You are on page 1of 196

University of Pisa

An Invariant Approach
to Differential and Kinematic
Analysis of Gears

Doctoral dissertation
in Mechanical Engineering

Marco Gabiccini

February 2006
Università di Pisa
Scuola di Dottorato in Ingegneria “Leonardo da Vinci”

Dottorato di Ricerca in
Ingegneria Meccanica

An Invariant Approach
to Differential and Kinematic
Analysis of Gears
Tesi svolta per il conseguimento del titolo di Dottore di Ricerca

Settore Scientifico Disciplinare:


Meccanica Applicata alle Macchine ING-IND/13

Allievo: Ing. Marco Gabiccini

Tutori: Prof. Ing. Massimo Guiggiani


Ing. Francesca Di Puccio

III Ciclo
Anno 2002
To my wife Angela
We have not succeeded in answering all our problems. The answers we
have found only serve to raise a whole set of new questions. In some
ways we feel we are as confused as ever, but we believe we are confused
on a higher level, and about more important things.

Bernt Øskendal
Stochastic Differential Equations:
An Introduction with Applications

Acknowledgements
There is no amount of thanks that could adequately show how much I
appreciate everything Prof. Massimo Guiggiani has done for me while
wearing his many “hats”. As an advisor, he kept me interested while
keeping me on track. As a counselor, he provided me a ten–month
opportunity at the “GearLab”, Department of Mechanical Engineering,
The Ohio State University. As a research scientist, he is inspiring in
his ability to actively seek new projects, and pursue them creatively.
His broad knowledge and deep insight were of great importance to my
research.
I could not have done half the things I did without Ing. Francesca Di
Puccio. The countless insights she had while sitting in her office should
warrant a place in this thesis. I owe her a debt not easily repaid.
I have benefitted greatly from interactions with many friends and col-
leagues. They have provided both emotional and technical support
to me in the development of this material: Alessio Artoni, from the
University of Pisa, with the countless fruitful discussions and insights,
Martino Vimercati, from the Politecnico di Milano, with whom I shared
a period of three months at The Ohio State University, a fundamental
experience for my vocational education. I wish to thank also Dr. Ah-
met Kahraman, from The Ohio State University, for his support and
encouragement while at The Ohio State as a visiting scholar.
Finally, I would most especially like to thank my wife, Angela, for
all her support and encouragement during this endeavor. I hope she
knows there is no one in the world as beautiful and inspiring as she is.
Abstract

The present thesis aims to re-formulate the whole theory of gearing


without any recourse to reference systems. This goal is achieved by
means of an approach entirely based on geometric concepts (points,
lines, surfaces, vectors) which exist by themselves and can be treated
as such.
As demonstrated in the earliest chapters, reference systems play no role
in the theoretical development and their introduction can be postponed
till the very end, when actual computations have to be performed. The
final result is a more compact and, hopefully, clearer formulation of the
whole theory.
To show that the, so called, invariant approach can cover all the issues
involved in the theory of gearing, a detailed fresh derivation of many
classical results is provided.
The strength and elegance of the invariant approach emerges particu-
larly in the description of the generation of tooth surfaces with general
3D motions of the tool, which are feasible on nowadays CNC machines
like the ones employed in the face-milling process of hypoid gears.
Also the curvature analysis of the tooth surfaces and the development
of new algorithms for optimizing the micro-geometry of the tooth sur-
faces greatly benefit from the framework of the invariant approach as
far as compactness of the expressions and ease of implementation, as
demonstrated in the latest chapters.
Contents

1 Introduction 1
1.1 Literature review . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Dissertation overview . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2 Generation with fixed axes and constant gear ratio 7


2.1 Generating surface . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Rotating vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3 A first family of surfaces . . . . . . . . . . . . . . . . . . . . . . . . 10
2.4 The enveloping family of surfaces . . . . . . . . . . . . . . . . . . . 12
2.4.1 Preliminary definitions . . . . . . . . . . . . . . . . . . . . . 12
2.4.2 Gear ratio and parameter of motion . . . . . . . . . . . . . 12
2.4.3 The enveloping family of surfaces . . . . . . . . . . . . . . . 12
2.4.4 Derivatives of the enveloping family of surfaces . . . . . . . 13
2.5 Equation of meshing . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.5.1 Classical definition . . . . . . . . . . . . . . . . . . . . . . . 14
2.5.2 Simplified form of equation of meshing . . . . . . . . . . . . 14
2.5.3 Another simplified form of the equation of meshing . . . . . 16
2.6 The envelope surface Γg . . . . . . . . . . . . . . . . . . . . . . . . 17
2.6.1 Explicit definition of the envelope surface (sufficient condition) 17
2.7 Regularity of the envelope surface Γg . . . . . . . . . . . . . . . . . 18
2.8 The truly enveloping surface Γe . . . . . . . . . . . . . . . . . . . . 20
2.9 The surface of action Γf . . . . . . . . . . . . . . . . . . . . . . . . 21
2.9.1 A further note on singular points of the envelope surface Γg 22
2.10 Rotating surfaces and contact lines . . . . . . . . . . . . . . . . . . 23
2.11 Envelope of contact lines on the envelope surface Γg . . . . . . . . 24
2.11.1 The parameter space viewpoint . . . . . . . . . . . . . . . . 24
2.11.2 Regularity of the line lg of singular points . . . . . . . . . . 25
2.12 Envelope of contact lines on the generating surface Γe . . . . . . . 26
2.12.1 Envelope of contact lines in the parameter space . . . . . . 27
2.12.2 Regularity of the line le of singular points on the tool . . . 28
2.13 Sliding velocity between mating surfaces . . . . . . . . . . . . . . . 28
2.14 Application to spiral bevel gears . . . . . . . . . . . . . . . . . . . 29
viii Contents

2.14.1 Preliminary definitions . . . . . . . . . . . . . . . . . . . . . 29


2.14.2 Equation of meshing . . . . . . . . . . . . . . . . . . . . . . 31
2.14.3 Gear surface . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.14.4 Case studied . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.15 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

3 Generation with translating axes and variable gear ratio 35


3.1 Rotating vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.1.1 Algebraic properties of rotating vectors . . . . . . . . . . . 36
3.1.2 Differential properties of rotating vectors . . . . . . . . . . 36
3.2 Surface of the generating tool . . . . . . . . . . . . . . . . . . . . . 37
3.3 A fixed family of surfaces . . . . . . . . . . . . . . . . . . . . . . . 39
3.4 The enveloping family of surfaces . . . . . . . . . . . . . . . . . . . 41
3.4.1 Derivatives of the enveloping family of surfaces . . . . . . . 41
3.4.2 Normal vectors to the enveloping family of surfaces . . . . . 43
3.5 Equation of meshing . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.6 Derivatives of the equation of meshing . . . . . . . . . . . . . . . . 44
3.6.1 Derivatives with respect to the arc length se . . . . . . . . 45
3.7 The envelope surface . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.9 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

4 Generation with supplemental motions 49


4.1 Rotating vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.1.1 Rotation about a mobile axis . . . . . . . . . . . . . . . . . 50
4.1.2 Differential properties . . . . . . . . . . . . . . . . . . . . . 52
4.2 Generating surface . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.3 Generating process in the fixed space . . . . . . . . . . . . . . . . . 53
4.3.1 Moving tool surface . . . . . . . . . . . . . . . . . . . . . . 55
4.3.2 Useful derivatives . . . . . . . . . . . . . . . . . . . . . . . . 56
4.4 The enveloping family of surfaces . . . . . . . . . . . . . . . . . . . 57
4.4.1 Derivatives of the enveloping family of surfaces . . . . . . . 58
4.5 Equation of meshing . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.6 The envelope surface Γg . . . . . . . . . . . . . . . . . . . . . . . . 61
4.7 Kinematic analysis of the enveloping process . . . . . . . . . . . . . 63
4.7.1 Sliding velocity between mating surfaces . . . . . . . . . . . 64
4.7.2 Screw axis of the relative motion . . . . . . . . . . . . . . . 64
4.8 Application to the face-milling process . . . . . . . . . . . . . . . . 65
4.8.1 Vector calculations and numerical example . . . . . . . . . 67
4.9 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.10 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
Contents ix

5 Curvature analysis of gear surfaces 71


5.1 Surface of the generating tool . . . . . . . . . . . . . . . . . . . . . 73
5.2 Curvatures of the generated gear . . . . . . . . . . . . . . . . . . . 75
5.2.1 Normal curvature . . . . . . . . . . . . . . . . . . . . . . . . 76
5.2.2 Relative normal curvature . . . . . . . . . . . . . . . . . . . 77
5.2.3 Geodesic torsion . . . . . . . . . . . . . . . . . . . . . . . . 78
5.2.4 Relative geodesic torsion . . . . . . . . . . . . . . . . . . . . 78
5.2.5 Principal relative normal curvatures and directions . . . . . 79
5.2.6 Principal normal curvatures and directions . . . . . . . . . 79
5.3 Application to spiral bevel gears . . . . . . . . . . . . . . . . . . . 80
5.3.1 Preliminary definitions . . . . . . . . . . . . . . . . . . . . . 81
5.3.2 Preliminary computations . . . . . . . . . . . . . . . . . . . 83
5.3.3 Numerical example . . . . . . . . . . . . . . . . . . . . . . . 85
5.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.5 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

6 Curvature analysis: different methods within the same frame-


work 91
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
6.2 Theoretical background: curvatures of a surface . . . . . . . . . . . 92
6.2.1 Normal curvature and geodesic torsion of a surface . . . . . 92
6.2.2 Tensor of curvature . . . . . . . . . . . . . . . . . . . . . . . 94
6.2.3 Rodrigues’ formula . . . . . . . . . . . . . . . . . . . . . . . 94
6.3 Rotation operator . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
6.4 Surfaces of the gear pair . . . . . . . . . . . . . . . . . . . . . . . . 96
6.4.1 Surface of the first gear . . . . . . . . . . . . . . . . . . . . 96
6.4.2 Surface of the second gear . . . . . . . . . . . . . . . . . . . 97
6.4.2.1 Meshing case . . . . . . . . . . . . . . . . . . . . . 97
6.4.2.2 Generating case . . . . . . . . . . . . . . . . . . . 98
6.5 General definition of the meshing/generating process . . . . . . . . 98
6.5.1 Meshing process . . . . . . . . . . . . . . . . . . . . . . . . 100
6.5.1.1 Conjugate curves in the meshing process . . . . . 101
6.5.2 Generating process . . . . . . . . . . . . . . . . . . . . . . . 102
6.5.2.1 Conjugate curves in the generating process . . . . 103
6.6 Litvin’s approach to curvature: indirect meshing method . . . . . . 103
6.7 Chen’s approach to curvature: indirect generating method . . . . . 108
6.8 New approach to curvature: direct method . . . . . . . . . . . . . 111
6.9 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 114
6.10 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
x Contents

7 Third order analysis of gear surfaces 119


7.1 Third order differential properties of the generated gear . . . . . . 120
7.1.1 Normal curvature . . . . . . . . . . . . . . . . . . . . . . . . 122
7.1.2 Derivative of the normal curvature . . . . . . . . . . . . . . 125
7.2 Third order analysis of the gap surface . . . . . . . . . . . . . . . . 127
7.3 Application to spiral bevel gears . . . . . . . . . . . . . . . . . . . 128
7.3.1 Preliminary definitions . . . . . . . . . . . . . . . . . . . . . 129
7.3.2 Preliminary computations . . . . . . . . . . . . . . . . . . . 131
7.3.3 Numerical example . . . . . . . . . . . . . . . . . . . . . . . 133
7.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
7.5 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138

8 Identification of CNC parameters for micro-geometry optimiza-


tion 139
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
8.2 Gear generation with supplemental motions . . . . . . . . . . . . . 140
8.3 Synthesis of the tooth micro-geometry . . . . . . . . . . . . . . . . 142
8.4 Least squares solution . . . . . . . . . . . . . . . . . . . . . . . . . 145
8.4.1 The full column rank case . . . . . . . . . . . . . . . . . . . 145
8.4.2 The column rank deficiency case . . . . . . . . . . . . . . . 145
8.5 Solution by subset selection using the SVD . . . . . . . . . . . . . 147
8.5.1 Pseudocode of the algorithm . . . . . . . . . . . . . . . . . 149
8.6 Quadratic programming solution . . . . . . . . . . . . . . . . . . . 151
8.7 Linear programming solution . . . . . . . . . . . . . . . . . . . . . 151
8.7.1 Minimization of the L∞ norm . . . . . . . . . . . . . . . . . 152
8.7.2 Minimization of the L1 norm . . . . . . . . . . . . . . . . . 154
8.8 Sequential refinement approach . . . . . . . . . . . . . . . . . . . . 157
8.8.1 A first procedure . . . . . . . . . . . . . . . . . . . . . . . . 162
8.8.2 A second procedure . . . . . . . . . . . . . . . . . . . . . . 163
8.8.3 A third procedure . . . . . . . . . . . . . . . . . . . . . . . 164
8.9 Numerical example . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
8.9.1 Design variables . . . . . . . . . . . . . . . . . . . . . . . . 166
8.9.2 Target ease-off . . . . . . . . . . . . . . . . . . . . . . . . . 167
8.9.3 First results . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
8.10 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173

9 Conclusions 175

Bibliography 184
Chapter 1

Introduction

Theory of gearing looks like a pretty difficult subject. It is based on the envelope
of surfaces and it involves a lot of geometry and complex relative motions. Some
books like Theory of Gearing by F. L. Litvin [48] are now classical, not to mention
more recent contributions, still in the form of books, like [49] and [50] by the
same author. This theory has been widely employed and it would be impossible
to cite here all the papers dealing with this subject. Among these very many
contributions we may recall [53, 45, 74, 79, 30, 70], to mention but a few.
All these analyses are based on a deep recourse to (Cartesian) reference sys-
tems. Indeed, reference systems are typically introduced at the very beginning,
almost implying that the whole theory cannot be properly explained without them.
This idea results in a formulation which looks rather complicated. For instance,
a variety of formulae, one for each reference system, are usually provided for the
same geometrical quantity.
A different approach has been pursued by Dooner [14, 15], which employed the
theory of screws. However, several reference systems are introduced in the first
pages of [15] as well.
This thesis aims to formulate the whole theory of gearing without any recourse
to reference systems. This goal is achieved by means of an approach entirely
based on geometric concepts (points, lines, surfaces, vectors) which exist by them-
selves. All vector functions are treated as such. The principal tool that allows
the description of rotating vectors and their manipulation without the use of their
components in some reference systems is the Euler formula for expressing the ro-
tation of a vector about an axis, which apparently dates back to 1775 and was
first published in [16]. The final result is a more compact and general formulation
of the theory of gearing. To show that the proposed approach can cover all the
main issues involved in the theory of gearing, a detailed fresh derivation of many
classical results is provided. For the sake of comparison, precise citation to former
contributions is given.
Since reference systems play no role at all in the theoretical development, their
2 1. Introduction

introduction can be postponed till the very end, when actual computations have
to be done. Moreover, and quite surprisingly, all computations can be carried out
employing just one reference system, without having to bother about the chain of
fixed or rotating frames typical of the traditional approach.

1.1 Literature review


The Theory of Gearing has its principal manifest in the classical books by Profes-
sor F.L. Litvin [48], [49] and [50] recently joined by his last monumental work [51].
The same author and many other researchers, most of them former Prof. Litvin’s
students, have contributed to the gear theory with a countless number of papers
and technical reports both on the theoretical aspects of gear generation and on
many of its applications: innovative geometrical shapes of the tools and new kine-
matic layouts of the generating processes of several types of gears were described.
Contributes were given also in the development of numerical algorithms for com-
puterized gear cut simulation and contact analysis.
Among the papers more focused on the theoretical issues we can mention [46]
where the envelope of a two parameter family of surfaces is investigated, [6] where
the concept of axis of meshing is employed in gear generation and [52, 53, 59, 5]
dealing with the analysis of lines of singular point, edges of regression and envelopes
of contact lines on the generating surface.
An important aspect is the determination of the relationship between the cur-
vatures of the tool surface and the ones of the gear. This was the topic of many
papers like [38, 39, 67]. Usually the study was based on the hypothesis of fixed axes
and constant gear ratio. Wu and Luo [84] proposed curvature equations in terms of
the limiting functions of the first kind and considering the mating surfaces subject
to relative screw motion with constant translational and rotational velocities. Yan
and Cheng [85] applied the approach proposed in [84] to some cam-follower mecha-
nism. Chen et al. [8] generalized the study of [39] to general 3D motions. Chen [9]
followed Litvin’s approach introducing non-principal parametric coordinates of the
surfaces and developed curvature expressions in a non-orthogonal reference frame.
Similar results on this topic where found by [71] while modelling the kinematics
of grasping an object by a robotic hand with smooth surface fingertips.
Several are the application of the theory of gearing to simulate the cut and
determine the geometry of complex and innovative gear types. Investigations on
the geometry of face gear drives were performed in [55, 86, 64, 63, 58, 41], while
improvements to the geometry of face-milled spiral bevel gears for the localization
of the bearing contact and the synthesis of the instantaneous contact ellipse were
presented in [60, 4, 62].
Suggestions about profile modifications to other types of gears and the subse-
quent analysis were presented in [57, 54, 61, 56, 44].
The presentation of dedicated computer codes suitable for the geometrical sim-
ulation of the cutting process of various types of gears and some suggestions on
1.1 Literature review 3

how to find reasonable initial guesses for the numerical procedures were presented
in [40, 3, 66, 65].
All the above mentioned articles and reports employ matrix methods. Several
reference frames are introduced in the first pages of all the papers and the vectors
are represented by their components. The transformation matrices among the
reference systems of the chain account for the relative motion between the tool
and the gear blank where the tooth surface is to be enveloped.
Different approaches to the formulation of the theory of gearing are due Dooner,
Grill, Miller, Wu and Luo and Vogel.
Dooner [15, 14] proposed three laws of gearing employing the theory of screws
and the cylindroidal coordinates to parametrize the kinematic geometry of gen-
eralized motion transmission between skew axes. The three laws define: (1) a
relationship between the instantaneous motion of two toothed bodies in contact;
(2) a relation between the instantaneous gear ratio and the apparent radii of the
hyperboloidal pitch surface in contact; (3) the instantaneous relationship for the
relative curvature of two conjugate surfaces in direct contact showing that this
relation is independent of the tooth profile geometry. An early establishment of
reference frames from the very beginning of the analysis is evident from the first
pages of [15] as well.
Grill [26] employed kinematic relations in vectorial form and derived the cur-
vature properties of the generated tooth surface by employing a spatial extension
of the Euler-Savary theorem.
Miller [69] employed geometric algebra, a unified framework to express geo-
metric concepts with algebraic symbols extensively treated by Hestenes [29, 28].
The equation of meshing in the sole focus of [69], though.
Wu and Luo [84] formulated the theory of gearing by employing a vectorial
notation but the introduction of some operations like the relative differentiation
and the use of kinematic concepts link them closely to the use of reference systems.
The topics in [84] are numerous: from the determination of the function of the first
and second kind, to the analysis of the induced curvatures and their derivatives to
the extension of the Euler-Savary formula to 3D motions.
Vogel et al. [83] proposed a new methodology to describe the generating process
and perform the tooth contact analysis, that they named direct. Based on concepts
borrowed from differential topology, some classical results are re-proposed in a
mathematically elegant manner. For example, the line of action is considered as a
first order singularity of a certain operator equation. Moreover, with the proposed
use of automatic differentiation of algorithms [25], direct dependencies between
contact ellipses and machine parameters are available and may be quantified in
order to perform optimizations.
The problem of performing a sensitivity analysis of the gear surface to the
machine-tool settings and to misalignments dates back to 1983 in the paper by
Litvin et al [42].
Krenzer [31] performed computer aided inspection of bevel and hypoid gears
4 1. Introduction

to provide a quantitative way to verify the correctness of the tooth surfaces as a


support to the rolling check. The same author proposed in [32] a methodology
to find corrective machine settings to minimize the manufacturing errors. He
approximated the error surface as a polynomial surface subdividing the error types
into first, second and progressively higher orders. Improvements to the methods
were performed by Litvin et al. in [47] and [43] and by Fong and Tsay in [20] by
applying the least squares method.
Studies on the effects of ease-off modifications on the bearing pattern were
performed by Stadtfeld in [75, 76] and by the same author in collaboration with
Gleason’s engineers [77].
Gosselin et al. [24] proposed five quality indexes to characterize the devia-
tions of the real surface from the theoretical one and proposed the Surface Match
algorithm to identify the corrective machine settings.
Fong at al. proposed in [36, 37] alternative algorithms based on the SVD
method and the Sequential Quadratic Programming method to minimize real tooth
surface deviations.
Latest development regarding the identification of machine corrections to op-
timize the contact pattern under load (problem which is closely related with the
minimization of deviations of real tooth surfaces) have been published by Acht-
mann and Bär [1] and by Klingelberg [81].
In [1], the technique of response surfaces was employed to optimize the tooth
bearing ellipse, which is defined as the ellipse that best fits the contact pattern
during the whole mesh cycle. Due to the nature of the methodology used though,
the optimization can be performed only with respect to two parameters at a time
(in the paper the effect of helical motion and modified roll was investigated),
therefore neglecting the crossed dependencies among the whole set of available
correction parameters.
In [81], a bearing pattern optimization procedure was proposed in conjunction
with a minimal representation of the six axis CNC machine kinematics. The
numerical techniques used to perform the identification of the corrections were
not discussed in detail though.

