, 5
28.43 0.83 1.10 2.19 2.77 19 25 20 13.91 17.54 11.05 0.99 0.75 0.81 4, 5
31.74 0.78 1.09 2.20 2.79 16 18 20 11.70 13.74 12.73 1.19 0.98 1.09 4, 5, 9, 15
35.64 0.73 1.08 2.20 2.81 16 16 22 11.17 12.45 13.00 1.25 1.09 1.06 4, 5, 9, 15

evaluation must be conducted on a wide range of gear design problems before broader
conclusions can be drawn.

7. Conclusion

As stated in Section 1, it is believed that the approach described here represents a


mathematical formalization of design considerations which are, perhaps, well-understood in an
intuitive sense by experienced gear designers, but which are in many respects extremely dicult
to express in any other framework. The methodology described here provides a formal
procedure for assimilation and evaluation of a wide range of tradeo€ issues in preliminary gear
design. As acknowledged in the Introduction, the speci®c details of the problem statement
outlined in this paper are based upon what are essentially undergraduate machine design
formulations and assumptions (as in Ref. [6]) and obviously do not contain the full detail of
applicable AGMA standards and practices. Nevertheless, it is asserted that the basic
methodology is valid, and potentially of signi®cant value, in the development of future gear
design methodologies.
Potential extensions of this work would include incorporation of more detailed (e.g.,
AGMA) design standards and extension to larger-scale, generalized problems which could
include helical gearing, multiple planetary gear sets, and other more complex systems. Potential
D.F. Thompson et al. / Mechanism and Machine Theory 35 (2000) 609±627 627

computational advancements would include incorporation of the Normal-Boundary


Intersection method [12], the Compromise DSP approach [13], and other emerging
computational techniques for multiobjective optimization and design tradeo€ analysis.

References

[1] R.C. Johnson, Optimum Design of Mechanical Elements, Wiley, New York, 1961.
[2] M. Savage, J.J. Coy, D.P. Townsend, Optimal tooth numbers for compact standard spur gear sets, Journal of
Mechanical Design 104 (1982) 749±758.
[3] R.K. Carroll, G.E. Johnson, Optimal design of compact spur gear sets, Journal of Mechanisms Transmissions,
and Automation in Design 106 (1984) 95±101.
[4] Anon, A Rational Procedure for the Preliminary Design of Minimum Volume Gears, American Gear
Manufacturers Association (AGMA) Information Sheet, AGMA 901-A92, American Gear Manufacturers
Association, Alexandria, VA, 1992.
[5] M.A. Branch, A. Grace, MATLAB Optimization Toolbox User's Guide, The Math Works, Natick, MA, 1996.
[6] R.C. Juvinall, K.M. Marshek, Fundamentals of Machine Component Design, 2nd ed., Wiley, New York, 1991.
[7] P.Y. Papalambros, D.J. Wilde, Principles of Optimal Design: Modeling and Computation, Cambridge
University Press, Cambridge, UK, 1998.
[8] J.S. Arora, Introduction to Optimal Design, McGraw-Hill, New York, 1989.
[9] S.S. Rao, Engineering Optimization: Theory and Practice, Wiley, New York, 1996.
[10] D.G. Luenberger, Linear and Nonlinear Programming, 2nd ed, Addison Wesley, Reading, MA, 1984.
[11] P.E. Gill, W. Murray, M.H. Wright, Practical Optimization, Academic Press, London, 1981.
[12] I. Das, J.E. Dennis, Normal-boundary intersection: a new method for generating the Pareto surface in non-
linear multicriteria optimization problems, SIAM Journal on Optimization, 8 (1999) 631±657.
[13] B.A. Bras, F. Mistree, A compromise decision support problem for robust and axiomatic design, Journal of
Mechanical Design 117 (1995) 10±19.

You might also like