1.2 Dissertation overview


The structure of the present dissertation reflects the path followed by the author
to formulate the so called invariant approach to the theory of gearing. Therefore,
it presents the evolution of the analysis of the gear tooth generation from simple
to fully general relative motions between the tool and the gear blank.
We will now give a brief synopsis of the various chapters in the dissertation.
In Chapter 2, all the main issues of the theory like the enveloping family
of (tool) surfaces, envelope (tooth) surface, surface of action, undercutting line,
envelope of contact lines on the tool surface and the analysis of their regularity
1.2 Dissertation overview 5

conditions is developed in a fresh way for the case of generation with fixed axes
and constant gear ratio.
In Chapter 3, the theory already presented is extended to the generation of
gear surfaces with translating axes and variable gear ratio. For this purpose only
slight variations to the basic theory are necessary.
The development and the use of computer numerically controlled (CNC) ma-
chines for manufacturing gears, like hypoid and spiral bevel gear for aerospace ap-
plications, has demanded for general three-dimensional motions to be addressed.
Therefore, in Chapter 4 the invariant approach to gear generation presented in
the previous chapters is extended to cope with fully general enveloping motions,
like those employed in the, so called, free-form cutting machines. Actually, more
than six parameters (somehow like the eight correction mechanisms in the Phoenix
Gleason machine [77]) are included in the analysis to retain the physical meaning
of some degrees of freedom and to make the whole formulation easier to apply to
some practical cases. First, a detailed analysis to obtain the basic equations of the
generated surface is presented. The proposed approach allows for a very compact
and simple formulation since it does not require reference systems. With respect
to the analyses presented in the previous chapters, several new aspects have to be
addressed, making the generalization far from obvious.
In Chapter 5, a short explanation on how to obtain the curvatures and some
kinematic relationships is provided. Again, some new subtle points have to be
taken into account. To show how the proposed method really works, its application
to the face-milling process is presented.
Chapter 6 is concerned with the comparison of different methods in curvature
analysis already proposed in the gear literature. To better point out the analogies
and the differences of the existing methods, the investigation is performed within
the common framework of the invariant approach.
Chapter 7 is devoted to the third order analysis of gear surfaces, that is the
analysis of the geometry of tooth surfaces and that of the contact up to the deriv-
ative of the curvature. Maybe, this is the least “practical” among the chapters.
Nevertheless, the procedures here outlined to perform the third order analysis of
conjugate surfaces reveal a common structure with the ones presented in Chapter
5 and represent their natural generalization.
In Chapter 8, the last chapter, the invariant approach is employed to solve the
problem of global surface synthesis. The problem could be stated as follows: let
us suppose we are given a basic design of a gear tooth surface, which means that
we are given a set of parameters defining the geometry of the tool and the CNC
machine kinematic settings. How should the machine-tool settings be (slightly)
varied in order for the new surface to be, in some norm, as close as possible to
an assigned surface, with an allowable variation of the settings? Obviously, it is
implied that the objective surface is “very” close, in some norm, to the basic one.
The new surface is a corrected surface and the problem of finding feasible
machine-tool settings which allow for arbitrary corrections of the tooth surface is
6 1. Introduction

one of the topics in the theory of gearing which, at present, stirs the most lively
research activity by industry researchers, as demonstrated by recent papers from
Gleason and Klingelnberg as [81], [17] and [35].
It is believed that, by employing the invariant approach, a better understand-
ing of the correction mechanism can be developed and a deeper insight of the
computational algorithms for surface synthesis can be gained.
Chapter 2

Generation with fixed axes


and constant gear ratio

The purpose of this chapter is to formulate the Theory of Gearing by employing


a new approach. The classical approach that deals with the components of the
vectors and relies deeply on the use of transformation matrices is abandoned in
favor of a new method which employs vectors as such, without their representation
in a particular reference system. The principal tool used to represent rotations of
position vectors and their derivatives is the Euler formula [16].
All the main topics of the theory of gear generation with fixed axes and constant
gear ratio are presented taking a fresh approach. The issues investigated range
from the basic definitions of the surface of action, the equation of meshing and
the envelope surface to the analysis of the undercutting condition and the study
of existence and regularity of the envelope of the contact lines on the tool surface.
Also the kinematic analysis of the generating process is presented within the
same invariant framework.
The result is a compact and clearer formulation where robust definitions of the
different geometrical entities are given.
To assess the ease of use of the proposed approach and to show the insights it
gives, a numerical application to a real spiral bevel gear set is presented.

2.1 Generating surface


In the Euclidean space E3e we define the generating tool to be a regular surface
Σe . Its generic point will be denoted by Pe (ξ, θ), with (ξ, θ) ∈ A, where A is an
open set of R2 .
Once a fixed point Oe in E3e has been selected, it is possible to associate, as
8 2. Generation with fixed axes and constant gear ratio

usual, to each point Pe a position vector pe


pe (ξ, θ) = Pe (ξ, θ) − Oe (2.1)
It is worth noting that the position vectors pe , like all vectors, belong to the linear
space R3 , while the points Pe belong, as already stated, to the Euclidean (i.e.,
affine) space E3e .
By definition, the normal vector me to Σe is given by
me (ξ, θ) = pe ,ξ × pe ,θ (2.2)
Owing to the assumed regularity of the surface Σe we always have
me 6= 0 (2.3)
We see that there has been no need of a reference system to define the gener-
ating surface and its normal vector.

2.2 Rotating vectors


We will employ extensively rotations of position vectors and their derivatives.
Therefore we briefly outline some of their general properties.
Let us consider an axis a, that is a directed straight line in an Euclidean space
E3 . Typically, an axis is defined by means of one of its points O and a unit vector
a that marks its direction.
We introduce the compact notation R(p, a, α̂) to denote the rigid rotation of
a position vector p = P − O around the unit vector a by an angle α̂ (positive
if counterclockwise), that is the rotation of point P around axis a. The result of
such rotation is the position vector

p̂(ξ, θ; α̂) = P̂ − O = R p(ξ, θ), a, α̂
= (p · a)a + [p − (p · a)a] cos α̂ + a × [p − (p · a)a] sin α̂ (2.4)

where P̂ is the image of P after the rotation as shown in Fig. 2.1.


According to [10], equation (2.4) should be called Euler finite rotation formula,
first published in [16], although it is often referred to as Euler–Rodrigues formula.
In [73], the vectorial notation employed to express (2.4) (see Eqn. (11.53)) is im-
mediately abandoned in favor of the matrix notation, considered more comfortable
for later developments.
Equation (2.4) is easy to demonstrate. With reference to Fig. 2.1 we can write
p̂ = OP̂ = OQ + QN + N P̂ = u0 + u1 + u2 (2.5)
Since
OP = p, OQ = (p · a)a, QP = OP − OQ = p − (p · a)a, (2.6)
2.2 Rotating vectors 9

a
a Q
Q
N P a
P u1
u0
p
p
c N u2
P
P
O

Figure 2.1: Rotation of p about a through α̂.

vector u1 = QN , which is parallel to QP , is indeed


 
u1 = QN = QP cos α̂ = p − (p · a)a cos α̂ (2.7)

and u2 = N P̂ has the following expression


 
u2 = N P̂ = a × p − (p · a)a sin α̂ (2.8)
because it is perpendicular to both p − (p · a)a and a and has length |QP | sin α̂.
Therefore we have
p̂ = u0 + u1 + u2 (2.9)
= (p · a)a + [p − (p · a)a] cos α̂ + a × [p − (p · a)a] sin α̂ (2.10)
which is identical to (2.4).
Several properties follow at once from equation (2.4); in particular let
û = R(u, a, α̂), v̂ = R(v, a, α̂), ŵ = R(w, a, α̂)
we immediately have the following relations
û + v̂ = R(u + v, a, α̂) (2.11)
û · v̂ = u · v (2.12)
v̂ × ŵ = R(v × w, a, α̂) (2.13)
 
û · (v̂ × ŵ) = u · (v × w) = u v w (2.14)
10 2. Generation with fixed axes and constant gear ratio

From (2.13), being obviously a = R(a, a, α̂), we also obtain

a × R(w, a, α̂) = R(a × w, a, α̂) (2.15)

The derivatives of the rotated vector p̂ with respect to the parameters ξ and θ
are given by

p̂,ξ (ξ, θ; α̂) = R p,ξ (ξ, θ), a, α̂

p̂,θ (ξ, θ; α̂) = R p,θ (ξ, θ), a, α̂ (2.16)

whereas the derivative with respect to α̂ (rigid rotation) is


 
p̂,α̂ (ξ, θ; α̂) = R p(ξ, θ), a, α̂ ,α̂ = a × R p(ξ, θ), a, α̂ = a × p̂(ξ, θ; α̂) (2.17)

In a more general case, where both the initial vector and the rotation depend on
the same parameter φ, like p̂(ξ, θ, φ) = R(p(ξ, θ, φ), a, φ), a composition of the
previous results is required

p̂,φ (ξ, θ, φ) = R p,φ (ξ, θ, φ), a, φ + a × p̂(ξ, θ, φ) (2.18)

2.3 A first family of surfaces


Let us consider another Euclidean space E3f . In this new space we define a first
fixed axis a (i.e., a directed straight line) by means of one of its points Oa and a
unit vector a. Similarly, we define a second fixed axis b again by means of one of
its points Ob and a unit vector b. In practical terms they are the two axes of the
gear pair under investigation (the generating tool and the to be generated gear).
To each point P of E3f we may associate the following two position vectors
of R3
pa = P − Oa and pb = P − Ob (2.19)
which are related as follows
pb = pa − dba (2.20)
where
dba = Ob − Oa (2.21)
As already stated, Oa and Ob can be taken anywhere on the corresponding
axes. However, when a and b are skew axes, we may take the two points Oa and
Ob on the line of shortest distance between them, so that
d
dba = (a × b) (2.22)
sin γ
where d is the signed distance between the two axes and γ is the angle between a
and b. In case of parallel axes, a = ±b and Oa and Ob may be taken on one of
2.3 A first family of surfaces 11

the infinite lines of shortest distance between a and b, while in case of intersecting
axes it is convenient to take Oa = Ob at the common point.
In the fixed space E3f we define a regular surface Σ̂(ψ̂) isomorphic to the surface
Σe and rigidly rotating around the first axis a by an angle ψ̂ (as usual a counter-
clockwise rotation is taken as positive). Typically we may have Σe = Σ̂(0). In
gear generation Σ̂(ψ̂) is the moving surface of the generating tool.
Denoting by P̂ the generic point of Σ̂, its position vector p̂a can be given by
(cf. (2.4)) 
p̂a (ξ, θ; ψ̂) = P̂ (ξ, θ; ψ̂) − Oa = R pe (ξ, θ), a, ψ̂ (2.23)
where Oa = Oe and pe (ξ, θ) was defined in equation (2.1).
Of course, it is equally possible to employ the position vectors p̂b (ξ, θ; ψ̂) given
by
p̂b (ξ, θ; ψ̂) = P̂ (ξ, θ; ψ̂) − Ob = p̂a (ξ, θ; ψ̂) − dba (2.24)
where dba is constant. It should be noted that the relationship between pe and p̂b
is not a rotation.
Both position vectors p̂a (ξ, θ; ψ̂) and p̂b (ξ, θ; ψ̂) of R3 describe the same family
of surfaces Φf in the space E3f , that is a sequence of regular surfaces Σ̂(ψ̂), one for
each value of ψ̂. It is a family of a special kind, since it only involves rigid-body
rotations of a given surface around a fixed axis a. To avoid any misunderstanding,
it is worth mentioning that Φf is not the family of surfaces whose envelope we are
looking for.
As a consequence of equations (2.23) and (2.24), along with the general prop-
erties (2.16) and (2.17), we have that

p̂a ,ξ = p̂b ,ξ = R(pe ,ξ , a, ψ̂)

p̂a ,θ = p̂b ,θ = R(pe ,θ , a, ψ̂) (2.25)


p̂a ,ψ̂ = p̂b ,ψ̂ = a × p̂a = a × R(pe , a, ψ̂) = R(a × pe , a, ψ̂) (2.26)

By definition the (rotating) normal vector m̂ to each regular surface Σ̂ of the


family is given by
m̂(ξ, θ; ψ̂) = p̂a ,ξ × p̂a ,θ = R(me (ξ, θ), a, ψ̂) 6= 0 (2.27)
where equations (2.2), (2.3), (2.16) and (2.13) were employed. We see from the
last result that m̂ is never zero, thus confirming the assumed regularity of each
surface of the family.
From equations (2.27) and (2.26) it is easy to show that (rigid-body rotation)
m̂,ψ̂ = a × m̂ (2.28)

A relationship similar to equations (19) and (21) can be found at page 69,
eqn. (95), in [84], although obtained by a kinematic approach, i.e., by means of
time derivatives.
12 2. Generation with fixed axes and constant gear ratio

2.4 The enveloping family of surfaces


2.4.1 Preliminary definitions
Let us consider yet another Euclidean space E3g . In this new space we define an
axis b by means of one of its points Og = Ob and a unit vector b. Axis b is the
same already introduced at the beginning of Sec. 2.3. In applications, E3g is the
space where the gear surface will be defined as a fixed one.
In R3 we define a new vector function p̃g by imposing to p̂b a rotation around
b by an angle ϕ̂ (as usual, positive if counterclockwise)
 
p̃g (ξ, θ, ψ̂; ϕ̂) = R p̂b (ξ, θ; ψ̂), b, ϕ̂ = R p̂a (ξ, θ; ψ̂) − dba , b, ϕ̂ (2.29)

2.4.2 Gear ratio and parameter of motion


The two angles ψ̂ and ϕ̂ have been introduced as independent quantities; however,
in gear generation they are related to the parameter of motion φ

ψ̂ = φ/τ = ηφ, ϕ̂ = −φ (2.30)

where τ = 1/η is the (signed) gear ratio. The condition ϕ̂ = −φ means that (2.29)
and (2.30) define a rotation opposite to the one of the gear in order to generate
the gear surface as it were fixed.
A positive value of the gear ratio τ means that a positive rotation ηφ = φ/τ
of the pinion around axis a yields a positive rotation φ of the gear around b.
For instance, in case of transmission between parallel axes with a = b, a pair of
external gears has τ < 0.
It could be easily proven that the screw axis of relative motion between the
two gears is directed like c = ηa − b.

2.4.3 The enveloping family of surfaces


It is now a simple matter to obtain the family of surfaces Φg in E3g whose envelope
we are interested in. If Pg denotes the generic point of Φg , we have from equations
(2.29) and (2.30) that the corresponding position vectors pg = Pg − Og of R3 are
given by any of the following expressions

pg (ξ, θ, φ) = p̃g (ξ, θ, ηφ; −φ)



= R p̂b (ξ, θ; ηφ), b, −φ
 
= R R pe (ξ, θ), a, ηφ − dba , b, −φ (2.31)

This definition as well only involves vectors and therefore does not require reference
systems.
2.4 The enveloping family of surfaces 13

2.4.4 Derivatives of the enveloping family of surfaces


Combining equations (2.25) and (2.31) and the general properties (2.16) we imme-
diately obtain that the derivatives with respect to ξ and θ of the (position vectors
associated to the) family of surfaces Φg are given by the following expressions
 
pg ,ξ = R p̂b ,ξ (ξ, θ; ηφ), b, −φ = R R(pe ,ξ (ξ, θ), a, ηφ), b, −φ
 
pg ,θ = R p̂b ,θ (ξ, θ; ηφ), b, −φ = R R(pe ,θ (ξ, θ), a, ηφ), b, −φ (2.32)

These results clearly show the relationship between the derivatives of the two
families of surfaces Φg and Φf and also of the generating surface Σe .
According to equations (2.2), (2.27) and (2.32), the normal vector mg to each
regular surface of the family Φg is given by

mg (ξ, θ, φ) = pg ,ξ × pg ,θ

= R m̂(ξ, θ; ηφ), b, −φ

= R R(me (ξ, θ), a, ηφ), b, −φ (2.33)

which also shows the relationship between the normal vectors defined so far.
The derivative of pg with respect to the parameter of motion φ can be obtained
as in eqn. (2.18)
 
pg ,φ (ξ, θ, φ) = R p̂b (ξ, θ; ηφ), b, −φ ,φ

= R p̂b ,φ (ξ, θ; ηφ), b, −φ − b × pg

= R ηa × p̂a (ξ, θ; ηφ) − b × p̂b (ξ, θ; ηφ), b, −φ (2.34)

where the last step comes from equations (2.26) and (2.31).
If we define the vector function

ĥ(ξ, θ, ψ̂) = ηa × p̂a (ξ, θ; ψ̂) − b × p̂b (ξ, θ; ψ̂)


= (ηa − b) × p̂a (ξ, θ; ψ̂) + b × dba
= c × p̂a (ξ, θ; ψ̂) + b × dba (2.35)

the derivative pg ,φ can be simply written as



pg ,φ (ξ, θ, φ) = R ĥ(ξ, θ, ηφ), b, −φ (2.36)

Quite useful may turn out also the alternative expression, holding for ψ̂ = ηφ

ĥ(ξ, θ, ηφ) = R(ηa × pe , a, ηφ) − R(b × pg , b, φ) (2.37)

which follows directly from equations (2.31) in the form p̂b = R(pg , b, φ) and (2.35).
14 2. Generation with fixed axes and constant gear ratio

2.5 Equation of meshing


2.5.1 Classical definition
Since in gear generation we are looking for the envelope Γg of the family of surfaces
Φg in the Euclidean space E3g , the equation of meshing f = 0 must involve the
triple product of the partial derivatives of the position vectors pg (ξ, θ, φ)
h i
pg ,ξ pg ,θ pg ,φ = mg · pg ,φ = f (ξ, θ, φ) = 0 (2.38)

with (ξ, θ, φ) ∈ B ⊂ R3 .
For the equation of meshing to be satisfied, the three vectors in (2.38) must
belong to the same two-dimensional sub-space (i.e, the same plane), that is to be
such that
pg ,φ = α pg ,ξ + β pg ,θ (2.39)

for suitable coefficients α = α(ξ, θ, ηφ) and β = β(ξ, θ, ηφ) as better explained in
the following subsection.
Unfortunately, the equation of meshing is usually defined, at least initially
(as in [48], pages 167–168, in [49], page 108, or in [50], page 3), in terms of the
components r2 (ξ, θ, φ) of pg in a Cartesian reference system, say S2 , fixed in E3g
h i
r2 ,ξ r2 ,θ r2 ,φ = f (ξ, θ, φ) = 0 (2.40)

We wish to point out here again that the reference systems are by no means
necessary and it is our goal to show that they play no role in the theoretical
treatment. In other words, all the results expressed in vector form hold true
regardless of the reference system employed.
For instance, providing the equation of meshing in the form (2.40) is somehow
misleading, since it emphasizes the role of S2 while any other reference system
would do as well. The key point is selecting the family of surfaces whose envelope
we are interested in, while the reference system is just an aid and therefore is
completely arbitrary.

2.5.2 Simplified form of equation of meshing


It is now possible to considerably simplify the equation of meshing (2.38) by taking
into account equations (2.32) and (2.36), along with the general result (2.14)
h i
f (ξ, θ, φ) = R(p̂b ,ξ , b, −φ) R(p̂b ,θ , b, −φ) R(ĥ, b, −φ)
h i h i
= p̂b ,ξ p̂b ,θ ĥ = p̂a ,ξ p̂a ,θ ĥ

= m̂ · ĥ = 0 (2.41)
2.5 Equation of meshing 15

with ψ̂ = ηφ in all vector functions. This formulation of the equation of meshing


basically corresponds to the so-called “engineering approach”, as introduced in
[49], page 109, although it has been obtained here in quite a different way. For
instance we do not need any kinematical concept like the relative velocity.
Obviously, the three definitions (2.38), (2.40) and (2.41) provide the same
equation of meshing. However, the last one is simpler than (2.38), since it fully
exploits the rigid body motions involved in the generation of gears. The major
simplification is that we only employ the position vectors p̂a or p̂b , that is the
fixed family of surfaces Φf . Moreover, there are no reference systems involved and
thus it is a more general statement than (2.40).
We wish also to point out that the equation of meshing given here does not
need the concept of “relative differentiation”, introduced in [84], page 48.
Other forms of the equation of meshing follow immediately from (2.35) and
(2.41)
h i h i
f (ξ, θ, φ) = η m̂ a p̂a − m̂ b p̂b

= m̂ · [(ηa − b) × p̂a + b × dba ]


= m̂ · (c × p̂a + b × dba )
h i h i
= m̂ c p̂a + m̂ b dba = 0 (2.42)

Exactly like in (2.39) for the classical equation of meshing, the three vectors
in (2.41) must stay on the same plane to have f = 0, that is be such that

ĥ = α p̂a ,ξ + β p̂a ,θ (2.43)

where α and β are the same of equation (2.39).


It may be noteworthy underlining that the most general form for ĥ is

ĥ(ξ, θ, ψ̂) = α p̂a ,ξ (ξ, θ; ψ̂) + β p̂a ,θ (ξ, θ; ψ̂) + γ m̂(ξ, θ; ψ̂) (2.44)

that reduces to (2.43) at points where f = 0 holds. In any case to obtain α(ξ, θ, ψ̂)
and β(ξ, θ, ψ̂) just consider that

[(α p̂a ,ξ + β p̂a ,θ + γ m̂) × p̂a ,θ ] · m̂ = α m̂ · m̂

which provides h i
ĥ p̂a ,θ m̂
α(ξ, θ, ψ̂) = (2.45)
m̂ · m̂
in a similar way h i
p̂a ,ξ ĥ m̂
β(ξ, θ, ψ̂) = (2.46)
m̂ · m̂
16 2. Generation with fixed axes and constant gear ratio

and simply
ĥ · m̂
γ(ξ, θ, ψ̂) = (2.47)
m̂ · m̂
When α and β are calculated for ψ̂ = ηφ, they are the same of equation (2.39),
since all vectors in eqn. (2.43) are simply rotated with respect to those in (2.39).

2.5.3 Another simplified form of the equation of meshing


The ideas employed to obtain the simplified equation of meshing (2.41) can be
further pursued. If we define

he (ξ, θ, ψ̂) = R ĥ(ξ, θ, ψ̂), a, −ψ̂

= ηa − R(b, a, −ψ̂) × pe (ξ, θ) + R(b × dba , a, −ψ̂)
= R(c, a, −ψ̂) × pe (ξ, θ) + R(b × dba , a, −ψ̂) (2.48)
and consider equation (2.25), we obtain
h i h i
f (ξ, θ, φ) = p̂a ,ξ p̂a ,θ ĥ = pe ,ξ (ξ, θ) pe ,θ (ξ, θ) he (ξ, θ, ηφ)

= me (ξ, θ) · he (ξ, θ, ηφ) = 0 (2.49)


which is another possible form of the equation of meshing, with the parameter of
motion φ only appearing in the last vector.
From equations (2.44) and (2.48), another expression for he (ξ, θ, ψ̂) is easily
obtained
he (ξ, θ, ψ̂) = α pe ,ξ (ξ, θ) + β pe ,θ (ξ, θ) + γ me (ξ, θ) (2.50)
from which it is also possible to obtain α(ξ, θ, ψ̂), β(ξ, θ, ψ̂) and γ(ξ, θ, ψ̂) in an
alternative way, that is
h i
he pe ,θ me
α(ξ, θ, ψ̂) =
me · me
h i
pe ,ξ he me
β(ξ, θ, ψ̂) =
me · me
he · me
γ(ξ, θ, ψ̂) = (2.51)
me · me
Summing up we have obtained, in a systematic way and without any refer-
ence system, that the equation of meshing is given by any of the following scalar
products
f (ξ, θ, φ) = mg (ξ, θ, φ) · pg ,φ (ξ, θ, φ)
= m̂(ξ, θ; ηφ) · ĥ(ξ, θ, ηφ)
= me (ξ, θ) · he (ξ, θ, ηφ) = 0 (2.52)
2.6 The envelope surface Γg 17

where the last two are simpler because fully exploit the rigid body motions involved
in gear generation.

2.6 The envelope surface Γg


As well known [48, 49], the family of surfaces Φg (defined in (2.31)) with the
equation of meshing (2.52) may define the sought for envelope surface Γg (of E3g )
with position vectors sg in R3
( 
sg = pg ξ, θ, φ
(2.53)
f (ξ, θ, φ) = 0

Actually, the fulfillment of the equation of meshing is only a necessary condition


for the existence on an envelope surface.

2.6.1 Explicit definition of the envelope surface (sufficient


condition)
According to Dini’s theorem on implicit functions, the condition

f,ξ (ξ, θ, φ) 6= 0 (2.54)

assuming f ∈ C1 , guarantees the local existence of the explicit function

ξ = ξ(θ, φ) (2.55)

which satisfies the equation of meshing. Moreover it has partial derivatives

∂ξ f, ∂ξ f,φ
= − θ, and =− (2.56)
∂θ f,ξ ∂φ f,ξ

Of course, in (2.56) f,θ means the following function of θ and φ



∂f (ξ, θ, φ)
f,θ =
∂θ
ξ=ξ(θ,φ)

where the derivative has to be done before inserting (2.55).


Analogous formulas hold for the other partial derivatives in (2.56).
From now on we will employ ξ = ξ(θ, φ), but θ = θ(ξ, φ) would be another
possible choice whenever f,θ (ξ, θ, φ) 6= 0. Therefore, we have to assume that

|f,ξ (ξ, θ, φ)| + |f,θ (ξ, θ, φ)| =


6 0 (2.57)

On the other hand φ = φ(ξ, θ) cannot be employed in general since, as we will see
in (2.75), we have to allow the possibility for f,φ (ξ, θ, φ) = 0.
18 2. Generation with fixed axes and constant gear ratio

Condition (2.57), along with the assumption f ∈ C1 , is precisely the sufficient


condition for the envelope process to generate a surface Γg . Indeed, we see that
if, e.g., (2.54) holds true, it is possible to define the envelope surface Γg by means
of position vectors sg given by the explicit function

sg (θ, φ) = pg ξ(θ, φ), θ, φ (2.58)

which employs equation (2.55).


It is worth noting that in the unique linear space R3 where all vectors are
defined, the envelope surface sg is fixed (exactly like pe ).

2.7 Regularity of the envelope surface Γg


The regularity of a surface is guaranteed wherever its normal vector does not
vanish. Points where the surface is not regular are said to form lines of singular
points.
According to (2.58), the normal vector ng (θ, φ) to Γg is, by definition, given
by
ng (θ, φ) = sg ,θ × sg ,φ (2.59)
where, employing Dini’s theorem

f,θ f,φ
sg ,θ = − p , + pg ,θ and sg ,φ = − p , + pg ,φ (2.60)
f,ξ g ξ f,ξ g ξ

Therefore, the normal vector ng can be expressed as



1     
ng (θ, φ) = pg ,ξ × pg ,θ f,φ + pg ,φ × pg ,ξ f,θ + pg ,θ × pg ,φ f,ξ
f,ξ ξ=ξ(θ,φ)

1 
= f,φ − α f,ξ − β f,θ mg
f,ξ ξ=ξ(θ,φ)

1

= (f,φ − α f,ξ − β f,θ ) R(m̂, b, −φ) ξ=ξ(θ,φ)
f,ξ ψ̂=ηφ
1
= (f,φ − α f,ξ − β f,θ ) R(R(me , a, ηφ), b, −φ) ξ=ξ(θ,φ) (2.61)
f,ξ

where in the last expressions the geometric meaning (2.39) of the equation of
meshing was taken into account and the last step comes directly from (2.33).
These expressions should be compared with those obtained in [48].
Since me 6= 0, we see that the vector condition for the regularity of the envelope
surface Γg
ng (θ, φ) 6= 0 (2.62)
2.7 Regularity of the envelope surface Γg 19

is equivalent indeed to the scalar inequality


 
g(ξ, θ, φ) = f,φ − α f,ξ − β f,θ 6= 0 (2.63)

with ξ = ξ(θ, φ). Coefficients α(ξ, θ, ηφ) and β(ξ, θ, ηφ) are the functions intro-
duced in equation (2.39) and obtained in (2.45) and (2.46).
It may be useful, as sometimes found in the literature, to express the scalar
function g(ξ, θ, φ) by means of vectors. To this purpose we introduce a new vector
  
q̂(ξ, θ, φ; ψ̂) = p̂a ,ξ × p̂a ,θ f,φ + ĥ × p̂a ,ξ f,θ + p̂a ,θ × ĥ f,ξ (2.64)

so that, according to equations (2.32) and (2.35) we may write eqn. (2.61) in a
more compact way

1 1
ng (θ, φ) = R(q̂, b, −φ) = g R(m̂, b, −φ) (2.65)
f,ξ f,ξ

which provides an alternative expression to compute g(ξ, θ, φ)

q̂ · m̂
g(ξ, θ, φ) = f,φ − α f,ξ − β f,θ = (2.66)
m̂ · m̂

where all functions are evaluated at ψ̂ = ηφ. The following dot product, expanded
using the Lagrange identity,

p̂ , · p̂ , p̂a ,ξ · p̂a ,θ p̂a ,ξ · ĥ
aξ aξ

q̂(ξ, θ, φ; ηφ) · m̂(ξ, θ; ηφ) = p̂a ,θ · p̂a ,ξ p̂a ,θ · p̂a ,θ p̂a ,θ · ĥ (2.67)

f, ξ f, θ f, φ

is often encountered in the literature. In [84, p. 57] it is called the limit function
of the first kind. However, in [84] it is obtained employing the concept of relative
differentiation which plays no role in the present paper. In [50, §1.4] two equivalent
conditions for singular points are obtained (namely g1 = 0 and g2 = 0). They are
in the form of (2.39) but written employing components, thus showing that the
use of reference systems may lead to a proliferation of formulas for the very same
concept.
Even in this case we may take advantage of rotating vectors and property
(2.13), writing

q̂(ξ, θ, φ; ηφ) · m̂(ξ, θ; ηφ) = q e (ξ, θ, φ; ηφ) · me (ξ, θ)

where
  
q e (ξ, θ, φ; ψ̂) = pe ,ξ × pe ,θ f,φ + he × pe ,ξ f,θ + pe ,θ × he f,ξ (2.68)
20 2. Generation with fixed axes and constant gear ratio

It is now easy to provide what is perhaps the most general definition for lines
of singular points lg on the generated surface Γg :

lg = pg (ξ, θ, φ)

f (ξ, θ, φ) = 0 (2.69)


g(ξ, θ, φ) = 0

which collects the results obtained in (2.31), (2.52) and (2.66). All equations are
valid regardless of the reference system actually employed. As already mentioned,
the same problem was addressed, e.g., in [50], §1.4 and the final results are basi-
cally the same. However, the approach here proposed provides maybe a clearer
framework and a more systematic treatment.
It will be shown in Sec. 2.13 that at points of no sliding we have ĥ = 0.
Therefore, at the same points, condition g = 0 simply becomes f,φ = 0 (which is
called limit function of the second kind in [84]) since q̂ = m̂f,φ .

2.8 The truly enveloping surface Γe


By means of equation (2.1) and the equation of meshing we can now define, back
in E3e of Sec. 2.1, the truly enveloping surface Γe , that is that part of Σe which
actually comes in contact with the gear to be generated
( 
se = pe ξ, θ
(2.70)
f (ξ, θ, φ) = 0

If from the equation of meshing we can obtain an explicit function like, e.g.,
(2.55), the truly enveloping surface can be defined in R3 by means of position
vectors 
se (θ, φ) = pe ξ(θ, φ), θ (2.71)
The two surfaces (i.e., the sets of points) Σe and Γe are such that Γe ⊆ Σe . In R3
the surface se (θ, φ) is fixed.
This new surface Γe has, by definition, normal ne (θ, φ)

ne (θ, φ) = se ,θ × se ,φ (2.72)

However, according to (2.71) and Dini’s theorem (2.55), we also have


 
∂ξ ∂ξ f,φ   f,
φ
ne (θ, φ) = pe ,θ + pe ,ξ × pe ,ξ = pe ,ξ × pe ,θ = me (2.73)
∂θ ∂φ f,ξ f,ξ

where, after the derivatives have been performed (i.e., in the final expressions) all
occurrences of ξ must be replaced by ξ(θ, φ). Equation (2.73) clearly shows the
link between the normal vectors me (ξ(θ, φ), θ) and ne (θ, φ). Incidentally, we note
that se ,φ and pe ,ξ are parallel vectors.
2.9 The surface of action Γf 21

We can also investigate under which conditions the new surface Γe may have
singular points, that is points where the normal vector ne becomes zero. As it
immediately arises from equations (2.2) and (2.73), ne = 0 whenever

f,φ ξ(θ, φ), θ, φ = 0 (2.74)
which defines an implicit function between θ and φ. Therefore, differently from
Σe , this new surface Γe may have (lines of) singular points. It will be shown in
Sec. 2.12 that they are also the envelope of contact lines on Γe itself.
In more general terms, the line le of singular points on Γe is fully defined by
 
le = pe ξ, θ

f (ξ, θ, φ) = 0 (2.75)


f,φ (ξ, θ, φ) = 0
which can also be found in [84, p. 58].

2.9 The surface of action Γf


By consideration of the equation of meshing, e.g. in the form ξ = ξ(θ, φ), along
with the condition ψ̂ = ηφ, it is possible to define in E3f the surface of action Γf .
Its points are given in R3 by any of the two position vectors
sf a (θ, φ) = p̂a (ξ(θ, φ), θ, ηφ) = R pe (ξ(θ, φ), θ), a, ηφ) and

sf b (θ, φ) = p̂b (ξ(θ, φ), θ, ηφ) = R pg (ξ(θ, φ), θ, φ), b, φ
= sf a (θ, φ) − dba (2.76)
with respect to Oa and Ob .
The normal vector nf to the surface of action Γf is, by definition, given by
nf (θ, φ) = sf a ,θ × sf a ,φ = sf b ,θ × sf b ,φ (2.77)
where, according to (2.56) and (2.76)

1 
sf a ,θ = sf b ,θ = − p̂ , f, − p̂a ,θ f,ξ ξ=ξ(θ,φ) (2.78)
f,ξ a ξ θ ψ̂=ηφ

and
1 
sf a ,φ = sf b ,φ = − p̂ , f, − η p̂a ,ψ̂ f,ξ ξ=ξ(θ,φ)
f,ξ a ξ φ ψ̂=ηφ
Therefore, nf (θ, φ) can be expressed as

1  
nf (θ, φ) = (p̂a ,ξ × p̂a ,θ )f,φ + η(p̂a ,ψ̂ × p̂a ,ξ )f,θ + η(p̂a ,θ × p̂a ,ψ̂ )f,ξ ξ=ξ(θ,φ)
f,ξ ψ̂=ηφ
(2.79)
22 2. Generation with fixed axes and constant gear ratio

Owing to equations (2.26), (2.27), (2.35), (2.43) and (2.63), the normal vector
nf (θ, φ) to the surface of action is also equal to any of the following expressions

f,φ
nf (θ, φ) = m̂ − (ηa × p̂a ) × t
f,ξ
= R(ne , a, ηφ) − (ηa × p̂a ) × t
g
= m̂ − (b × p̂b ) × t
f,ξ
= R(ng , b, φ) − (b × p̂b ) × t (2.80)

where
1 
t(θ, φ) = sf a ,θ (θ, φ) = − p̂ , f, − p̂a ,θ f,ξ ξ=ξ(θ,φ) (2.81)
f,ξ a ξ θ ψ̂=ηφ

is the tangent vector to the contact line sf a (θ, φ), with fixed φ. From equations
(2.27) and (2.57) we clearly see that t is never zero. This formula for t corresponds
to equations (1.6.9) and (1.6.10) in [50].

2.9.1 A further note on singular points of the envelope sur-


face Γg
Quite interestingly, whenever ng (θ, φ) = 0, i.e. when g = 0, we obtain from
eqn. (2.80) (with ξ = ξ(θ, φ) and ψ̂ = ηφ)

nf (θ, φ) = − b × p̂b × t (2.82)

which states that, no matter t, at singular points for Γg the normal nf to the
contact surface, when applied in p̂b , goes through axis b. More precisely, we have,
with ψ̂ = ηφ h i
nf b p̂b = 0 (2.83)

which means that the three vectors belong to the same two-dimensional sub-space.
Therefore, a line through the point P̂ = Ob + p̂b directed like nf necessarily
intersect the line through Ob with direction b (unless nf and b are parallel).
Equation (2.82) corresponds to eqn. (17) in [52], where, however, reference
systems are deeply rooted in the procedure.
Also of interest might be to observe that, if f,φ = 0, we obtain from eqn. (2.80)

nf (θ, φ) = − ηa × p̂a × t (2.84)

and hence h i
nf a p̂a = 0 (2.85)
2.10 Rotating surfaces and contact lines 23

2.10 Rotating surfaces and contact lines


Summing up, we have obtained three relevant surfaces:
• the envelope (generated) surface Γg ∈ E3g , defined by position vectors sg (θ, φ) =
Pg − Og , with Og = Ob (eqn. (2.58));
• the truly enveloping (generating) surface Γe ∈ E3e , with position vectors
se (θ, φ) = Pe − Oe , with Oe = Oa (eqn. (2.71));
• the surface of action Γf ∈ E3f , with position vectors sf a (θ, φ) = P − Oa and
sf b (θ, φ) = P − Ob (eqn. (2.76)),
where φ is the parameter of motion. Of each surface we have obtained the normal
vector and discussed the possible occurrence of singular points.
Actually, in the fixed space E3f we also see two moving surfaces. Surface Γ̂e (ψ̂)
of the tool that rigidly rotates around axis a by an angle ψ̂, and surface Γ̂g (ν̂)
of the gear that rigidly rotates around axis b by an angle ν̂. Surface Γ̂e (ψ̂) is
described in R3 by the following position vectors

ŝea (θ, φ; ψ̂) = R(se (θ, φ), a, ψ̂), with respect to Oa


ŝeb (θ, φ; ψ̂) = ŝea (θ, φ; ψ̂) − dba , with respect to Ob (2.86)

Similarly, surface Γ̂g (ν̂) is described by the following position vectors

ŝgb (θ, φ; ν̂) = R(sg (θ, φ), b, ν̂), with respect to Ob


ŝga (θ, φ; ν̂) = ŝgb (θ, φ; ν̂) + dba , with respect to Oa (2.87)

During meshing and according to (2.30), the two rotations are related and we
have ψ̂ = ηφ and ν̂ = φ. The surface of action Γf was already in E3f and stays
fixed. For a fixed value of φ, the three surfaces Γ̂e (ηφ), Γf and Γ̂g (φ) touch each
other along the contact line, with position vectors

ca (θ, φ) = R(se (θ, φ), a, ηφ) = sf a (θ, φ) = R(sg (θ, φ), b, φ) + dba and

cb (θ, φ) = ca (θ, φ) − dba (2.88)

with respect to Oa and Ob . More precisely, Γ̂e (ηφ) and Γ̂g (φ) are tangent while
intersecting Γf along the contact line. The tangent vector t to each contact line
was obtained in equation (2.81).
On the other hand, contact lines on each surface Γe and Γg are obtained by
simply fixing φ in the corresponding position vectors se (θ, φ) and sg (θ, φ), respec-
tively. Therefore, we have on each surface a family of curves with parameter φ.
It is interesting to determine under which conditions these contact lines may have
an envelope on the surface itself.
24 2. Generation with fixed axes and constant gear ratio

2.11 Envelope of contact lines on the envelope


surface Γg
From the general theory of envelope of curves, the necessary condition for the
existence of the envelope of the family of curves sg (θ, φ) (contact lines), with
parameter φ, on the surface Γg is
sg ,θ × sg ,φ = 0 (2.89)
which, according to equation (2.59), corresponds to ng = 0. By taking into
account the analysis presented in Section 2.7, we see that condition (2.89) can be
given as the system of equations
(
f (ξ, θ, φ) = 0
(2.90)
g(ξ, θ, φ) = f,φ − α f,ξ − β f,θ = 0

like in (2.69). Therefore, the necessary condition for the existence of the envelope
of contact lines on Γg exactly corresponds to the necessary condition for the exis-
tence of lines of singular points on the same surface. This line lg of E3g was defined
by equations (2.69).

2.11.1 The parameter space viewpoint


The system of equation (2.90) defines a curve in the two-dimensional parameter
space (ξ, θ) which corresponds to a line of singular points on Γg . The parametric
equations of this line can take the form
(
ξ = ξg (φ)
(2.91)
θ = θg (φ)
if, according to Dini’s theorem on implicit functions

D(f, g) f,ξ f,θ
D(ξ, θ) g,ξ g,θ = f,ξ g,θ − f,θ g,ξ 6= 0
= (2.92)

which is therefore the sufficient condition for their existence as functions of φ.


Moreover, the tangent vector to the curve (2.91) has components
 
f,θ f,φ f,ξ f,φ 
tg (φ) = , −
g,ξ g,φ = f,θ g,φ − f,φ g,θ , f,φ g,ξ − f,ξ g,φ ξ=ξg (φ)

g,θ g,φ θ=θ (φ)
g
(2.93)
and hence the curve (2.91) in (ξ, θ) is regular whenever tg =6 0. More precisely we
have  dξ dθ 
g g tg (φ)
, = (2.94)
dφ dφ f,ξ g,θ − f,θ g,ξ
with ξ = ξg (φ) and θ = θg (φ) in the denominator.
2.11 Envelope of contact lines on the envelope surface Γg 25

2.11.2 Regularity of the line lg of singular points


Equations (2.90) are the necessary conditions for the existence of a line of singular
points, as defined in (2.69), on the envelope surface Γg . This line lg is also the
envelope of contact lines on the surface itself if condition (2.92) is fulfilled. As a
matter of fact, condition (2.92) ensures the existence of the two functions (2.91),
and hence that this line lg can be given a parametrization in terms of φ

lg (φ) = pg ξg (φ), θg (φ), φ (2.95)

It is worth noting that inequality (2.92) implies (2.57), but not the other way
around.
The regularity of lg (φ) is equivalent to the condition

lg ,φ 6= 0 (2.96)

which can be formulated through a simple, although a bit long calculation as

(f,ξ g,θ − f,θ g,ξ ) lg ,φ



h i

= pg ,ξ (f,θ g,φ −f,φ g,θ )+pg ,θ (f,φ g,ξ −f,ξ g,φ )+pg ,φ (f,ξ g,θ −f,θ g,ξ ) ξ=ξg (φ) 6= 0
θ=θg (φ)
(2.97)

This result can also be found in [53], eqn. (41).


Moreover, we can take into account that pg ,φ = αpg ,ξ + βpg ,θ and f,φ =
αf,ξ + βf,θ thus obtaining a more compact form



(f,ξ g,θ − f,θ g,ξ ) lg ,φ = (g,φ − αg,ξ − βg,θ )(pg ,ξ f,θ − pg ,θ f,ξ ) ξ=ξg (φ) 6= 0 (2.98)
θ=θg (φ)

The same subject was treated in § 1.7 of [50], but in a fairly different way.
Conditions (2.97) and (2.98) for the regularity of the envelope of contact lines
can be considerably simplified by means again of equations (2.32) and (2.36) with
ξ = ξg (φ), θ = θg (φ) and ψ̂ = ηφ, thus getting

p̂a ,ξ (f,θ g,φ − f,φ g,θ ) + p̂a ,θ (f,φ g,ξ − f,ξ g,φ ) + ĥ(f,ξ g,θ − f,θ g,ξ ) 6= 0 (2.99)

which can also be written as





 
g,φ − αg,ξ − βg,θ p̂a ,ξ f,θ − p̂a ,θ f,ξ ξ=ξg (φ) =
6 0 (2.100)
θ=θg (φ)
ψ̂=ηφ
26 2. Generation with fixed axes and constant gear ratio

Comparing equations (2.81) and (2.100) we obtain, as expected, that contact lines
are tangent to their envelope curve. In [53], eqn. (42), a similar conclusion is
reached, though in a different way.
Moreover, since t is never zero, we see that the regularity condition for the
envelope of contact lines lg can be simply stated as




s(φ) = g,φ − αg,ξ − βg,θ ξ=ξg (φ) 6= 0 (2.101)
θ=θg (φ)
ψ̂=ηφ
According to equations (2.45) and (2.46), we can take a further step and write the
above condition as (cf. (2.67))

p̂ , · p̂ , p̂a ,ξ · p̂a ,θ p̂a ,ξ · ĥ
aξ aξ

(m̂ · m̂) s(φ) = p̂a ,θ · p̂a ,ξ p̂a ,θ · p̂a ,θ p̂a ,θ · ĥ 6= 0 (2.102)

g,ξ g,θ g,φ
or as

p , · p , pe ,ξ · pe ,θ pe ,ξ · he
eξ eξ

(me · me ) s(φ) = pe ,θ · pe ,ξ pe ,θ · pe ,θ pe ,θ · he 6= 0 (2.103)

g,ξ g,θ g,φ

with ξ = ξg (φ), θ = θg (φ) and ψ̂ = ηφ. Basically the same result can also
be found in [53], eqn. (33). However in that paper all developments are based on
vector components and hence need reference systems.

2.12 Envelope of contact lines on the generating


surface Γe
Similar steps provide the necessary condition for the existence of the envelope of
the family of curves se (θ, φ), with parameter φ, on the surface Γe
se ,θ × se ,φ = 0. (2.104)
From equation (2.72) we see that the above condition requires the normal vector
ne (θ, φ) to the surface to be zero, that is the surface to be singular. As already
mentioned and according to eqn. (2.73), ne (θ, φ) = 0 if and only if equation (2.74)
holds, that is f,φ (ξ(θ, φ), θ, φ) = 0, or more generally (cf. [48, § 9.7]
(
f (ξ, θ, φ) = 0
(2.105)
f,φ (ξ, θ, φ) = 0
Therefore the necessary condition for the existence of the envelope of contact lines
le on Γe exactly corresponds to the necessary condition for the existence of lines of
singular points (defined in equation (2.75)). This topic is also discussed in [53], §5.
2.12 Envelope of contact lines on the generating surface Γe 27

2.12.1 Envelope of contact lines in the parameter space


Conditions (2.105) have another interesting interpretation. In the space R2 of the
parametric coordinates (ξ, θ), the equation of meshing f (ξ, θ, φ) = 0 can be inter-
preted as the implicit definition of a family of curves with parameter φ. Therefore,
the fulfillment of the additional equation f,φ (ξ, θ, φ) = 0 provides the necessary
condition for the existence of the envelope of the image of contact lines in the
parameter space (ξ, θ). Basically, equations (2.105) may define in the parameter
space the envelope curve (
ξ = ξe (φ)
(2.106)
θ = θe (φ)
Therefore, we see that the envelope of contact lines on the truly generating surface
Γe has a counterpart envelope curve in the parameter space. This is not surprising
owing to the regularity of the starting surface Σe .
According to Dini’s theorem on implicit functions, for the curve (2.106) to exist
(sufficient condition) we need, along with f ∈ C2

D(f, f,φ ) f,ξ f,θ
D(ξ, θ) f,φξ f,φθ = f,ξ f,φθ − f,θ f,φξ 6= 0
= (2.107)

Its tangent vector has components


 
f,
f,φ = 0 f,ξ f,φ = 0 
te (φ) = θ

, − = f,θ , −f,ξ f,φφ ξ=ξe (φ)
f,φθ f,φφ f,φξ f,φφ θ=θe (φ)
(2.108)
and hence the curve (2.106) in (ξ, θ) is regular whenever te 6= 0, that is

f,φφ ξe (φ), θe (φ), φ 6= 0 (2.109)

More precisely we have


 dξ dθe  te (φ)
e
, = (2.110)
dφ dφ f,ξ f,φθ − f,θ f,φξ

with ξ = ξe (φ) and θ = θe (φ) in the denominator.


The point where this line (ξe (φ), θe (φ)) matches line (ξg (φ), θg (φ)) (eqn. (2.91))
is given by 
f (ξ, θ, φ)
 =0
f,φ (ξ, θ, φ) = 0 (2.111)


g(ξ, θ, φ) =0
These two lines have the same direction at their common point (ξp , θp ): indeed
from equations (2.108) and (2.93) we see that tg is parallel to te when f,φ = 0.
28 2. Generation with fixed axes and constant gear ratio

2.12.2 Regularity of the line le of singular points on the tool


Equations (2.105) are the necessary conditions for the existence of a line of singular
points, as defined in (2.75), on the generating surface Γe . This line le is also the
envelope of contact lines on the surface itself if condition (2.107) is fulfilled. As a
matter of fact, condition (2.107) ensures the existence of the two functions (2.106)
and hence that line le can be given a parametrization in terms of φ

le (φ) = pe ξe (φ), θe (φ) (2.112)
The regularity of the curve le (φ) is equivalent to the condition
le ,φ 6= 0 (2.113)
which, according to equation (2.110), means

 

(f,ξ f,φθ − f,θ f,φξ ) le ,φ = f,φφ pe ,ξ f,θ − pe ,θ f,ξ ξ=ξe (φ) 6= 0 (2.114)
θ=θe (φ)

Owing to the conditions (2.3) and (2.57), we see that the above inequality is
fulfilled if and only if f,φφ 6= 0, exactly like in (2.109). The analysis here presented
is the counterpart of §1.8 in [50].

2.13 Sliding velocity between mating surfaces


It is quite important in many applications to compute the sliding velocity between
the surface Γ̂e of the tool and the surface Γ̂g of the gear at their mating points.
Of course, we have to assume here that the parameter of motion is a function of
time, that is φ = φ(t).
First, let us consider the space E3g where the generated surface Γg is fixed. In
this space a generic point pg (ξ, θ, φ(t)) of the family of surfaces Φg has velocity

v(ξ, θ, φ) = φ̇ pg ,φ (2.115)

The sliding velocity v sg of the shaper with respect to the gear is therefore given
by the above velocity computed at those points where the equation of meshing is
verified, that is a points of contact during the envelope process

s

v g (θ, φ) = v(ξ(θ, φ), θ, φ) = φ̇ pg ,φ = φ̇ R ĥ(ξ(θ, φ), θ, ηφ), b, −φ
ξ=ξ(θ,φ)
(2.116)
where the last step is due to equation (2.36).
To map the sliding velocity v sf in the space E3f , that is on the surface of action
Γf it suffices to apply to v sg a rotation φ around axis b

v sf (θ, φ) = R(v sg , b, φ) = φ̇ ĥ(ξ(θ, φ), θ, ηφ) (2.117)


2.14 Application to spiral bevel gears 29

which also shows the physical meaning of vector ĥ. We have no sliding between
the two mating gears at those points where ĥ(ξ(θ, φ), θ, ηφ) = 0. The same result
can be obtained through a different route. Considering equations (2.86), (2.87)
and also (2.88), with ψ̂(t) = ηφ(t) and ν̂(t) = φ(t) we have
!
s

v f (θ, φ) = φ̇ ηŝea ,ψ̂
− ŝgb ,ν̂
(2.118)
ψ̂=ηφ ν̂=φ

Finally, to obtain the sliding velocity v se in the space E3e where the generating
surface Γe is fixed, we simply have to do (cf. equation (2.48))

v se (θ, φ) = R(v sf , a, −ηφ) = φ̇ R ĥ(ξ(θ, φ), θ, ηφ), a, −ηφ = φ̇ he (ξ(θ, φ), θ, ηφ)
(2.119)

2.14 Application to spiral bevel gears


In order to show the convenience of the proposed approach, we employ it to obtain
the equation of meshing and the parametric equations of the active flanks and root
fillets for a generated (face-milled) spiral bevel gear and compare them with the
results obtained in [4] in a classical way.
In particular, the proposed equations have been employed to study a real trans-
mission for aerospace applications.

2.14.1 Preliminary definitions


According to the notation introduced in Section 2.3, let a be the axis of the machine
cradle (generating tool) and b the axis of the gear blank. As well known, during
generation of a spiral bevel gear a and b can be skew axes. Points Oa and Ob are
not taken in this case on the line of shortest distance, as suggested in eqn. (2.22),
but, according to [4] and also to common practice, they are displaced with respect
to such line. More precisely, point Oa is moved along a by the so-called sliding
base ∆XB2 , while point Ob is moved along b by the machine center to back ∆XD2 ,
as shown in Figure 2.2. Quantity ∆EM2 , still in Figure 2.2, called blank offset, is
indeed the shortest distance between axes a and b.
In order to actually apply the proposed approach (i.e., for computational pur-
poses) we need to define, in the linear space R3 , a unique (fixed) reference frame
S = (O; x, y, z), with unit vectors (i, j, k). It is worth noting that all computations
will be performed using just this single reference system. We take full advantage
of the fact that our analysis is based only on vectors and they always belong to
the same linear space R3 .
For instance we may select k = a and
1 a×b
j= a×b= (2.120)
sin γ |a × b|
30 2. Generation with fixed axes and constant gear ratio

head-cutter
machine
frame

cutter
ye axis

Sr

q2 yq
xe
ψc2
cradle Oe= Oa blank
axis a
xq axis
ψ2
∆E m2 ze
γm2
zq
∆ X b2 O q= O b b
∆ X D2

Figure 2.2: Geometric set up for the generation of spiral bevel gears.

where γ is the angle between a and b. Alternatively to γ the machine root angle
γm2 = π/2 − γ is often employed (Figure 6 in [4]).
In the reference system S the components of axis unit vectors a and b are
a = (0, 0, 1) and b = (cos γm2 , 0, sin γm2 ), respectively, and the components of
dba = Ob − Oa become

dba = (∆XD2 cos γm2 , −∆EM2 , ∆XB2 + ∆XD2 sin γm2 ) (2.121)

It is worth noting that a, b and dba are all fixed vectors and, accordingly, they
have in S constant components.
In this application we consider a generating tool Σe with a straight blade
profile and a circular fillet at the top. The geometric parameters of the tool and
the machine settings will appear here with the same labels as in [4] (see also
Tables 2.1, 2.2 and 2.3 below). The parametric equations in S of the active flanks
2.14 Application to spiral bevel gears 31

(a) (a)
pe (ξ, θ) of the tool and their unit normal vector me (ξ, θ) are
   
(Rg ± ξ sin αg ) cos θ + Sr2 cos q2 cos αg cos θ
me(a) (θ) =  cos αg sin θ  ,
pe(a) (ξ, θ) =  (Rg ± ξ sin αg ) sin θ + Sr2 sin q2  ,
−ξ cos αg ± sin αg
(2.122)
1−sin αg
where ξ ∈ [ρw cos αg , s̄g (θ)], θ ∈ [0, 2 π) and the upper and lower signs refer to
concave and convex side, respectively.
The parametric equations of the fillets at the top of the tool and their unit
normal vectors, also in S, are
   
(Xw ± ρw sin ξ) cos θ + Sr2 cos q2 sin ξ cos θ
p(b)
e (ξ, θ) =
 (Xw ± ρw sin ξ) sin θ + Sr2 sin q2  , m(b)
e (ξ, θ) = sin ξ sin θ
 
−ρw (1 − cos ξ) ± cos ξ
(2.123)
where ξ ∈ [0, π2 −αg ], θ ∈ [0, 2π) and the upper and lower signs refer to the concave
and convex side, respectively.
Again they are all fixed vectors. In [4] equations (2.122) and (2.123) would be
in the reference system Sc2 .

2.14.2 Equation of meshing


To compute the equation of meshing for the active flanks and the fillets of the gear,
it is sufficient to plug the previous expressions in (2.49). For the active flanks we
obtain
f (a) (ξ, θ, φ) = me(a) (θ) · he(a) (ξ, θ, ηφ) = 0, (2.124)
where
he(a) = (η a − R(b, a, −ηφ)) × pe(a) + R(b × dba , a, −ηφ). (2.125)
Equation (2.124) appears already in quite a compact form and it is easy to solve
with respect to the parametric coordinate ξ obtaining
 
ξ (a) (θ, φ) = csc (θ + η φ) sec γm2 cos αg ∆EM2 cos(θ + η φ) sin γm2
 
+ Sr2 (−η + sin γm2 ) sin(q2 − θ) ∓ cos γm2 sin αg (∆EM2

+ Sr2 sin(q2 + η φ) ) ± (∆XB2 cos αg ± Rg sin αg ) sin(θ + η φ) ,
(2.126)

where the upper/lower signs must be chosen accordingly with the equations for
the concave/convex side.
The equation of meshing for the root fillets of the gear is obtained in much the
same way as
f (b) (ξ, θ, φ) = m(b) (b)
e (ξ, θ) · he (ξ, θ, η φ) = 0, (2.127)
32 2. Generation with fixed axes and constant gear ratio

where
h(b) (b) b
e = (η a − R(b, a, −η φ)) × pe + R(b × da , a, −η φ). (2.128)
Equation (2.127) can be solved with respect to the parametric coordinate ξ
obtaining " p #
(b) A(θ, φ) ∓ B(θ, φ)
ξ (θ, φ) = 2 arctan , (2.129)
C(θ, φ)
where the upper/lower signs refer to a right/left handed gear. The coefficients
in (2.129) have the following expressions

A(θ, φ) = 2 −∆EM2 cos(θ + η φ) sin γm2 +Sr2 (η−sin γm2 ) sin(q2 −θ)

+ (∆XB2 +ρw ) cos γm2 sin(θ+η φ) ,
 2
 2
B(θ, φ) = 4 cos γm 2
∆EM2 +Sr2 sin(q2 +η φ)+Xw sin(θ+η φ)
 (2.130)
+ ∆EM2 cos(θ+η ψ2 ) sin γm2 +Sr2 (−η+sin γm2 ) sin(q2 −θ)
2
− (∆XB2 + ρw ) cos γm2 sin(θ + η φ) ,
C(θ, φ) = ±2 cos γm2 [∆EM2 + Sr2 sin(q2 + η φ) + Xw sin(θ + η φ)],

where the upper/lower signs are to be chosen for the concave/convex side.

2.14.3 Gear surface


Employing equations (2.31) and (2.58), the envelope sg for the active flanks and
the root fillets is quite straightforward to compute. The parametric equations in
S for the active flanks are
 
(a) (a) b
sg (θ, φ) = R R(pe (ξ, θ), a, ηφ) − da , b, −φ (2.131)
ξ=ξ (a) (θ,φ)

and for the root fillets are




s(b)
g (θ, φ) = R(R(p(b)
e (ξ, θ), a, η φ) − dba , b, −φ) (2.132)
ξ=ξ (b) (θ,φ)

It is worth noting that in the unique linear space R3 where all vectors are
defined, the envelope surface sg (θ, φ) is fixed and in line contact with the fixed
(a) (a)
surface defined by the position vectors pe (ξ(θ, φ), θ) − dba = se (θ, φ) − dba .
It is more common to have the components of the gear surface sg in another
reference frame, say Sq , with an axis parallel to b; more precisely we could take,
e.g., Sq = (Oq ≡ Ob ; xq , yq , zq ) with kq = b, j q = j and iq = j q × kq . All we need
is simply a change of coordinates from S to Sq , corresponding to a rotation, which
can be conveniently expressed by a 3 × 3 rotation matrix Lq
s̃g = Lq sg (θ, φ) (2.133)
2.15 Conclusions 33

Since the rotation of the gear around its axis is automatically taken into account in
these vector expressions, the generated gear sg is fixed in both S and Sq reference
systems.
Equation (2.133) is the expression that is usually obtained in the classical
approach, where a moving reference frame with the third axis coincident with
the axis b of the to be generated gear is generally employed. For instance, in
[4] equation (2.133) would be in the reference system S2 (one of the six reference
systems employed there for the gear generation), and indeed it is exactly the same.
To obtain the equation of the gear surfaces sg directly in Sq we can use the
following form

s̃g = R R(Lq pe , Lq a, η φ) − Lq dba , Lq b, −φ (2.134)

which leads to the very same vector of R3 as (2.133). In this form all the vectors
involved are first expressed in the reference frame Sq , which has the third axis
aligned with b, and then employed to perform the computations.
The generation of the mating pinion would follow precisely the same steps.
As confirmed by this example, the proposed approach can be carried out theo-
retically without any reference system, thus providing for a very compact formu-
lation. Moreover, all the actual computations can be easily performed using only
one reference system.

2.14.4 Case studied


The proposed approach has been employed to model a transmission for aerospace
application in the Avio firm. A picture of this transmission is represented in
Fig. 2.3. The main data of the transmission, the geometric parameters of the
grinding tool and the machine settings for the gear are given in Tables 2.1, 2.2
and 2.3.

2.15 Conclusions
The theory of gearing has been developed taking a fresh approach. Contrary to
common practice, vectors are not dealt with by means of their components, but
are treated as such. As a consequence no reference systems are needed and the
overall formulation becomes more compact.
All major aspects in the theory of gearing have been analyzed, thus showing
that the proposed method can provide a complete formulation.
As also shown by means of a numerical example, actual computations are
performed using just one reference system. The difficult process of defining a chain
of reference systems, typical of the traditional approach, is completely avoided.
34 2. Generation with fixed axes and constant gear ratio

Figure 2.3: Aerospace transmission modelled.

Parameter name Symbol Value S.A.F. Ref. (Record, Item)


Pinion tooth number N1 27 (1, 1)
Gear tooth number N2 38 (1, 2)
Module m 4.950 mm –
Mean spiral angle β 35.0 deg (8, 4)
Face width Fw 32.0 mm (1, 5)
Outer cone distance A0 97.8706 mm (3, 10)
Gear face angle γa 2 76.371 deg (7, 6)

Table 2.1: Main input data of the transmission modelled.

Parameter name Symbol Value S.A.F. Ref. (Record, Item)


Cutter point radius (concave side) Rg 77.5335 mm (25, 8)
Cutter point radius (convex side) Rg 74.8665 mm (21, 8)
Blade angle (concave side) αg 20.0 deg (27, 4)
Blade angle (convex side) αg 20.0 deg (23, 4)
Edge radius (concave side) ρw 1.27 mm (26, 15)
Edge radius (convex side) ρw 1.27 mm (22, 15)

Table 2.2: Gear grinding wheel parameters.

Parameter name Symbol Value S.A.F. Ref. (Record, Item)


Radial setting Sr 2 73.162 mm (25,1)
Blank offset ∆EM2 0.00 mm (25,4)
Root angle γm 2 70.288 deg (25,5)
Mach. center to back ∆XD2 0.00 mm (25,6)
Sliding base ∆XB2 -0.0067564 mm (25,7)
Cradle angle q2 58.5578 deg (26,9)
Ratio of roll τ = 1/η 1.038512 (25,13)

Table 2.3: Machine settings employed for gear finishing.


Chapter 3

Generation with translating


axes and variable gear ratio

In this chapter we extend the theory developed in Chapter 2 to include the case
when the axes of the tool and the gear blank are subjected, during generation,
to a relative translating motion. Moreover, particular restrictions are imposed to
the ratio of roll, which can vary as a function of the parameter of motion. With
the results of this chapter, it is possible to analyze, for example, the generation of
hypoid gears with correction motions like modified roll, helical and vertical motion.

3.1 Rotating vectors


In this analysis we will employ extensively rotations of position vectors and their
derivatives. Therefore, we recall the definition of the rotation operator R given in
Chapter 2 refreshing some of its general properties.
Let us consider an axis a, that is a directed straight line in an Euclidean
space E3 , typically defined by means of one of its points O (not necessarily fixed)
and a unit vector a that marks its direction. We introduce the compact notation
R(p, a, α(φ)) to denote the rigid rotation of a position vector p = P −O around the
unit vector a by an angle α(φ) (positive if counterclockwise), that is the rotation
of point P around axis a of an angle α(φ). The result of such rotation is the
position vector p̂

p̂(ξ, θ, φ) = P̂ − O = R p(ξ, θ), a, α(φ)
= (p · a)a + [p − (p · a)a] cos(α(φ)) + a × p sin(α(φ)) (3.1)

where P̂ is the image of P after the rotation. It is worth noting that α(φ) can
be an arbitrary function of a general parameter of motion φ, which represents the
independent variable controlling the motion.
36 3. Generation with translating axes and variable gear ratio

Position vectors p and p̂, like all vectors, belong to the linear space R3 , while
points belong, as already stated, to Euclidean (i.e., affine) spaces E3 .

3.1.1 Algebraic properties of rotating vectors


Several properties follow at once from Eq. (3.1); in particular let

û = R(u, a, α) v̂ = R(v, a, α) ŵ = R(w, a, α) (3.2)

we immediately have the following relations

u = R(R(u, a, α), a, −α) (3.3)


û + v̂ = R(u + v, a, α) (3.4)
û · v̂ = u · v (3.5)
v̂ × ŵ = R(v × w, a, α) (3.6)
 
û · (v̂ × ŵ) = u · (v × w) = u v w (3.7)

From (3.6), being obviously a = R(a, a, α), we also obtain

a × R(w, a, α) = R(a × w, a, α) (3.8)

3.1.2 Differential properties of rotating vectors


The derivatives of the rotated vector p̂ defined in (3.1) with respect to the para-
meters ξ and θ are simply given by

p̂,ξ (ξ, θ, φ) = R p,ξ (ξ, θ), a, α(φ)

p̂,θ (ξ, θ, φ) = R p,θ (ξ, θ), a, α(φ) (3.9)

whereas the derivative with respect to φ that controls the rigid rotation is

p̂,φ (ξ, θ, φ) = R p(ξ, θ), a, α(φ) ,φ

= α0 a × R p(ξ, θ), a, α(φ) = α0 a × p̂(ξ, θ, φ), (3.10)

where α0 (φ) = dα(φ)/dφ. In a more general case, where both the initial vector
and the rotation depend on the same parameter φ like in

p̂(ξ, θ, φ) = R(p(ξ, θ, φ), a, α(φ)), (3.11)

a combination of the previous results is required



p̂,φ (ξ, θ, φ) = R p,φ (ξ, θ, φ), a, α(φ) + α0 a × p̂(ξ, θ, φ) (3.12)

which generalizes Eq. (2.18) of Chapter 2 and Eq. (12) in [11].


3.2 Surface of the generating tool 37

3.2 Surface of the generating tool


In the Euclidean space E3e we define the generating tool to be a (fixed) regular
surface Σe . Its generic point will be denoted by Pe (ξ, θ), with (ξ, θ) ∈ A, where A
is an open set of R2 .
Once a fixed point Oe has been selected, it is possible to associate, as usual, to
each point Pe of E3e a position vector pe of R3

pe (ξ, θ) = Pe (ξ, θ) − Oe (3.13)

By definition, the normal vector me to Σe is given by

me (ξ, θ) = pe,ξ × pe,θ (3.14)

Owing to the assumed regularity of the surface Σe we always have

me 6= 0 (3.15)

Therefore, the unit normal vector mue is


me (ξ, θ) me (ξ, θ) me (ξ, θ)
mue (ξ, θ) = =√ = , (3.16)
|me (ξ, θ)| EG − F 2 D(ξ, θ)
where E, F, G are the coefficients of the first fundamental form of Σe [33, p. 82],
that is
E = pe,ξ · pe,ξ F = pe,ξ · pe,θ G = pe,θ · pe,θ (3.17)

and D = EG − F 2 .
The coefficients of the second fundamental form are given by [33, p. 119]

L = −mue ,ξ ·pe,ξ M = −mue ,ξ ·pe,θ = −mue ,θ ·pe,ξ N = −mue ,θ ·pe,θ


(3.18)
To investigate some geometric properties of surfaces, like normal curvatures, it
is often convenient to employ suitable curves on the surfaces involved [33, p. 121].
Therefore, let us consider an arbitrary regular curve ce on the tool surface Σe
defined by the position vectors

ce (se ) = pe (ξe (se ), θe (se )), (3.19)

where ξe (se ) and θe (se ) are analytic functions and se is the arc length of ce .
The unit tangent vector to ce is given by
dce
te (se ) = = pe,ξ ξe,se + pe,θ θe,se = pe ,se (3.20)
dse
and the derivative along ce of the unit normal vector is
dmue
= mue ,ξ ξe,se + mue ,θ θe,se = mue ,se (3.21)
dse
38 3. Generation with translating axes and variable gear ratio

The normal curvature kne and the geodesic torsion τ e of Σe along ce can be
easily computed by means of the following well known formulas
kne = −mue ,se ·te (3.22)
 
τ e = mue mue ,se te (3.23)
It is worth mentioning that in most cases we are merely interested in the
computation of kne and τ e at a selected point, say pe (ξ, ˇ θ̌) with ξˇ = ξe (še ) and
θ̌ = θe (še ), and along a given direction marked in (3.20) by the two numbers
eξ = ξe,se (še ) and eθ = θe,se (še ). Therefore, we do not really need the expressions
of the two functions ξe (se ) and θe (se ). Similar ideas can also be found in [9, Eqs.
(3) and (4)].
At a given point on Σe the principal directions g e1 and g e2 and the corresponding
principal curvatures k1e and k2e can be easily obtained [13]. Let λ1 , λ2 , (ε1 , χ1 ) and
(ε2 , χ2 ) be the eigenvalues and eigenvectors of the matrix
   −1  
a11 a12 E F L M
A= =− (3.24)
a21 a22 F G M N
Then we have
k1e = −λ1 k2e = −λ2 (3.25)
and
ε1 pe,ξ + χ1 pe,θ
g e1 =
|ε1 pe,ξ + χ1 pe,θ |
ε2 pe,ξ + χ2 pe,θ
g e2 = (3.26)
|ε2 pe,ξ + χ2 pe,θ |
Therefore, once obtained at a given point on Σe the principal directions g e1 and
g e2 and the corresponding principal curvatures k1e and k2e , we may write
te (se ) = cos β g e1 + sin β g e2 (3.27)
and, employing Rodrigues’ formula,
mue ,se = −(k1e cos β g e1 + k2e sin β g e2 ) = −(k1e (te · g e1 ) g e1 + k2e (te · g e2 ) g e2 ) (3.28)
where β is the angle, in the tangent plane, between g e1 and te , positive if a counter-
clockwise rotation around mue aligns g e1 onto te (we remind that g e1 × g e2 = mue ).
Moreover the normal curvature and the geodesic torsion defined in Eqs. (3.22)
and (3.23) can be calculated by means of Euler’s and Bertrand’s formulas, respec-
tively [33, p. 132 and p. 159],
kne = k1e cos2 β + k2e sin2 β (3.29)
1 e
τe = k = (k2e − k1e ) sin β cos β (3.30)
2 n,β
3.3 A fixed family of surfaces 39

3.3 A fixed family of surfaces


Let us consider another Euclidean space E3f . As shown in Fig. 6.1, in this new
space we define a first translating axis a by means of one of its (moving) points
Oa (φ) and a fixed unit vector a, and a second translating axis b again by means
of one of its points Ob (φ) and a fixed unit vector b. In practical terms they are
the two axes (i.e., directed straight lines) of the gear pair under investigation (the
generating tool and the to be generated gear).
The dependance of Oa and Ob on the parameter of motion φ shows that in the
following treatment the two axes a and b are allowed to translate with respect to
each other, while their directions remain fixed, that is a and b are constant unit
vectors. This is a more general (and more realistic) setting with respect to the
analysis presented in Chapter 2 and published in [11], where the two axes were
assumed to be fixed in E3f .
In the fixed space E3f we define a regular surface Σ̂(φ), isomorphic to the surface
Σe and rigidly rotating around the first axis a by an angle ψ(φ) while moving with
the transfer motion of Oa . In gear generation Σ̂(φ) is the moving surface of the
generating tool in the fixed space.
The sequence of regular surfaces Σ̂(φ), one for each value of φ, defines a one-

S (f)

Oa(f) y(f)

Gg (f) pa
a

mu
d (f)

pb

j (f)
Ob(f)

Figure 3.1: Mating surfaces in the fixed space E3f .


40 3. Generation with translating axes and variable gear ratio

parameter family of surfaces Φf in the space E3f . As already stated, it is a family


of a special kind, since it only involves rigid-body motions, that is a rigid-body
rotation of a given surface around an axis a plus a translation due to Oa (φ). To
avoid any misunderstanding, we mention that Φf is not the family of surfaces
whose envelope we are looking for.
Denoting by P̂ (ξ, θ, φ) the generic point of Σ̂(φ), its position vector p̂a with
respect to the moving point Oa (φ) is given by the following expression which
employs the compact notation introduced in (3.1)

p̂a (ξ, θ, φ) = P̂ (ξ, θ, φ) − Oa (φ) = R pe (ξ, θ), a, ψ(φ) (3.31)
and where pe (ξ, θ) was defined in Eq. (3.13).
Of course, it is equally possible to relate P̂ (ξ, θ, φ) to the position vector
p̂b (ξ, θ, φ) with respect to the moving point Ob (φ)
p̂b (ξ, θ, φ) = P̂ (ξ, θ, φ) − Ob (φ) = p̂a (ξ, θ, φ) − d(φ) (3.32)
where (Fig. 6.1)
d(φ) = Ob (φ) − Oa (φ) (3.33)
is a variable vector depending arbitrarily on φ. It should be noted that the rela-
tionship between pe and p̂b is a rotation plus a translation, i.e. a general affine
transformation. Both position vectors p̂a (ξ, θ, φ) and p̂b (ξ, θ, φ) of R3 are related
to the same family of surfaces Φf of E3f .
As a consequence of Eqs. (3.31) and (3.32), along with the general properties
(3.9) and (3.10), we have the following derivatives
p̂a,ξ = p̂b,ξ = R(pe ,ξ , a, ψ(φ))
p̂a,θ = p̂b,θ = R(pe ,θ , a, ψ(φ)) (3.34)
0 0 0
p̂a,φ = ψ a × p̂a = ψ a × R(pe , a, ψ(φ)) = R(ψ a × pe , a, ψ(φ)) (3.35)
0
p̂b,φ = p̂a,φ − d (3.36)

By definition the normal vector m̂ to each regular surface Σ̂(φ) of the family
(hence for fixed φ) is given by
m̂(ξ, θ, φ) = p̂a,ξ × p̂a ,θ = p̂b,ξ × p̂b ,θ = R(me (ξ, θ), a, ψ(φ)) 6= 0 (3.37)
where Eqs. (3.6), (3.9), (3.14) and (3.15) were employed. We see from the last
result that m̂ is never zero, thus confirming the assumed regularity of each surface
of the family.
From Eqs. (3.10) and (3.37) we can obtain the derivative of m̂ with respect to
the parameter of motion φ (rigid-body rotation)
m̂,φ = ψ 0 a × m̂ (3.38)
We remark that vectors, including position vectors and their derivatives, re-
gardless on how they were defined, all belong to the very same linear space R3 .
3.4 The enveloping family of surfaces 41

3.4 The enveloping family of surfaces


Let us consider yet another Euclidean space E3g . In this new space we define a
fixed axis b by means of one of its points Og and a unit vector b, which is the same
already introduced at the beginning of Sec. 3.3. In applications, E3g is the moving
space where the gear surface Γg will be obtained as a fixed one.
At this point we need to define the (signed, possibly non-constant) gear ratio
τ (φ)
ϕ0 (φ)
τ (φ) = 0 (3.39)
ψ (φ)
between the mating tool and the to be generated gear. Both rotations ϕ(φ) and
ψ(φ) may depend on the parameter of motion φ, and therefore the so-called mod-
ified roll [77] is included in the present analysis (whereas in Chapter 2 a constant
τ was assumed).
In E3g we define the one-parameter family of surfaces Φg whose envelope Γg we
are interested in. It is given by a sequence of surfaces Σg (φ), again isomorphic to
Σe . If Pg denotes the generic point of Φg , we may define in R3 the position vector
pg = Pg − Og by imposing to p̂b a rotation around b by an angle −ϕ(φ)

pg (ξ, θ, φ) = R p̂b (ξ, θ, φ), b, −ϕ(φ)

= R p̂a (ξ, θ, φ) − d(φ), b, −ϕ(φ)

= R R(pe (ξ, θ), a, ψ(φ)) − d(φ), b, −ϕ(φ) (3.40)

The reason why we have a minus sign in front of the scalar function ϕ(φ) is to
define a rotation opposite to the one actually experienced by the gear. This is
done in order to observe the actual enveloping family, that is the one which is seen
under the transfer motion with the gear.
An equivalent form of Eq. (3.40) is

R(pg (ξ, θ, φ), b, ϕ(φ)) = p̂b (ξ, θ, φ) = R(pe (ξ, θ), a, ψ(φ)) − d(φ) (3.41)

It is worth noting that this definition of the enveloping family pg (ξ, θ, φ) does
not require reference systems, since all position vectors belong to the linear space
R3 regardless of the Euclidean space employed for their definition.
It could be easily proven that the screw axis of relative motion between the
two gears is directed like the possibly non-constant vector

w(φ) = ψ 0 (φ) a − ϕ0 (φ) b (3.42)

3.4.1 Derivatives of the enveloping family of surfaces


Combining Eqs. (3.34) and (3.40) and the general properties (3.9), we immedi-
ately obtain that the derivatives with respect to ξ and θ of the (position vectors
42 3. Generation with translating axes and variable gear ratio

associated to the) family of surfaces Φg are given by the following expressions


 
pg,ξ = R p̂b,ξ (ξ, θ, φ), b, −ϕ(φ) = R R(pe ,ξ (ξ, θ), a, ψ(φ)), b, −ϕ(φ)
 
pg,θ = R p̂b ,θ (ξ, θ, φ), b, −ϕ(φ) = R R(pe ,θ (ξ, θ), a, ψ(φ)), b, −ϕ(φ) (3.43)
These results clearly show the relationship between the derivatives of the two
families of surfaces Φg and Φf and also of the generating surface Σe .
The derivative of pg with respect to the parameter of motion φ can be obtained
as in Eq. (3.12)
 
pg,φ (ξ, θ, φ) = R p̂b (ξ, θ, φ), b, −ϕ(φ) ,φ

= R p̂a,φ (ξ, θ, φ) − d0 (φ), b, −ϕ(φ) − ϕ0 (φ)b × pg (ξ, θ, φ)

= R ψ 0 (φ)a × p̂a (ξ, θ, φ) − d0 (φ) − ϕ0 (φ)b × p̂b (ξ, θ, φ), b, −ϕ(φ)
(3.44)
where the last step comes from Eqs. (3.35) and (3.40). If we define the vector
function
ĥ(ξ, θ, φ) = ψ 0 (φ)a × p̂a (ξ, θ, φ) − ϕ0 (φ)b × p̂b (ξ, θ, φ) − d0 (φ)
= (ψ 0 (φ)a − ϕ0 (φ)b) × p̂a (ξ, θ, φ) + ϕ0 (φ)b × d(φ) − d0 (φ)
= w(φ) × p̂a (ξ, θ, φ) + ϕ0 (φ)b × d(φ) − d0 (φ) (3.45)
the derivative pg,φ can be simply written as

pg,φ (ξ, θ, φ) = R ĥ(ξ, θ, φ), b, −ϕ(φ) (3.46)
formally similar to Eq. (2.36) and to Eq. (30) in [11].
It will also turn out useful to define the following vector function

he (ξ, θ, φ) = R ĥ(ξ, θ, φ), a, −ψ(φ)
 
= R w(φ), a, −ψ(φ) × pe (ξ, θ) + R ϕ0 (φ)b × d(φ) − d0 (φ), a, −ψ(φ)
= we (φ) × pe (ξ, θ) + q e (φ) (3.47)
where 
we (φ) = R w(φ), a, −ψ(φ) (3.48)
is the direction of the screw axis as seen by the tool surface, and

q e (φ) = R ϕ0 (φ)b × d(φ) − d0 (φ), a, −ψ(φ) (3.49)
vanishes whenever d = d0 = 0.
Eq. (3.47) generalizes Eq. (2.48) in Chapter 2 and Eq. (42) in [11] and allows
us to write  
pg,φ (ξ, θ, φ) = R R he (ξ, θ, φ), a, ψ(φ) , b, −ϕ(φ) (3.50)
As shown in Section 14 in [11], ĥ and he are directly related to the sliding
velocity.
3.5 Equation of meshing 43

3.4.2 Normal vectors to the enveloping family of surfaces


According to Eqs. (3.14), (3.37) and (3.43), the normal vector mg to each regular
surface Σg (φ) of the family Φg (hence for fixed φ) is given by

mg (ξ, θ, φ) = pg,ξ × pg,θ = R m̂(ξ, θ, φ), b, −ϕ(φ)

= R R(me (ξ, θ), a, ψ(φ)), b, −ϕ(φ) (3.51)
which also shows the relationship between the normal vectors defined so far. Sim-
ilarly, the unit normal vector to each surface of the enveloping family Φg can be
expressed as

mug (ξ, θ, φ) = R R(mue (ξ, θ), a, ψ(φ)), b, −ϕ(φ) (3.52)
The derivatives of mug with respect to ξ and θ are readily obtained from (3.52)

mug ,ξ (ξ, θ, φ) = R R(mue ,ξ (ξ, θ), a, ψ(φ)), b, −ϕ(φ)

mug ,θ (ξ, θ, φ) = R R(mue ,θ (ξ, θ), a, ψ(φ)), b, −ϕ(φ) (3.53)
while, according to (3.12), (3.38) and (3.42)
 
mug ,φ (ξ, θ, φ) = R m̂u,φ (ξ, θ, φ), b, −ϕ(φ) − ϕ0 (φ)b × R m̂u (ξ, θ, φ), b, −ϕ(φ)

= R m̂u,φ − ϕ0 b × m̂u , b, −ϕ

= R ψ 0 a × R(mue , a, ψ) − ϕ0 b × R(mue , a, ψ), b, −ϕ

= R w(φ) × R(mue (ξ, θ), a, ψ(φ)), b, −ϕ(φ)
 
= R R we (φ) × mue (ξ, θ), a, ψ(φ) , b, −ϕ(φ) (3.54)

3.5 Equation of meshing


Since in gear generation [48] we are looking for the envelope Γg of the family of
surfaces Φg in the Euclidean space E3g , the definition of the equation of meshing
f = 0 must involve the triple product of the partial derivatives of the position
vectors pg (ξ, θ, φ)
 
pg,ξ pg,θ pg,φ = mg · pg,φ = f (ξ, θ, φ) = 0 (3.55)
with (ξ, θ, φ) ∈ B ⊂ R3 .
Following [11], it is now possible to considerably simplify the equation of mesh-
ing (3.55) by taking into account Eqs. (3.43), (3.46) and (3.47), along with the
properties (3.7)
 
f (ξ, θ, φ) = R(p̂b,ξ , b, −ϕ) R(p̂b ,θ , b, −ϕ) R(ĥ, b, −ϕ)
   
= p̂b,ξ p̂b ,θ ĥ = p̂a,ξ p̂a ,θ ĥ = m̂(ξ, θ, φ) · ĥ(ξ, θ, φ)
 
= R(pe ,ξ , a, ψ) R(pe ,θ , a, ψ) R(he , a, ψ)
 
= pe ,ξ pe ,θ he = me (ξ, θ) · he (ξ, θ, φ) = 0 (3.56)
44 3. Generation with translating axes and variable gear ratio

This formulation of the equation of meshing basically corresponds to the so-called


“engineering approach”, as introduced in [49], although here we do not need any
kinematical concept like the relative velocity, nor we need to use time as the
independent variable. Moreover, there are no reference systems involved.
Obviously, the two definitions (3.55) and (3.56) provide the same equation
of meshing. However the last one is more convenient since it fully exploits the
properties of rigid body motions involved in the generation of gears. Moreover it
is ready to be computed thanks to the definition (3.47) and it returns expressions
already simplified because the parameter of motion φ only appears in the last
vector. The major simplification is that we only employ the position vector pe of
the fixed surface Σe .
Instead of (3.56), a slightly different, although completely equivalent, form of
the equation of meshing with the unit normal vector mue (ξ, θ) can be employed

f˜(ξ, θ, φ) = mue (ξ, θ) · he (ξ, θ, φ) = 0 (3.57)

According to [84], page 46, f˜ would be called a “contact function in the strict
sense”, while f would just be the “contact function”. Similar ideas are also dis-
cussed at page 18, still in [84].

3.6 Derivatives of the equation of meshing


For later use (see Chapter 5, page 75) we obtain the expressions of the derivatives
of the function f˜ in the equation of meshing.
By employing the definition (3.57) of f˜, the partial derivative f˜,ξ is

f˜,ξ = mue ,ξ ·he + mue · he,ξ (3.58)

where the derivative he,ξ is easily computed according to the general property
(3.9) and definition (3.47) as

he,ξ (ξ, θ, φ) = R(w(φ), a, −ψ(φ)) × pe,ξ (ξ, θ) = we (φ) × pe,ξ (ξ, θ) (3.59)

Similarly we obtain

f˜,θ = mue ,θ ·he + mue · he,θ (3.60)

with

he,θ (ξ, θ, φ) = R(w(φ), a, −ψ(φ)) × pe,θ (ξ, θ) = we (φ) × pe,θ (ξ, θ) (3.61)

The derivative f˜,φ is


f˜,φ = mue · he,φ (3.62)
3.6 Derivatives of the equation of meshing 45

where, owing to (3.12) and (3.47)

he,φ = R(w0 − ψ 0 a × w, a, −ψ) × pe +



R ϕ00 b × d − ψ 0 a × (ϕ0 b × d) + (ψ 0 a + ϕ0 b) × d0 − d00 , a, −ψ
= w0e (φ) × pe (ξ, θ) + q 0e (φ) (3.63)

which, although fairly long, is an easily computable function. Derivatives of the


equation of meshing are calculated also in [9, Eq. (9)], obtaining similar relation-
ships.
Given any two functions ξ(s), θ(s), where s is not necessarily an arc length, let
us take φ(s) such that they all identically satisfy

f˜(ξ(s), θ(s), φ(s)) = f˜(s) = 0 (3.64)

As a consequence, we can also define the total derivative

df˜
= f˜,ξ ξ,s +f˜,θ θ,s +f˜,φ φ,s = 0 (3.65)
ds
which must be identically zero as well. We can interprete (3.65) as a condition
among the derivatives of ξ(s), θ(s) and φ(s).
From Eq. (3.65) and according also to (3.20), (3.21), (3.58) and (3.60)

df˜
= mue ,s · he + mue · we × pe ,s +f˜,φ φ,s = 0 (3.66)
ds
which, for instance, provides a way to compute f˜,φ φ,s . Employing also Eqs. (3.62)
and (3.63) we obtain

df˜
= mue ,s ·he + mue · we × pe ,s +(w0e × pe + q 0e ) · mue φ,s = 0 (3.67)
ds
which may provide φ,s . Equivalent expressions, although in kinematic terms, are
Eq. (13.1.19) in [48], Eq. (8.4.35) in [49] and Eq. (10) in [9].

3.6.1 Derivatives with respect to the arc length se


Eqs. (3.66) and (3.67) can be further developed. Although not strictly necessary,
let s be the arc length se of a curve ce on Σe as in (3.19). Taking into account
Eqs. (3.20), (3.27) and (3.28) and introducing, like in [84, p. 77], a new vector

k = −λ(ξ, θ, φ) g e1 − µ(ξ, θ, φ)g e2 (3.68)

where
" # " # " #
λ(ξ, θ, φ) k1e (ξ, θ) he (ξ, θ, φ) we (φ) g e1 (ξ, θ)
= · (3.69)
µ(ξ, θ, φ) −we (φ) k2e (ξ, θ) he (ξ, θ, φ) g e2 (ξ, θ)
46 3. Generation with translating axes and variable gear ratio

we obtain, as shown in the Appendix


mue ,se ·he + mue · we × te = k · te (3.70)
Accordingly, Eq. (3.66) and (3.67) become

df˜
= k · te + f˜,φ φ,se = 0 (3.71)
dse
df˜
= k · te + (w0e × pe + q 0e ) · mue φ,se = 0 (3.72)
dse
the last one being equivalent to Eq. (18) in [9], where b − ω 12 × n corresponds to
k.
It is worth noting that if we choose ce to be a contact line we have φ,se = 0
and the previous equation gives
k · te = 0 (3.73)
showing that k is always orthogonal to contact lines.
Should s not be an arc length, it just suffices to replace te by pe,s .

3.7 The envelope surface


The envelope surface Γg , that is the generated gear, is given by the combination of
the enveloping family of surfaces Φg and the equation of meshing. More precisely,
in R3 the position vectors sg of Γg are given by
(
sg = pg (ξ, θ, φ)
(3.74)
f (ξ, θ, φ) = 0

which employ equations (3.40) and (3.56). Similarly, the normal vector ng to Γg
is (
ng = mg (ξ, θ, φ)
(3.75)
f (ξ, θ, φ) = 0
An in depth analysis of the location of singular points on Γg , that is where
ng = 0, was presented in [11].

3.8 Conclusions
In this chapter the theory developed in Chapter 2 has been extended to gener-
ation with translating axes and variable gear ratio. The results obtained reveal
that this extension easily fits into the invariant approach framework. Only slight
changes to the notation are necessary and the basic structure of the theory remains
unchanged.
3.9 Appendix 47

3.9 Appendix
Proof of Eq. (3.70)
Equation (3.70) can be proven in the following way

k · te = mue ,se ·he + mue · we × te



= − k1e cos β g e1 + k2e sin β g e2 · he + mue × we · te

= − k1e (g e1 · he ) cos β + k2e (g e2 · he ) sin β + mue × we · te

= − k1e (g e1 · he )(g e1 · te ) + k2e (g e2 · he )(g e2 · te ) + mue × we · te

= − k1e (g e1 · he )g e1 + k2e (g e2 · he )g e2 · te + mue × we · te

= − k1e (g e1 · he )g e1 + k2e (g e2 · he )g e2 − mue × we · te

= − k1e (g e1 · he )g e1 + k2e (g e2 · he )g e2 − mue × ((we · g e1 )g e1 + (we · g e2 )g e2 ) · te

= − k1e (g e1 · he )g e1 + k2e (g e2 · he )g e2 − (we · g e1 )g e2 + (we · g e2 )g e1 · te
   
= − k1e (g e1 · he ) + (we · g e2 ) g e1 + k2e (g e2 · he ) − (we · g e1 ) g e2 · te

= −λg e1 − µg e2 · te (3.76)

since
me × g e1 = g e2 me × g e2 = −g e1 (3.77)
Chapter 4

Generation with
supplemental motions

The use of Computer Numerically Controlled machines for manufacturing gears


has demanded for general three-dimensional motions to be addressed. Kinematic
equations, involving the instantaneous screw axis in the relative motion, are pre-
sented in [8] and [26]. A mathematical model of a universal hypoid generator,
based on Litvin’s approach, is developed in [19]. Similar ideas can be found in
[34], but specifically applied to Klingelnberg bevel gears.

In this chapter the vector approach to gear generation presented Chapters 2


and 3 and published in [11] and [12] is extended to cope with fully general en-
veloping motions, like those employed in free-form cutting machines. Actually,
more than six parameters (somehow like the eight correction mechanisms in the
Phoenix Gleason machine [77]) are included in the analysis to retain the physical
meaning of some degrees of freedom and to make the whole formulation easier to
apply to some practical cases.

First a detailed analysis to obtain the basic equations of the generated surface
is presented. The proposed approach allows for a very compact and simple for-
mulation since it does not require reference systems. With respect to the analyses
presented in [11, 12], several new aspects have to be addressed, making the gener-
alization far from obvious. In the last part, a brief explanation on how to obtain
some kinematic relationships is provided. Again, some new subtle points have to
be taken into account. The theoretical results obtained in this chapter will be
tested in Chapter 5 where the simulation of the grinding process of a spiral bevel
pinion is performed.
50 4. Generation with supplemental motions

4.1 Rotating vectors


The new approach to the theory of gearing, proposed in [11] and extended in this
chapter, is deeply based on the rotation operator R, representing a rigid rotation
in vectorial form.
Let us consider an axis a, that is a directed straight line in an Euclidean space
E3 , defined by means of one of its points O and a unit vector a that marks its
direction. The rotation of a point P around the axis a of an angle α, positive if
counterclockwise with respect to a, is denoted by R(p, a, α) with p = P − O; such
rotation moves the point P onto P̂ which is given by the following formula

p̂(ξ, θ, α) = P̂ − O = R p(ξ, θ), a, α
= (p · a)a + [p − (p · a)a] cos α + a × [p − (p · a)a] sin α (4.1)

Some properties of the rotation operator are reported in Appendix 4.10.

4.1.1 Rotation about a mobile axis


One of the specific features of this chapter is the introduction of mobile axes,
meaning that in Eq. (4.1) also a can be dependent on the angle α, that is

p̂(ξ, θ, α) = P̂ − O = R p(ξ, θ), a(α), α (4.2)

The current orientation of the axis a, described by a(α), can be obtained in dif-
ferent ways, but it may be convenient to employ (elementary) rotations about
fixed axes. Hence, a fixed orthogonal reference frame of unit vectors i, j and k is
introduced (although only two axes are really employed). It is defined as follows:
denoting by a0 the initial orientation of a and choosing arbitrarily a unit vector
k, the unit vectors i and j are obtained as
k × a0
j= ; i=j×k (4.3)
|k × a0 |
In this way the current unit vector a(α) is given by the following expression,
employing two elementary rotations

a(α) = R R(a0 , j, σ(α)), k, ζ(α) (4.4)

where the angles σ and ζ are shown in Fig. 4.1 and are given by the following
equations
sin(δ) = |k × a0 |

cos(δ + σ(α)) = a(α) · k


(4.5)
i · (a(α) − (a(α) · k)k)
cos(ζ(α)) =
sin(δ + σ(α))
4.1 Rotating vectors 51

As it is well known, a sequence of rotations, about mobile or fixed axes, is


again a rotation. Thus Eq. (4.4) can also be written as

a(α) = R a0 , r(α), ε(α) (4.6)

being [2],
!
i · î + j · ĵ + k · k̂ − 1
ε(α) = arccos
2
 
cos ζ cos σ + cos ζ + cos σ − 1
= arccos (4.7)
2
1 h i
r(α) = (k · ĵ − j · k̂)i + (i · k̂ − k · î)j + (j · î − i · ĵ)k
2 sin ε
1 
= − sin ζ sin σ, sin σ(1 + cos ζ), sin ζ(1 + cos σ) (4.8)
2 sin ε

where σ(α) and ζ(α) were defined in (4.5) and î, ĵ and k̂ are the unit vectors,
initially overlapped to i, j and k, after the sequence of rotations.

a0

d
s
a
s
j

Figure 4.1: Representation of parameters describing the rotating vector a.


52 4. Generation with supplemental motions

4.1.2 Differential properties


The derivatives of the rotated vector p̂ introduced in Eq. (4.2) with respect to the
parameters ξ and θ are given by

p̂,ξ (ξ, θ, α) = R p,ξ (ξ, θ), a(α), α
 (4.9)
p̂,θ (ξ, θ, α) = R p,θ (ξ, θ), a(α), α

whereas the derivative with respect to α (rigid rotation), that can be obtained
from Eqs. (4.1) and (4.2), is

p̂,α (ξ, θ, α) = a × p̂ + (1 − cos α)[(p · a0 )a + (p · a)a0 ] + a0 × p sin α (4.10)

Similarly, from (4.4), the derivative of the unit vector a(α) is



a0 = ζ 0 k × a + R R(a0 , j, σ(α)),α , k, ζ(α)

= R (σ 0 j + ζ 0 k) × R(a0 , j, σ(α)), k, ζ(α) (4.11)

Therefore, indicating with


 
ω(α) = R ζ 0 k + σ 0 j, k, ζ(α) = ζ 0 k + R σ 0 j, k, ζ(α) (4.12)

a very simple form for the derivative of a is obtained

a0 = ω(α) × a(α) (4.13)

that, inserted in Eq. (4.10), gives

p̂,α = a × p̂ + (1 − cos α)[(p · (ω × a))a + (p · a)(ω × a)] + (ω × a) × p sin α (4.14)

Introducing a new vector $(α)

$(α) = ω − R(ω, a, α)
= (1 − cos α)(ω − (ω · a) a) + sin α (ω × a) (4.15)

it is possible to show that the following relationship holds

$(α) × p̂ = (1 − cos α)[(p · (ω × a))a + (p · a)(ω × a)] + (ω × a) × p sin α (4.16)

Therefore Eq. (4.14) simply becomes

p̂,α = (a + ω − R(ω, a, α)) × p̂ = (a + $) × p̂ (4.17)

which generalizes a result presented in [11] for fixed a (and hence ω = 0).
In a even more general case, where both the initial vector p(ξ, θ, α) and the ro-
tation depend on the same parameter α, an additional term is required in Eq. (4.17)

p̂,α = (a + $) × p̂ + R(p,α (ξ, θ, α), a(α), α) (4.18)


4.2 Generating surface 53

4.2 Generating surface


The generating surface Σe (i.e., the tool surface) is introduced in a Euclidean space
E3e and is assumed to be a regular surface. Selected a fixed point Oe , it is possible
to associate to each point Pe of Σe a position vector pe

pe (ξ, θ) = Pe (ξ, θ) − Oe (4.19)

It is worth noting that the position vectors pe , like all vectors, belong to the
linear space R3 , while the points Pe belong, as already stated, to the Euclidean
(i.e., affine) space E3e .
By definition, the normal vector me to Σe is given by

me (ξ, θ) = pe,ξ × pe,θ (4.20)

and, owing to the assumed regularity of the surface Σe , we always have me 6= 0.


Since in curvature analysis some concepts of differential geometry are necessary,
the coefficients E, F, G of the first fundamental form of Σe [33, p. 82] are introduced

E = pe ,ξ · pe ,ξ F = pe ,ξ · pe ,θ G = pe ,θ · pe ,θ (4.21)

where D = EG − F 2 . Accordingly the unit normal vector to the surface mue is
given by
me (ξ, θ) me (ξ, θ) me (ξ, θ)
mue (ξ, θ) = =√ = (4.22)
|me (ξ, θ)| EG − F 2 D(ξ, θ)

4.3 Generating process in the fixed space


Let E3f be the Euclidean fixed space where the generating (i.e., cutting) process
takes place. In this space, as shown in Fig. 4.2, we introduce the axes of the gear
pair: the tool axis a (i.e., the axis of the virtual generating gear), defined by means
of one of its points Oa and a unit vector a, and the to-be-generated gear axis b,
again defined by means of one of its points Ob and a unit vector b. Both axes are
well defined in a practical setting. In case of generalized motion we assume that
the points Oa and Ob as well as the axes unit vectors a and b all depend on the
parameter of motion φ.
Denoted by a0 and b0 the unit vectors corresponding to the initial orientations
of axes a and b, their current orientations can be obtained according to Eq. (4.4),
once a suitable unit vector k has been selected,

a(φ) = R R(a0 , j a , σa (φ)), k, ζa (φ) (4.23)

b(φ) = R R(b0 , j b , σb (φ)), k, ζb (φ) (4.24)
54 4. Generation with supplemental motions

S (y)

Oa(f) y(f)

Gg pa
a(f)

mu
d (f)

pb

j (f)
Ob(f)

b (f)

Figure 4.2: Mating surfaces in the fixed space E3f .

where, like in (4.3),

k × a0
ja = , ia = j a × k (4.25)
|k × a0 |
k × b0
jb = , ib = j b × k (4.26)
|k × b0 |
It could be convenient to choose a0 = k and
a0 × b0
ja = jb = j = , i=j×k (4.27)
|a0 × b0 |
but we stick to the more general case defined above in (4.23).
As already mentioned, it is possible also to get to the current orientation with
just one rotation, according to Eqs. (4.6), (4.7) and (4.8),
a(φ) = R(a0 , r a (φ), εa (φ)), b(φ) = R(b0 , r b (φ), εb (φ)) (4.28)
In the space E3f , the position vector of each point P can be referred to Oa or
Ob , that is
pa = P − Oa (φ) and pb = P − Ob (φ) (4.29)
4.3 Generating process in the fixed space 55

and, denoting by d(φ) = Ob (φ) − Oa (φ), they are related as follows

pb = pa − d(φ) (4.30)

Again all position vectors belong to R3 .

4.3.1 Moving tool surface


The moving tool surface in E3f is defined as a regular surface Σ̂(ψ) isomorphic
to Σe and rigidly rotating around the mobile axis a by an angle ψ (as usual a
counterclockwise rotation is taken as positive), depending on the parameter of
motion, i.e., ψ = ψ(φ). Denoting by P̂ the generic point of Σ̂, we assume that in
the initial position ψ(0) = 0 and P̂ − Oa = (P − Oe ) + (Oe − Oa ) = pe (ξ, θ) + e(0),
where, in general, e(φ) = Oe (φ) − Oa (φ).
During their motion, the position vectors p̂a = P̂ − Oa of the tool surface can
be given in two equivalent ways:
a) first aligning the tool axis onto the unit vector a and then rotating the tool
surface around it

p̂a (ξ, θ, φ) = P̂ (ξ, θ, φ) − Oa (φ) = R p̃e (ξ, θ, φ), a(φ), ψ(φ) (4.31)

where, as in (4.28),

p̃e (ξ, θ, φ) = R R(pe (ξ, θ) + e(φ), j a , σa (φ)), k, ζa (φ)

= R pe (ξ, θ) + e(φ), r a (φ), εa (φ) (4.32)

b) first applying the rotation ψ(φ) in the initial position, i.e. about a0 , and then
aligning a0 onto a(φ), that is
 
p̂a (ξ, θ, φ) = R R(p̂e (ξ, θ, φ), j a , σa (φ)), k, ζa (φ) = R p̂e (ξ, θ, φ), r a (φ), εa (φ)
(4.33)
with 
p̂e (ξ, θ, φ) = R pe (ξ, θ) + e(φ), a0 , ψ(φ) (4.34)
In the following, to stress the role of the mobile axis a being the axis of the
virtual generating gear, we will adopt the definition (4.31), and make use of a
compact notation for expressing the sequence of rotations

Ra (v) = R R(v, r a (φ), εa (φ)), a(φ), ψ(φ) (4.35)

which allows Eq. (4.31), that is the moving tool surface, to be written, referred to
Oa or Ob respectively, as

p̂a (ξ, θ, φ) = Ra (pe (ξ, θ) + e(φ)) (4.36)


p̂b (ξ, θ, φ) = Ra (pe (ξ, θ) + e(φ)) − d(φ) (4.37)
56 4. Generation with supplemental motions

It will be useful also the inverse sequence of rotations



Rai (v) = R R(v, a(φ), −ψ(φ)), r a (φ), −εa (φ) (4.38)

For the sake of clarity, it is worth stressing that Eq. (4.35) is written as a sequence
of two rotations but, for Eqs. (4.4) and (4.6), it represents three elementary ro-
tations. Therefore, to describe the motion of the gear around a mobile axis, a
three-parameters representation is employed, with three elementary rotations of
the Euler’s kind, that is 3–2–3.

4.3.2 Useful derivatives


The derivatives of the vector functions p̂a (ξ, θ, φ) and p̂b (ξ, θ, φ), can be easily
obtained employing (4.9), (4.31) and (4.35),

p̂a ,ξ = p̂b ,ξ = R p̃e ,ξ (ξ, θ, φ), a(φ), ψ(φ)

= R R(pe,ξ (ξ, θ), r a (φ), εa (φ)), a(φ), ψ(φ) = Ra (pe ,ξ )
 (4.39)
p̂a ,θ = p̂b ,θ = R p̃e ,θ (ξ, θ, φ), a(φ), ψ(φ)

= R R(pe ,θ (ξ, θ), r a (φ), εa (φ)), a(φ), ψ(φ) = Ra (pe ,θ )

According to (4.18), the derivative with respect to the parameter of motion φ is



p̂a ,φ = p̂b ,φ + d0 = R(p̃e ,φ (ξ, θ, φ), a(φ), ψ(φ) + (ψ 0 a + $ a (φ)) × p̂a (4.40)

where, employing (4.15),

$ a (φ) = ω a − R(ω a , a, ψ) (4.41)

with, from (4.12),



ω a = ζa0 k + R (σa0 j a ), k, ζa (φ) (4.42)

In addition, repeating the steps leading to Eq. (4.13), we have



p̃e ,φ = R R(pe (ξ, θ) + e(φ), j, σa (φ)), k, ζa (φ) ,φ

= ζa0 k × p̃e + R R(pe + e, j, σa (φ)),φ , k, ζa (φ)
 
= R (σa0 j + ζa0 k) × R(pe + e, j a , σa (φ)), k, ζa (φ) + R R(e0 , j a , σa (φ)), k, ζa (φ)
= ω a × p̃e + R(e0 , r a , εa ) (4.43)

Therefore Eq. (4.40) simply becomes



p̂a ,φ = p̂b ,φ + d0 = R(ω a , a, ψ) + ψ 0 a + $ a × p̂a + Ra (e0 )
= (ψ 0 a + ω a ) × p̂a + Ra (e0 ) = wa × p̂a + Ra (e0 ) (4.44)
4.4 The enveloping family of surfaces 57

with
wa (φ) = ψ 0 a + ω a (4.45)
Equations (4.44) and (4.45) are two simple expressions corresponding to well
known kinematic results, the first one giving the velocity of P̂ relative to Oa
and the second one the composition of the angular velocities. They generalize
Eq. (35) in [12].
By definition the (moving) normal vector m̂ to each regular surface Σ̂ of the
family is given by

m̂(ξ, θ, φ) = p̂a ,ξ × p̂a ,θ = Ra (pe ,ξ × pe ,θ ) = Ra (me (ξ, θ)) 6= 0 (4.46)

where equation (4.20), with properties (4.9) and (4.131) were employed, along with
the regularity of Σe . In a similar way, m̂u = Ra (mue ). We see from the last result
that m̂ is never zero, thus confirming the assumed regularity of each surface of
the family. As in (4.44), the derivative of the normal vector with respect to φ is

m̂,φ = wa × m̂ (4.47)

4.4 The enveloping family of surfaces


Usually we want to study the envelope surface (i.e., the generated gear) as it were
fixed. Therefore we introduce another Euclidean space E3g , where we define a fixed
axis b (i.e., the axis of the generated gear) by means of one of its points Og and a
unit vector b0 .
In E3g we define a new vector function pg by imposing to p̂b a rotation around
b by an angle −ϕ(φ) (opposite to the one really experienced by the gear) and then
restoring the initial orientation of b to b0

pg (ξ, θ, φ) = R R(p̂b (ξ, θ, φ), b(φ), −ϕ(φ)), r b (φ), −εb (φ) (4.48)

where r b (φ) and εb (φ) were defined in Eq. (4.28). In gear generation the two
angles ψ and ϕ of the tool and of the generated gear are related to the parameter
of motion φ; since the possibility of a modified roll, i.e. variable gear ratio τ (φ),
is taken into account, generic expressions for the two angles ψ(φ) and ϕ(φ) are
assumed.
Equation (4.48) represents the family of surfaces Φg whose envelope we are
interested in. If Pg denotes the generic point of Φg , the corresponding position
vectors pg = Pg −Og are therefore given by (4.48) as well as by any of the following
expressions

pg (ξ, θ, φ) = R R(p̂a (ξ, θ, φ) − d(φ), b(φ), −ϕ(φ)), r b (φ), −εb (φ)

= R R(R(R(pe (ξ, θ) + e(φ), r a (φ), εa (φ)), a(φ), ψ(φ) +

− d(φ), b(φ), −ϕ(φ)), r b (φ), −εb (φ) (4.49)
58 4. Generation with supplemental motions

This definition as well is purely geometrical and does not require reference
systems. It is a little bit complicated because it requires a sequence of four rigid
rotations, two for each gear: one for positioning the working axis and another one
around it. As in Eq. (4.35), we introduce a compact operator for the sequence of
rotations 
Rb (v) = R R(v, r b (φ), εb (φ)), b(φ), ϕ(φ) (4.50)
and for the inverse case

Rbi (v) = R R(v, b(φ), −ϕ(φ)), r b (φ), −εb (φ) (4.51)

so that Eq. (4.49) becomes

pg (ξ, θ, φ) = Rbi (p̂b ) = Rbi (Ra (pe (ξ, θ) + e(φ)) − d) (4.52)

Moreover introducing a total rotation operator and its inverse

Rba (v) = Rai (Rb (v)) (4.53)


i
Rab (v) = Rba (v) = Rbi (Ra (v)) (4.54)

Eq. (4.52) becomes

pg (ξ, θ, φ) = Rab (pe (ξ, θ) + e(φ)) − Rbi (d(φ)) (4.55)

which generalized in a non trivial way Eq. (25) in [11].

4.4.1 Derivatives of the enveloping family of surfaces


The derivatives of the enveloping family of surfaces Φg with respect to ξ and θ can
be easily determined exploiting Eqs. (4.39), (4.55) and property (4.9),

pg ,ξ = Rbi (p̂b ,ξ ) = Rbi (Ra (pe ,ξ (ξ, θ)) = Rab (pe ,ξ (ξ, θ))
(4.56)
pg ,θ = Rbi (p̂b ,θ ) = Rbi (Ra (pe ,θ (ξ, θ)) = Rab (pe ,θ (ξ, θ))

According to equations (4.39), (4.46), (4.56) and (4.131), the normal vector
mg to each regular surface of the family Φg is given by

mg (ξ, θ, φ) = pg ,ξ × pg ,θ = Rbi (m̂(ξ, θ, φ)) = Rab (me (ξ, θ)) (4.57)

which also shows the relationship between the normal vectors defined so far.
The evaluation of the derivative of pg with respect to the parameter of motion φ
is more complicated but we can take advantage of result (4.44). Employing the
general property (4.18), let us first write the derivative of p̂b = Rb (pg ) as

p̂b ,φ = wb × p̂b + Rb (pg ,φ ) (4.58)


4.4 The enveloping family of surfaces 59

where

w b = ϕ0 b + ω b (4.59)

ωb = R (σb0 j b + ζb0 k), k, ζb (4.60)

It can be also proven that for the inverse sequence of rotations the derivative of
pg = Rbi (p̂b ) is

pg ,φ = wib × pg + Rbi (p̂b ,φ ) = −Rbi (wb ) × pg + Rbi (p̂b ,φ )


= Rbi (p̂b ,φ − wb × p̂b ) (4.61)

since

wib = −Rbi (wb ) = −Rbi (ϕ0 b + ω b ) = −(ϕ0 b0 + ω ib ) (4.62)



ω ib = R (σb0 j b + ζb0 k), j b , −σb (φ) (4.63)

According to Eq. (4.44), equation (4.61) becomes

pg ,φ = Rbi (p̂a ,φ − d0 − wb × p̂b ) (4.64)


= Rbi (wa 0
× p̂a + Ra (e ) − d − wb × p̂b ) 0
(4.65)

Introducing the vector function

ĥ(ξ, θ, φ) = wa × p̂a + Ra (e0 ) − d0 − wb × p̂b (4.66)


0 0
= (wa − wb ) × p̂a + Ra (e ) − d + wb × d
= w × p̂a + Ra (e0 ) − d0 + wb × d (4.67)

where w = wa − wb , the derivative pg ,φ can be simply written as

pg ,φ (ξ, θ, φ) = Rbi (ĥ(ξ, θ, φ)) (4.68)

This result generalizes Eq. (30) in [11].


In curvature analysis we will need also the derivatives of the unit normal vector

mug ,ξ = Rbi (m̂u ,ξ ) = Rbi (Ra (mue ,ξ ) = Rab (mue ,ξ ) (4.69)


mug ,θ = Rbi (m̂u ,θ ) = Rbi (Ra (mue ,θ ) = Rab (mue ,θ ) (4.70)
mug ,φ = Rbi (m̂u ,φ ) − Rbi (wb ) × mug = Rbi (wa × m̂u ) − Rbi (wb ) × mug
u u
= Rbi ((wa − wb ) × m̂ ) = Rbi (w × m̂ ) (4.71)
60 4. Generation with supplemental motions

4.5 Equation of meshing


In order to determine the envelope surface Γg of the family Φg in the Euclidean
space E3g , the equation of meshing f = 0 is required. It is usually defined (e.g., [48])
as the triple product of the partial derivatives of the position vectors pg (ξ, θ, φ)
h i
pg ,ξ pg ,θ pg ,φ = mg · pg ,φ = f (ξ, θ, φ) = 0 (4.72)

Taking advantage of previous results (4.56) and (4.68), along with the general
property (4.132), equivalent expressions, simpler than (4.72), can be obtained
h i
f (ξ, θ, φ) = Rbi (p̂b ,ξ ) Rbi (p̂b ,θ ) Rbi (ĥ)
h i h i
= p̂b ,ξ p̂b ,θ ĥ = p̂a ,ξ p̂a ,θ ĥ

= m̂ · ĥ = 0 (4.73)
corresponding to the so-called “engineering approach” [49, 50].
It is convenient to introduce a new vector

he (ξ, θ, φ) = R R(ĥ(ξ, θ, φ), a, −ψ), r a , −εa = Rai (ĥ) (4.74)
= Rai (w 0 0
× p̂a + Ra (e ) − d + wb × d)
= we (φ) × pe (ξ, θ) + q e (φ) (4.75)
where
q e (φ) = Rai (−d0 (φ) + wb (φ) × d(φ)) + e0 (φ) (4.76)
we (φ) = Rai (w(φ)) = Rai (wa (φ) − wb (φ)) (4.77)
so that, employing (4.39), we obtain
h i
f (ξ, θ, φ) = p̂a ,ξ p̂a ,θ ĥ
h i
= Rai (pe ,ξ ) Rai (pe ,θ ) Rai (he )
h i
= pe ,ξ (ξ, θ) pe ,θ (ξ, θ) he (ξ, θ, φ)

= me (ξ, θ) · he (ξ, θ, φ) = 0 (4.78)


It may be interesting to underline that in this last form of the equation of meshing
the parameter of motion φ appears only in one vector.
Sometimes a slightly different, although completely equivalent, form of the
equation of meshing can be employed where the normal vector is replaced by the
unit normal vector mue (ξ, θ), so that (4.78) becomes
f˜(ξ, θ, φ) = mue (ξ, θ) · he (ξ, θ, φ) = 0 (4.79)
4.6 The envelope surface Γg 61

For the equation of meshing to be satisfied, the three vectors in the triple
products (4.72), (4.73) and (4.78) must belong to the same two-dimensional
sub-space, that is to be such that

pg ,φ = α pg ,ξ + β pg ,θ (4.80)
ĥ = α p̂a ,ξ + β p̂a ,θ (4.81)
he = α pe ,ξ + β pe ,θ (4.82)

for suitable coefficients α = α(ξ, θ, φ) and β = β(ξ, θ, φ). Employing (4.82), it is


possible to obtain α(ξ, θ, φ) and β(ξ, θ, φ) as
h i
he pe ,θ me
α(ξ, θ, φ) = (4.83)
me · me
h i
he pe ,ξ me
β(ξ, θ, φ) = − (4.84)
me · me
formally like Eq. (45) in [11], but where he hides a much more general kind of
relative motion.

4.6 The envelope surface Γg


In the previous sections, the basic foundations for the general spatial motion have
been given. It is now rather straightforward to follow the steps in gear generation
described in [11], where more details and comments are given, especially for the
comparison with the literature. First of all let us express the position vectors sg
of the sought for envelope surface Γg as
( 
sg = pg ξ, θ, φ
(4.85)
f˜(ξ, θ, φ) = 0

Since the fulfillment of the equation of meshing is only a necessary condition for
the existence on an envelope surface, we may need also to express the sufficient
condition. It means that we must assume the local existence of

ξ = ξ(θ, φ) or θ = θ(ξ, φ) (4.86)

while φ = φ(ξ, θ) cannot be employed in general since, as we will see, we have to


allow the possibility for f˜,φ (ξ, θ, φ) = 0. According to Dini’s theorem on implicit
functions, the conditions (4.86), assuming f˜ ∈ C1 , correspond to

|f˜,ξ (ξ, θ, φ)| + |f˜,θ (ξ, θ, φ)| =


6 0 (4.87)
62 4. Generation with supplemental motions

which is exactly the sufficient condition for the envelope generating process. The
expressions for the derivatives of f˜ are reported in the Appendix 4.10.
Adopting the first condition in (4.86), it is possible to define the envelope
surface Γg (i.e., the gear surface) by means of position vectors sg given by the
explicit function 
sg (θ, φ) = pg ξ(θ, φ), θ, φ (4.88)
The regularity of the envelope surface Γg can be investigated analyzing its
normal vector ng (θ, φ) which, by definition, is given by

ng (θ, φ) = sg ,θ × sg ,φ (4.89)

where, employing Dini’s theorem

f˜,θ f˜,φ
sg , θ = − pg ,ξ + pg ,θ and sg ,φ = − pg ,ξ + pg ,φ (4.90)
f˜,ξ f˜,
ξ

Substituting in Eq. (4.90) in Eq. (4.89), the normal vector ng becomes



1  
˜

˜
 
˜
ng (θ, φ) = pg ,ξ × pg ,θ f ,φ + pg ,φ × pg ,ξ f ,θ + pg ,θ × pg ,φ f ,ξ
f˜,ξ ξ=ξ(θ,φ)

1 
= f,φ − α f˜,ξ − β f˜,θ mg
˜
f ,ξ ξ=ξ(θ,φ)

1 ˜
= (f ,φ − α f˜,ξ − β f˜,θ ) Rab (me ) ξ=ξ(θ,φ) (4.91)
˜
f, ξ

where (4.57) and (4.80) have been taken into account respectively in the second
and in the last steps.
The condition for the regularity of the envelope surface Γg (i.e., no undercut-
ting) is
ng (θ, φ) 6= 0 (4.92)
that, according to (4.91) and since me 6= 0, is equivalent to the scalar inequality
 
g(ξ, θ, φ) = f˜,φ − α f˜,ξ − β f˜,θ 6= 0 (4.93)

with ξ = ξ(θ, φ). The scalar function g(ξ, θ, φ) can also be obtained as
q · me
g(ξ, θ, φ) = f˜,φ − α f˜,ξ − β f˜,θ = (4.94)
me · me
where
  
q(ξ, θ, φ) = pe ,ξ × pe ,θ f˜,φ + he × pe ,ξ f˜,θ + pe ,θ × he f˜,ξ (4.95)
4.7 Kinematic analysis of the enveloping process 63

Equation (4.94) follows from the observation that the first and the last lines in
(4.91) can be written as

1 1
Rab (q) = g Rab (me ) (4.96)
˜
f ,ξ ˜
f ,ξ

from which g(ξ, θ, φ) can be obtained doing a dot product of both terms with
Rab (me ) and employing property (4.130).
It is worth noting that the dot product q(ξ, θ, φ) · me (ξ, θ) corresponds to the
so called limit function of the first kind, sometimes written as

p , · p , pe ,ξ · pe ,θ pe ,ξ · he E F pe ,ξ ·he
eξ eξ
pe ,θ · he = F pe ,θ ·he
q(ξ, θ, φ)·me (ξ, θ) = pe ,θ · pe ,ξ pe ,θ · pe ,θ G

f˜,ξ f˜,θ f˜,φ f˜,ξ f˜,θ f˜,φ
(4.97)
From the investigation on the regularity of the surface Γg , employing Eqs. (4.49),
(4.79) and (4.94), we can obtain the following definition for lines of singular
points lg (that is points of undercutting, where ng vanishes)

lg = pg (ξ, θ, φ)

f˜(ξ, θ, φ) = 0 (4.98)


g(ξ, θ, φ) = 0

4.7 Kinematic analysis of the enveloping process


Since most of the literature dealing with gear generation employs kinematic re-
lationships, this section may be useful to complete this study and ease the com-
parison with other papers. We assume that time acts through the parameter of
motion, that is φ = φ(t).
Selecting a fixed point Of in the fixed space E3f , we can define the position
vectors of the points Oa and Ob as

da = Oa (φ) − Of , db = Ob (φ) − Of (4.99)

therefore their velocities are

d˙a = φ̇ d0a , d˙b = φ̇ d0b = φ̇ (d0a + d0 ) (4.100)

where, we recall, d = Ob − Oa . Still in the fixed space, the velocity of a point


P̌e (ξˇe , θ̌e ) of the moving tool Σ̂e is

v e (ξˇe , θ̌e , φ) = φ̇ wa × Ra (pe (ξˇe , θ̌e ) + e) + Ra (e0 ) + d0a

= φ̇ wa × p̂a (ξˇe , θ̌e , φ) + Ra (e0 ) + d0a (4.101)
64 4. Generation with supplemental motions

and, according to (4.45), the angular velocity of Σ̂e is

ω e = φ̇ wa (4.102)

In a similar way the velocity of a point P̌g (ξˇg (θ̌g , φ̌g ), θ̌g , φ̌g ), with f (ξˇg , θ̌g , φ̌g ) = 0,
of the generated surface is
  
v g (θ̌g , φ̌g , φ) = φ̇ wb × Rb pg (ξˇg (θ̌g , φ̌g ), θ̌g , φ̌g ) + d0b

= φ̇ wb × p̂ (ξˇg (θ̌g , φ̌g ), θ̌g , φ) + d0
b b (4.103)

and the gear angular velocity, for (4.59), is

ω g = φ̇ wb (4.104)

4.7.1 Sliding velocity between mating surfaces


It is interesting to find the expression of the sliding velocity between the surface
Σe of the tool and the surface Γg of the gear at their mating points, that is at
points of contact which are identified by the same parametric coordinates, i.e.

ξˇe = ξˇg = ξˇ θ̌e = θ̌g = θ̌ φ = φ̌g = φ̌ (4.105)

The sliding velocity is given by the difference of the velocity of the pinion (tool)
and of the gear
ˇ θ̌, φ̌), θ̌, φ̌) − v g (θ̌, φ̌, φ̌)
v seg (θ̌, φ̌) = v e (ξ(
 
= φ̇ wa × p̂a + Ra (e0 ) + d0a − wb × p̂b + d0b (4.106)

that according to Eqs. (4.66) and (4.100) can be written as

ˇ θ̌, φ̌), θ̌, φ̌)


v seg (θ̌, φ̌) = φ̇ ĥ(ξ( (4.107)

which clarifies the physical meaning of the vector ĥ. As a consequence he and
pg ,φ are parallel to the sliding velocity as seen from the pinion and from the gear
surfaces, respectively. From Eq. (4.107) we obtain that pure rolling between the
two mating gears happens at those points where ĥ(ξ( ˇ θ̌, φ̌), θ̌, φ̌) = 0.

4.7.2 Screw axis of the relative motion


In Eq. (4.67) the (non-unit) vector w = wa − wb has been introduced, which gives
the direction in E3f of the screw axis of the relative motion between the gear pair.
Therefore the relative angular velocity Ωr of the tool with respect to the gear is
given by
Ωr = φ̇ (wa − wb ) = φ̇ w (4.108)
4.8 Application to the face-milling process 65

In order to completely define the screw axis, one of its points H must be deter-
mined, knowing that, by definition, its relative velocity v H eg must be parallel to Ωr
or null that is
 H 
vH ˇ H ˇ
eg × Ωr = v e (ξ(θ̌, φ̌), θ̌, φ) − v g (ξ(θ̌, φ̌), θ̌, φ) × Ωr

ˇ θ̌, φ̌), θ̌, φ) + Ra (e0 ) + d0 +
= φ̇ wa × p̂a (ξ( a

ˇ 0
− wb × p̂b (ξ(θ̌, φ̌), θ̌, φ) + db × Ωr
 
= φ̇ w × p̂ (ξ(ˇ θ̌, φ̌), θ̌, φ) + Ra (e0 ) − d0 + wb × d × Ωr = 0 (4.109)
a

Since p̂a = H − Oa this equation can be written as


 
w × (H − Oa ) + Ra (e0 ) − d0 + wb × d × w = 0
  
w × (H − Oa ) × w = − Ra (e0 ) + d0 − wb × d × w (4.110)

w2 (H − Oa ) − (w · (H − Oa ))w = d0 − Ra (e0 ) − wb × d × w

Looking for the special point H satisfying the condition w · (H − Oa ) = 0 we have



d0 − Ra (e0 ) − wb × d × w
H − Oa = (4.111)
w2

4.8 Application to the face-milling process


In order to show the main applicative features of the proposed approach, even in
case of general motions, it is applied to the grinding process of a face-milled spiral
bevel pinion. The pinion modelled is a component of a real aerospace gearbox of
the Avio firm.
With reference to the notation already introduced in Section 4.3, let a be the
unit vector of the machine cradle axis and b(φ) the one of the pinion blank. It
is well known in gear literature, e.g. [77] and [19], that during generation of a
spiral bevel pinion axes a and b(φ) are skew and can continuously change their
relative position and orientation. The fixed reference point Oa and the moving
reference point Ob (φ) are not taken on the line of shortest distance between a and
b, but are displaced with respect to it, as shown in the schematic representation
of the geometric setup given in Fig. 4.3. Point Oa , which is a fixed point on the
fixed axis of the machine is shifted along a by a variable parameter called sliding
base ∆XB (φ), while point Ob (φ), fixed on the moving axis b, is moved along b(φ)
by the machine center to back ∆XD , which has a constant value. The quantity
∆EM (φ), called blank offset, is indeed the shortest distance between a and b(φ)
and is a function of the motion parameter φ. In the Gleason’s implementation
of the face-milling process, blank offset ∆EM (φ) and sliding base ∆XB (φ) are
polynomial functions of the motion parameter.
The cutter head surface is allowed to change its initial orientation with respect
to a reference position according to two subsequent rotations about the fixed unit
66 4. Generation with supplemental motions

head-cutter

machine
frame
zo
O e(f)
kc
cutter
axis
Se
so S r(f)
j

jc i
q k gear
blank
f
cradle Oa blank
axis axis
a j(f)
DE 
(f)
g DX
g


(f) 

DX 
(f) s (f)


Ob (f) b (f)

line of shortest
distance

Figure 4.3: Geometric setup for the generation of the spiral bevel pinion.

vectors j c and kc . In the proposed approach, it means that



pe (ξ, θ) = R R(p0e (ξ, θ), j c , σ0 ), kc , ζ0 , (4.112)

where the constant angles σ0 and ζ0 are called, respectively, tilt and swivel, and
j c and kc mark the tilt and swivel axes, as shown in Fig. 4.3. The vector function
p0e (ξ, θ) represents the tool surface in its reference configuration. As previously
stated, the distance between the cutter and the cradle axes, that is vector e(φ) =
Ôe (φ) − Oa , is a variable quantity called radial installment of the head cutter and
the motion related to its variation is called modified radial motion.
Functions ψ(φ) and ϕ(φ) controlling the rotation of the cradle and the blank,
respectively, specialize now in
 
2C 2 6D 3 24E 4 120F 5
ψ(φ) = φ; ϕ(φ) = m0 φ − φ − φ − φ − φ , (4.113)
2 6 24 120

where the last expression is generally referred to as modified roll function and the
coefficients within as modified roll coefficients [77]. In the most general framework
of UMC (Universal Motion Concept) type machines, also the machine root angle
4.8 Application to the face-milling process 67

can vary during generation and it is a function of the motion parameter of the
following form
1 1
γm (φ) = γm0 + G1 φ + G2 φ2 + G3 φ3 = γm0 − σb (φ), (4.114)
2 3
where γm0 is the basic machine root angle and σb (φ) is one of the angles employed
throughout the paper.
To actually perform the computations with the proposed approach we need to
define, in the linear space R3 , a single (fixed) reference frame S = (O; x, y, z), with
unit vectors (i, j, k). It is worth noting that all computations will be carried out
using just this unique reference system. We may select k = a and
a × b0 a × b0
j= = , i=j×k (4.115)
sin γ0 |a × b0 |
where γ0 is the angle between a and b0 = b(0). Angle γ0 is generally expressed
with respect to the machine root angle as γ0 = π/2 − γm0 .

4.8.1 Vector calculations and numerical example


In the unique reference system S of R3 the components of generating gear and
pinion blank unit vectors a, b0 and b are, respectively,
a = (0, 0, 1), b0 = (cos γm0 , 0, sin γm0 ), b = R(b0 , j, σb (φ)) (4.116)
since ζb (φ) = 0, and the components of the variable vector d(φ) = Ob (φ) − Oa
become
d(φ) = (∆XD cos γm (φ), −∆EM (φ), ∆XB (φ) + ∆XD sin γm (φ)) (4.117)
The components of the tilt and swivel axes j c and kc are then
j c = (cos q, sin q, 0), kc = ( 0, 0, −1) (4.118)
In this application a generating tool Σe is considered with a circular blade
profile and a circular fillet at the top. The parametric equations in S of the entire
surface of the grinder in its reference configuration are
   
Xf ± ρf sin ρξf cos θ
   
p0e (ξ, θ) =  Xf ± ρf sin ρξf sin θ  , 0 ≤ ξ ≤ ξ¯
 
(4.119)
     
ξ
−ρf 1 − cos ρf
   
Xp ∓ R1 cos π2 − λf + ξ− ξ̄
R1  cos θ
   
 ¯
p0e (ξ, θ) =  Xp ∓ R1 cos π2 − λf + ξ− ξ̄

R1 sin θ , ξ < ξ (4.120)
   
Zp − R1 sin π2 − λf + ξ−R1
ξ̄
68 4. Generation with supplemental motions

Figure 4.4: Aerospace pinion modelled.

where ξ¯ = λf ρf , θ ∈ [0, 2 π), λf is the pressure angle, the parameters Xf , ρf ,


Xp , Zp , R1 are related to the geometry of the grinding wheel and the upper and
lower signs refer to the concave (outside blade) and convex side (inside blade),
respectively.
In order to show the ease of use of the proposed approach we will be briefly
outline the methodology to obtain directly in the reference frame S the equations of
the generated surface of the pinion. By employing definitions (4.112) and (4.119)
the surface Σe is a known object along with its normal vector me (ξ, θ), with
components me (ξ, θ). Vector function he (ξ, θ, φ), whose expression in components
is given by

he (ξ, θ) = we (φ) × (pe (ξ, θ) + e(φ)) + qe (ξ, θ) (4.121)

is straightforward to calculate once the following quantities have been evaluated

wa = a, wb (φ) = σb0 (φ)j + ϕ0 (φ) b(φ), w(φ) = wa − wb (φ)


(4.122)

we (φ) = R(w(φ), a, −φ), qe (φ) = R(wb (φ) × d(φ) − d0 (φ), a, −φ) + e0 (φ)
(4.123)

Therefore, the equation of meshing is readily available as follows

f (ξ, θ, φ) = me (ξ, θ) · he (ξ, θ) = 0 (4.124)


4.9 Conclusions 69

The enveloping family of surfaces, according to (4.55), specializes now in the fol-
lowing expression

pg (ξ, θ, φ) = Rab (pe (ξ, θ) + e(φ)) − Rbi (d(φ)), (4.125)

where Rab (v) and Rbi (v) were defined in (4.50)–(4.53). The position vectors of the
actual surface of the generated gear are then sampled by solving the system
( 
sg = pg ξ, θ, φ
(4.126)
f (ξ, θ, φ) = 0

The CAD model of a spiral bevel pinion for aerospace application obtained with
the proposed approach is shown in Fig. 4.4.

4.9 Conclusions
In the first part the invariant approach to the theory of gearing formulated in
Chapters 2 and 3 has been extended to generation with variable axis distance and
generalized gear ratio. The enveloping process has been described, obtaining the
main features of the generated surface, like singular points and kinematic parame-
ters. The ease of use, compactness and computational efficiency of the proposed
approach have been illustrated with a numerical example of a real aerospace spiral
bevel pinion.
70 4. Generation with supplemental motions

4.10 Appendix
Rotating vectors
The rotation operator has many important and useful properties, that are em-
ployed in this approach.
Given the rotated vectors
û = R(u, a, α) v̂ = R(v, a, α) ŵ = R(w, a, α) (4.127)
we immediately have the following algebraic properties
u = R(R(u, a, α), a, −α) (4.128)
û + v̂ = R(u + v, a, α) (4.129)
û · v̂ = u · v (4.130)
v̂ × ŵ = R(v × w, a, α) (4.131)
 
û · (v̂ × ŵ) = u · (v × w) = u v w (4.132)

Derivatives of the equation of meshing


By employing the definition (4.79) of f˜, the partial derivative f˜,ξ is
f˜,ξ = mue ,ξ ·he + mue · he ,ξ (4.133)
where the derivative he,ξ is easily computed according to the general property
(4.9) and definition (4.75) as
he ,ξ (ξ, θ, φ) = we (φ) × pe ,ξ (ξ, θ) (4.134)
Similarly we obtain
f˜,θ = mue ,θ ·he + mue · he ,θ (4.135)
he ,θ (ξ, θ, φ) = we (φ) × pe ,θ (ξ, θ) (4.136)
The derivative f˜,φ is
f˜,φ = mue · he,φ (4.137)
where, owing to (4.18) and (4.74)
he,φ = w0e (φ) × pe (ξ, θ) + q 0e (φ) (4.138)
with
q 0e (φ) = Rai (−d00 + w0b × d + wb × d0 ) − Rai (wa ) × q e + e00 (φ) (4.139)

w0e (φ) = Rai (w(φ)),φ = Rai (w0 (φ)) − Rai (wa × w)

= Rai (w0 (φ) + wa × wb )) (4.140)


having employed Eqs. (4.61), (4.76) and (4.77).
Chapter 5

Curvature analysis of gear


surfaces

In order to evaluate contact stresses, contact deformations and minimum film


thickness of lubricant between mechanical components in direct contact, a curva-
ture analysis of the surfaces of the mating bodies has to be performed. It is not
difficult to find the normal curvature of a surface by means of well established
results of classical differential geometry [33]. Disadvantages in the application of
the classical differential geometry approach arise though when the surface under
investigation is generated as the envelope of a one parameter family of surfaces, as
in gear generation. One of the main problems is to solve the equation of meshing
with respect to one of the parameters involved and find an explicit function to be
substituted in the equation of the family [48]. Another problem is the extreme
complexity of the equations of the generated surface and its derivatives, which
inevitably require lengthy computations in order to be evaluated, lowering the
efficiency of the contact algorithms.
An alternative to the differential geometry approach has been sought by many
researchers. Since the generated surface is uniquely defined by the geometry of the
tool and the relative motion, Litvin proposed in [49, 39, 18] a systematic method-
ology to derive relationships between the principal normal curvatures of surfaces
in continuous tangency undergoing a relative motion with constant translational
and variable angular velocity. In this now classical approach, based on kinematic
relationships, curvatures are obtained following an indirect way, moving from the
analysis of motion of the contact points and requiring a line or a point contact
between the surfaces. Another feature of Litvin’s approach is that it does not
use the parametric coordinates of the surfaces, obtaining complex though general
expressions in terms of vector components expressed in an orthogonal reference
frame.
In [9], N. Chen follows Litvin’s approach introducing non-principal parametric
72 5. Curvature analysis of gear surfaces

coordinates of the surfaces and develops curvature expression in a given direction


in a non-orthogonal reference frame. Extensions to generalized motions in three
dimensions of the mating surfaces are due to C. H. Chen et al. in [8], again with
a deep recourse to vector components.
Wu and Luo in [84] proposed curvatures equations in terms of the limit func-
tions of the first kind and considering the mating surfaces subjected to relative
screw motion with constant translational and rotational velocities. Yan and Cheng
in [85] applied curvature analysis equations to some cam-follower mechanisms
based on the approach proposed in [84]. Ito and Takahashi in [30] investigated
curvatures in hypoid gears starting from a classical differential geometry point of
view, but then introducing kinematic relationships. By employing the theory of
screws, Dooner proposed in [15, 14] the third law of gearing and formulated the
limiting relationship between the radii of curvature of conjugate surfaces, which is
valid only for the reference pitch surfaces.
Formulae with strong assumptions about the possible motions of the axes of
the meshing bodies or on the functions controlling the rotation around the axes
are not adequate to represent the relationships between the normal curvatures of
surfaces generated on computer numerically controlled (CNC) machine with non-
uniform surface correcting motions. The present tendency on CNC gear cutting
or milling machines is, in fact, to employ simple tool geometry and adopt complex
correction motions in order to obtain favorably disposed LTCA contact patterns,
prescribed functions of transmission error, robustness under misalignments and
low vibration levels of the pair. It is therefore essential to develop a model for the
curvature relations capable to accomodate the most general conditions of motion.
In this chapter a rather general formulation for the analysis of curvature for
conjugate surfaces is proposed based on the invariant approach formulated in
Chapter 2 and extended in Chapters 3 and 4. The only assumptions are that
the orientation of the rotation axes of the mating bodies does not change during
generation. The functions controlling the relative position of the two axes and the
roll angle around the axes are not subjected to any restrictions. The main differ-
ence with respect to the previous approaches are that the formulation of equations
for the normal curvatures of the generated surface is accomplished without any
recourse to reference systems. Since, as shown in [11], reference systems play no
role in the theoretical development, their introduction can be postponed till the
very end of the treatment, that is when actual computations have to be done.
Moreover only one frame is required, thus avoiding the chain of reference systems
typical of the previous approaches. Another difference is that, since time does
not play any role in the geometry of the motion, that is the speed of the cutting
process does not affect the shape of the generated surface (at least when vibratory
motion effects on the cutting process are neglected), a parameter of motion will be
used as an independent variable controlling the relative motion between the tool
and the workpiece. Formulas for curvature and torsion of the generated surface are
obtained in a direct way, simply employing their definitions and some fundamental
5.1 Surface of the generating tool 73

properties due to the enveloping process.

5.1 Surface of the generating tool


In the Euclidean space E3e we define the generating tool to be a (fixed) regular
surface Σe . Its generic point will be denoted by Pe (ξ, θ), with (ξ, θ) ∈ A, where A
is an open set of R2 .
Once a fixed point Oe has been selected, it is possible to associate, as usual, to
each point Pe of E3e a position vector pe of R3

pe (ξ, θ) = Pe (ξ, θ) − Oe (5.1)

By definition, the normal vector me to Σe is given by

me (ξ, θ) = pe,ξ × pe,θ (5.2)

Owing to the assumed regularity of the surface Σe we always have

me 6= 0 (5.3)

Therefore, the unit normal vector mue is

me (ξ, θ) me (ξ, θ) me (ξ, θ)


mue (ξ, θ) = =√ = , (5.4)
|me (ξ, θ)| EG − F 2 D(ξ, θ)

where E, F, G are the coefficients of the first fundamental form of Σe [33, p. 82],
that is
E = pe,ξ · pe,ξ F = pe,ξ · pe,θ G = pe,θ · pe,θ (5.5)

and D = EG − F 2 .
The coefficients of the second fundamental form are given by [33, p. 119]

L = −mue ,ξ ·pe,ξ M = −mue ,ξ ·pe,θ = −mue ,θ ·pe,ξ N = −mue ,θ ·pe,θ


(5.6)
To investigate some geometric properties of surfaces, like normal curvatures, it
is often convenient to employ suitable curves on the surfaces involved [33, p. 121].
Therefore, let us consider an arbitrary regular curve ce on the tool surface Σe
defined by the position vectors

ce (se ) = pe (ξe (se ), θe (se )), (5.7)

where ξe (se ) and θe (se ) are analytic functions and se is the arc length of ce .
The unit tangent vector to ce is given by
dce
te (se ) = = pe,ξ ξe,se + pe,θ θe,se = pe ,se (5.8)
dse
74 5. Curvature analysis of gear surfaces

and the derivative along ce of the unit normal vector is

dmue
= mue ,ξ ξe,se + mue ,θ θe,se = mue ,se (5.9)
dse

The normal curvature kne and the geodesic torsion τ e of Σe along ce can be
easily computed by means of the following well known formulas

kne = −mue ,se ·te (5.10)


 
τ e = mue mue ,se te (5.11)

It is worth mentioning that in most cases we are merely interested in the


computation of kne and τ e at a selected point, say pe (ξ, ˇ θ̌) with ξˇ = ξe (še ) and
θ̌ = θe (še ), and along a given direction marked in (5.8) by the two numbers
eξ = ξe,se (še ) and eθ = θe,se (še ). Therefore, we do not really need the expressions
of the two functions ξe (se ) and θe (se ). Similar ideas can also be found in [9, Eqs.
(3) and (4)].
At a given point on Σe the principal directions g e1 and g e2 and the corresponding
principal curvatures k1e and k2e can be easily obtained [13]. Let λ1 , λ2 , (ε1 , χ1 ) and
(ε2 , χ2 ) be the eigenvalues and eigenvectors of the matrix
   −1  
a11 a12 E F L M
A= =− (5.12)
a21 a22 F G M N

Then we have
k1e = −λ1 k2e = −λ2 (5.13)
and
ε1 pe,ξ + χ1 pe,θ
g e1 =
|ε1 pe,ξ + χ1 pe,θ |
ε2 pe,ξ + χ2 pe,θ
g e2 = (5.14)
|ε2 pe,ξ + χ2 pe,θ |

Therefore, once obtained at a given point on Σe the principal directions g e1 and


g e2 and the corresponding principal curvatures k1e and k2e , we may write

te (se ) = cos β g e1 + sin β g e2 (5.15)

and, employing Rodrigues’ formula,

mue ,se = −(k1e cos β g e1 + k2e sin β g e2 ) = −(k1e (te · g e1 ) g e1 + k2e (te · g e2 ) g e2 ) (5.16)

where β is the angle, in the tangent plane, between g e1 and te , positive if a counter-
clockwise rotation around mue aligns g e1 onto te (we remind that g e1 × g e2 = mue ).
5.2 Curvatures of the generated gear 75

Moreover the normal curvature and the geodesic torsion defined in Eqs. (5.10)
and (5.11) can be calculated by means of Euler’s and Bertrand’s formulas, respec-
tively [33, p. 132 and p. 159],
kne = k1e cos2 β + k2e sin2 β (5.17)
1 e
τe = k = (k2e − k1e ) sin β cos β (5.18)
2 n,β

5.2 Curvatures of the generated gear


In this section we develop an invariant approach for the evaluation of curvature
features of Γg , in the neighborhood of any of its regular points, in terms of the
features of the tool Σe and the enveloping rigid-body motion, that is pe (ξ, θ), a,
b, ψ(φ), ϕ(φ) and d(φ).
Let us take a curve cg on the gear surface Γg , with arc length sg . Exactly like
in (3.64), we assume that three functions ξg (sg ), θg (sg ) and φg (sg ) exist such that
f˜(ξg (sg ), θg (sg ), φg (sg )) = 0 (5.19)
Therefore, the position vectors cg (sg ) of the curve cg are given by
cg (sg ) = pg (ξg (sg ), θg (sg ), φg (sg )) (5.20)
the unit tangent vector to cg is
dcg
tg (sg ) = = pg,ξ ξg,sg + pg,θ θg,sg + pg,φ φg,sg (5.21)
dsg
and the derivative along cg of the unit normal vector to Γg is
dmug
= mug ,ξ ξg,sg + mug ,θ θg,sg + mug ,φ φg,sg = mug ,sg (5.22)
dsg
As already stated in Section 5.1, typically we want to evaluate the normal
curvature at some selected points, say čg = pg (ξg (šg ), θg (šg ), φg (šg )), and along
a given direction ťg = tg (šg ). Therefore we select the point and the direction,
respectively, by means of the numbers (we do not really need the curve)
ξˇ = ξg (šg ), θ̌ = θg (šg ), φ̌ = φg (šg ) (5.23)
and
gξ = ξg,sg (šg ), gθ = θg,sg (šg ), gφ = φg,sg (šg ) (5.24)
ˇ
Of course they are not completely independent as (ξ, θ̌, φ̌) must satisfy the equation
of meshing f˜(ξ,
ˇ θ̌, φ̌) = 0, while the components (gξ , gθ , gφ ) in (5.21) must be such
that tg be of unit length (tg (šg ) · tg (šg ) = 1) and the total derivative (3.65) of the
equation of meshing vanishes (see page 45), that is
f˜,ξ gξ + f˜,θ gθ + f˜,φ gφ = 0 (5.25)
76 5. Curvature analysis of gear surfaces

5.2.1 Normal curvature


The normal curvature kng of Γg along cg is defined, like in (5.10), as

kng = −mug ,sg ·tg (5.26)

However, we wish to manipulate this formula in order to obtain a more convenient


expression. Owing to equations (3.43), (3.50), (3.53) and (3.54) (see page 42 and
thereafter), along with property (3.5) we obtain that
 
kng = − mue ,ξ gξ + mue ,θ gθ + we × mue gφ · pe,ξ gξ + pe,θ gθ + he gφ (5.27)

This is a computable expression where everything is evaluated at (ξ, ˇ θ̌, φ̌) and it
is in terms of the tool shape and the generating relative motion.
However, in many cases it is perhaps simpler to choose the direction for the
evaluation of kng on the tool surface Σe . Therefore we need the unit vector

ˇ θ̌, φ̌), b, ϕ(φ̌)), a, −ψ(φ̌)
te = R R(tg (ξ, (5.28)

ˇ θ̌, φ̌), the same direction


to mark, when the tool and the gear come in contact at (ξ,
of tg .
By construction, te is tangent to the tool surface Σe at pe (ξ,ˇ θ̌). Accordingly
it can be written as
te = pe,ξ eξ + pe,θ eθ (5.29)
that is as a linear combination of pe,ξ (ξ,ˇ θ̌) and p (ξ, ˇ θ̌) with suitable scalars eξ
e,θ
and eθ . We remark that in this framework they are numbers, although they could
also be seen as derivatives of functions like in (5.8). For te to be of unit length we
require (see (5.5))
te · te = E e2ξ + 2F eξ eθ + G e2θ = 1 (5.30)
In many cases it is perhaps simpler to choose the direction for the evaluation of
kng by employing eξ and eθ .
Combining Eqs. (5.21), (5.28) and (5.29), we obtain that

pe,ξ eξ + pe,θ eθ = pe,ξ gξ + pe,θ gθ + he gφ (5.31)

that is
pe,ξ (eξ − gξ ) + pe,θ (eθ − gθ ) = he gφ (5.32)
If we take the scalar product of equation (5.31) with pe ,ξ and pe ,θ and we
make use of (3.65), we get the following system of three linear equations

 E gξ

 + F gθ + he · pe,ξ gφ = E eξ + F eθ
F gξ + G gθ + he · pe,θ gφ = F eξ + G eθ (5.33)


 f˜, g + f˜, g + f˜, g =0
ξ ξ θ θ φ φ
5.2 Curvatures of the generated gear 77

which is solvable at non singular points on Γg [11, eqn. (60)] since



E F he · pe ,ξ

1
g̃(ξ, θ, φ) = 2 F G he · pe ,θ 6= 0 (5.34)
D
f˜,ξ f˜,θ f˜,φ
For instance, gφ is given by

E F E eξ + F eθ E F 0
1 1
gφ = 2 F G F eξ + G eθ = − 2 F
G 0

D g̃ D g̃
f˜,ξ f˜,θ 0 f˜,ξ f˜,θ ˜ ˜
f ,ξ eξ + f ,θ eθ
1 f˜,φ
= − (f˜,ξ eξ + f˜,θ eθ ) = φ,se (5.35)
g̃ g̃
Therefore for the computation of kng we can employ the following formula once
we have obtained the three quantities gξ , gθ and gφ from eξ and eθ

kng = − mue ,ξ gξ + mue ,θ gθ + we × mue gφ · (pe,ξ eξ + pe,θ eθ ) (5.36)

5.2.2 Relative normal curvature


The relative normal curvature kneg between surfaces Σe and Γg at contact points
[48, p. 286] is, by definition, given by the following equation (cf. (5.10) and (5.26))
kneg = kne − kng = −(mue ,se ·te − mug ,sg ·tg ) (5.37)
provided equation (5.28) holds.
Employing equations (5.6), (5.9), (3.54), (3.70), (5.29), (5.32) and (5.36) we
obtain

kneg = − mue ,ξ eξ + mue ,θ eθ − (mue ,ξ gξ + mue ,θ gθ ) − we × mue gφ · te

= − mue ,ξ (eξ − gξ ) + mue ,θ (eθ − gθ ) − we × mue gφ · te
= −(mue ,ξ eξ + mue ,θ eθ ) · (pe,ξ (eξ − gξ ) + pe,θ (eθ − gθ )) + te · we × mue gφ
= −((mue ,ξ eξ + mue ,θ eθ ) · he + mue · we × te )gφ
= −(mue ,se ·he + mue · we × te )gφ
= −(k · te )gφ (5.38)
which in the last lines provides fairly compact and computable expressions that
fully exploits the enveloping process.
Moreover, because of (3.71) and (5.35), we also have
1 2 1
kneg = f˜,φ φ,se gφ = f˜,φ φ,se = k · te )2 (5.39)
g̃ g̃
which, inserting equations (5.15) and (3.68), simply becomes (cf. [84, p. 98])
1 2
kneg = λ cos β + µ sin β (5.40)

78 5. Curvature analysis of gear surfaces

5.2.3 Geodesic torsion


The geodesic torsion τ g of Γg along cg is defined, like in (5.11), by
 
τ g = mug mug ,sg tg (5.41)
Taking advantage of property (3.7), Eqs. (3.51) and (5.28) and following the steps
defined in the manipulation of Eq. (5.27), another form of the previous relation
can be obtained
  
τ g = mue mue ,ξ gξ + mue ,θ gθ + we × mue gφ te (5.42)
which involves features of the tool surface and the generating relative motion.
Again, as for Eq. (5.36), for the computation of τ g we can employ the previous
formula once we have determined the three quantities gξ , gθ and gφ from eξ and
eθ , solving system (5.33).

5.2.4 Relative geodesic torsion


The relative geodesic torsion τ eg between surfaces Σe and Γg at contact points is,
by definition (see, e.g., [33], [84]), given by the following equation (cf. (5.11) and
(5.41))    
τ eg = τ e − τ g = mue mue ,se te − mug mug ,sg tg (5.43)
provided equation (5.28) holds. Introducing the modified form for τ g defined in
(5.42) we obtain

τ eg = (te × mue ) · mue ,se − mue ,ξ gξ + mue ,θ gθ + we × mue gφ (5.44)
which can be further developed as shown in the Appendix to Chapter 5

τ eg = (te × mue ) · mue ,ξ eξ + mue ,θ eθ − mue ,ξ gξ + mue ,θ gθ + we × mue gφ

= (te × mue ) · mue ,ξ (eξ − gξ ) + mue ,θ (eθ − gθ ) − we × mue gφ

= mue ,ξ (eξ − gξ ) + mue ,θ (eθ − gθ ) × (pe,ξ eξ + pe,θ eθ ) · mue
− (te × mue ) · (we × mue gφ )
 u 
= (me ,ξ eξ + mue ,θ eθ ) × pe,ξ (eξ − gξ ) + pe,θ (eθ − gθ )

+ (mue ,ξ ×pe,θ − mue ,θ ×pe,ξ ) eθ (eξ − gξ ) − eξ (eθ − gθ ) · mue − te · we gφ
  
= mue ,se he mue − te · we gφ
 
+ (mue ,θ ×pe,ξ − mue ,ξ ×pe,θ ) eξ (eθ − gθ ) − eθ (eξ − gξ ) · mue
  
= he mue (mue ,se +(k1e + k2e )te ) − te · we gφ
 
= k te mue gφ (5.45)
As for the relative normal curvature, because of (3.71) and (5.35), we also have
1 
τ eg = − k mue te (k · te ) (5.46)

5.2 Curvatures of the generated gear 79

and, again inserting equations (5.15) and (3.68),


1
τ eg = (λ cos β + µ sin β)(µ cos β − λ sin β) (5.47)

5.2.5 Principal relative normal curvatures and directions


By comparing expressions Eqs. (5.38) and (5.45), we can observe that both the
relative normal curvature and the relative geodesic torsion vanish when gφ = 0,
that is, according to (5.35), when either one of the following conditions occurs
(
φ,se = 0
kneg = τ eg = 0 ⇐⇒ (5.48)
f˜,φ = 0

The first condition means that along a contact line, identified by dφ = 0, two
conjugate surfaces have (obviously) the same normal curvature, which is also a
relative principal direction since τ eg = 0 [48, p. 287], and all contact points are
parabolic [33, p. 125], [84, p. 100]. The second condition states that at points where
f˜,φ = 0, that is along the envelope curve of contact lines on the tool surface (e.g.,
[11, Section 13]), kneg = 0 in every direction, which means that the two conjugate
surfaces have a flat contact point, as also reported in [84, p. 100] (see also [33,
p. 136]).
As just stated, the tangent to the contact line is a relative principal direction
with null curvature. Owing to Eq. (3.73), the other relative principal direction
will therefore correspond to k and the relative curvature canpbe easily computed
replacing te in (5.39) with the unit vector parallel to k, i.e. k/ λ2 + µ2 . Summing
up, the principal relative normal curvatures between the conjugate surfaces Σe and
Γg are
k1eg = 0 along the contact line

λ2 + µ2 (5.49)
k2eg = normal to the contact line

The mean relative curvature H eg and the Gaussian relative curvature K eg are
1 eg λ2 + µ2
H eg = (k1 + k2eg ) =
2 2 g̃ (5.50)
K eg
= k1eg k2eg =0

5.2.6 Principal normal curvatures and directions


The normal curvature kng of the generated gear Γg can be directly computed by
means of Eq. (5.36), or calculated from the relative normal curvature as in (5.37).
Defining
1 1
H e = (k1e + k2e ), Re = (k1e − k2e ), (5.51)
2 2
80 5. Curvature analysis of gear surfaces

and employing (5.17), the normal curvature of the generated gear Γg can be written
as
1
kng = kne − kneg = H e + Re cos 2β − (λ cos β + µ sin β)2 (5.52)

Then, by definition, the geodetic torsion τ g on Γg is given by the following expres-
sion
 2 
1 g λ − µ2 λµ
τ g = kn,β = − Re sin 2β − cos 2β (5.53)
2 2 g̃ g̃
By requiring kng to assume the extreme values, that is requiring that τ g van-
ishes, it is straightforward to obtain
2λµ
τg = 0 ⇐⇒ tan 2β = (5.54)
(λ2 − µ2 ) − 2 g̃ Re
The principal directions of Γg on the tangent plane to Σe are therefore identified
by the angles
 
g 1 2λµ
β1 = arctan
2 (λ2 − µ2 ) − 2 g̃ Re
π
β2g = β1g + (5.55)
2
g
and the corresponding principal normal curvatures k1,2 are therefore given by

λ2 + µ2 1p 2
k1g , k2g = H e − ∓ (λ − µ2 − 2g̃Re )2 + (2 λ µ)2 . (5.56)
2g̃ 2g̃
The mean curvature H g and the Gaussian curvature K g of the generated surface
Γg can be readily obtained as
λ2 + µ2
H g = H e − H eg = H e −
2 g̃
(5.57)
k e µ2 + k2e λ2
Kg = Ke − 1

The same formulas can be found in [84, p. 104].

5.3 Application to spiral bevel gears


The convenience of the proposed approach is further illustrated by a numerical ex-
ample of a face-milled spiral bevel pinion generated on a Gleason’s CNC machine.
The pinion modelled is a component of a real aerospace drive. The methodol-
ogy described here is the extension to the curvature relations of the procedure
presented in [11].
5.3 Application to spiral bevel gears 81

5.3.1 Preliminary definitions

Adopting the notation already introduced in Section 3.3, let a be the axis of
the machine cradle and b the axis of the pinion blank. As well known in gear
literature [77], during generation of a spiral bevel pinion axes a and b are skew
and can experience a relative translating motion. As suggested in [11], points Oa
and Ob (φ) are not taken on the line of shortest distance, but according to common
practice they are displaced with respect to such line. A schematic representation
of the geometric set up during generation is given in Fig. 5.1.
Point Oa , which is a fixed point on the fixed axis of the machine, is moved
along a by a variable parameter called sliding base ∆XB1 (φ), while point Ob (φ) is
moved along b by the machine center to back ∆XD1 , which has a constant value.
Quantity ∆EM1 (φ), called blank offset, is indeed the shortest distance between a
and b and is a function of the motion parameter φ.
In the Gleason’s implementation of the face-milling process, blank offset ∆EM1 (φ)
and sliding base ∆XB1 (φ) are polynomial functions of the motion parameters of

head-cutter
machine
frame

cutter
axis

y
Sr


q 

y(f)
cradle Oa blank
axis a z axis
j(f)
D E M (f)


g 

D XB (f)


Ob (f)
D XD 

line of shortest
distance

Figure 5.1: Geometric set up for the generation of the spiral bevel pinion.
82 5. Curvature analysis of gear surfaces

the following types


1 1
∆EM1 (φ) = ∆EM10 + V1 φ + V2 φ2 + V3 φ3 (5.58)
2 3
1 1
∆XB1 (φ) = ∆XB10 + H1 φ + H2 φ2 + H3 φ3 (5.59)
2 3
where ∆EM10 and ∆XB10 are constant values and Vi and Hi (with i = 1, 2, 3) are
called, respectively, vertical and helical motion coefficients.
Functions ψ(φ) and ϕ(φ) controlling the rotation of the cradle and the blank,
respectively, specialize now in

ψ(φ) = φ, (5.60)
 
2C 2 6D 3 24E 4 120F 5
ϕ(φ) = m0 φ − φ − φ − φ − φ , (5.61)
2 6 24 120

where the last expression is generally referred to as modified roll function and the
coefficients within as modified roll coefficients [77].
To perform the computations with the proposed approach we need to define,
in the linear space R3 , a single (fixed) reference frame S = (O; x, y, z), with unit
vectors (i, j, k). It is worth noting that all computations will be carried out using
just this reference system.
We may select k = a and

1 a×b
j= a×b= , i=j×k (5.62)
sin γ |a × b|

where γ is the angle between a and b. Angle γ is generally expressed with respect
to the machine root angle γm1 = π/2 − γ.
In the unique reference system S the components of a and b are, respectively,

a = (0, 0, 1), b = (cos γm1 , 0, sin γm1 ) (5.63)

and the components of the variable vector d(φ) and its derivatives d0 (φ) and d00 (φ)
become

d(φ) = (∆XD1 cos γm1 , −∆EM1 (φ), ∆XB1 (φ) + ∆XD1 sin γm1 )
d0 (φ) = (0, −∆EM
0
1
0
(φ), ∆XB 1
(φ))
00 00 00
d (φ) = (0, −∆EM1 (φ), ∆XB1 (φ)) (5.64)

It is worth stressing that the definitions given here correspond basically to those
given in [11], but the components of the vectors just defined are now functions of
the parameter of motion φ, since the relative displacement between axes a and b
during generation must be accounted for.
5.3 Application to spiral bevel gears 83

In this application we consider a generating tool Σe with a circular blade profile


and a circular fillet at the top. The geometric parameter will be labelled in the
(a)
same manner as in [4]. The parametric equations in S of the flanks pe (ξ, θ) of
the tool and their unit normal vectors mue (a) (ξ, θ) are
   
(Xp ± R1 cos ξ) cos θ + Sr1 cos q1 cos ξ cos θ
pe(a) (ξ, θ) =  (Xp ± R1 cos ξ) sin θ + Sr1 cos q1  , mue (a) =  cos ξ sin θ  ,
   
Zp − R1 sin ξ ∓ sin ξ
(5.65)
¯
where ξ ∈ [ρf (1−sin αp )/(cos αp ), ξ(θ)], θ ∈ [0, 2 π), and the upper and lower signs
refer to the convex (inside blade) and concave side (outside blade), respectively.
(b)
The parametric equations of the fillets at the top of the tool pe (ξ, θ) and their
u (b)
unit normal vectors me (ξ, θ), again in S, are
   
(Xf ∓ ρf sin ξ) cos θ + Sr1 cos q1 sin ξ cos θ
p(b) mue (b) =  sin ξ sin θ  ,
   
e (ξ, θ) =  (Xf ∓ ρf sin ξ) sin θ + Sr1 cos q1  ,
−ρf (1 − cos ξ) ∓ cos ξ
(5.66)

where ξ ∈ [0, π/2 − αp ], θ ∈ [0, 2 π). Again the upper and lower signs must be
taken when considering the convex and concave side, respectively.

5.3.2 Preliminary computations


According to the procedure described in Section 5.2.5, we have first to calculate
certain quantities useful in the rest of the development. First of all, we evaluate
the principal normal curvatures k1e and k2e and the components g1e and g2e in S
of the principal directions g e1 and g e2 of the tool surface Σe , along with its first
coefficients E, F, G. Explicit formulas and numerical results will be provided only
for the convex active flank (conjugate to the inside blade of the tool).
Since Σe is a surface of revolution, it is easy to verify that the coordinate lines
(dθ = 0, dξ = 0) mark the principal normal directions on Σe . To evaluate the
principal normal curvatures of Σe , we can therefore apply the following simplified
formulas
pe ,ξ (ξ, θ) · mue (ξ, θ) 1
k1e = − =− (5.67)
E R1
pe ,θ (ξ, θ) · mue (ξ, θ) cos ξ
k2e (ξ) = − =− , (5.68)
G(ξ) Xp + R1 cos ξ
where explicit expressions of the first coefficients of Σe are

E = R12 , F = 0, G(ξ) = (Xp + R1 cos ξ)2 , (5.69)


84 5. Curvature analysis of gear surfaces

and the components of the principal directions in S become


   
− cos θ sin ξ − sin θ
pe ,ξ (ξ, θ)  pe ,θ (ξ, θ) 
g1e (ξ, θ) = g2e (θ) = p
 
√ =  − sin θ sin ξ  , =  cos θ (5.70)
E G(ξ)
− cos ξ 0

The components w(φ) in S of the vector w(φ), marking the direction of the screw
axis, are readily computed as follows
 
6D 2 24E 3 120F 4
w(φ) = a − ϕ0 (φ) b = a − m0 1 − 2Cφ − φ − φ − φ b (5.71)
2 6 24

The components he (ξ, θ, φ) and its derivative he ,φ (ξ, θ, φ) are computed by em-
ploying definitions (3.47) and (3.63), which give

he (ξ, θ, φ) = R(w(φ), a, −φ) × pe (ξ, θ) + R(ϕ0 (φ)b × d(φ) − d0 (φ), a, −φ)


= we (φ) × pe (ξ, θ) + qe (φ) (5.72)

and

he,φ = R(w0 − a × w, a, −φ) × pe



+ R ϕ00 b × d − a × (ϕ0 b × d) + (a + ϕ0 b) × d0 − d00 , a, −φ (5.73)

Figure 5.2: Aerospace pinion modelled.


5.3 Application to spiral bevel gears 85

We are now ready to evaluate the equation of meshing in the strict sense by
means of the following expression

f˜(ξ, θ, φ) = mue (ξ, θ) · he (ξ, θ, φ) = 0 (5.74)

Then, the important scalar functions λ(ξ, θ, φ) and µ(ξ, θ, φ) are computed taking
advantage of Eqs. (3.69), which specialize in
" # " # " #
λ(ξ, θ, φ) k1e he (ξ, θ, φ) we (φ) g1e (ξ, θ)
= · (5.75)
µ(ξ, θ, φ) −we (φ) k2e (ξ) he (ξ, θ, φ) g2e (θ)

Equations (3.58), (3.60) allow us to determine f˜,ξ and f˜,θ , which become

f˜,ξ (ξ, θ, φ) = − E λ(ξ, θ, φ) = −R1 λ(ξ, θ, φ) (5.76)
p
f˜,θ (ξ, θ, φ) = − G(ξ) λ(ξ, θ, φ) = −(Xp + R1 cos ξ) µ(ξ, θ, φ), (5.77)

while derivative f˜,φ is evaluated by employing (??) as follows

f˜,φ (ξ, θ, φ) = mue (ξ, θ) · he ,φ (ξ, θ, φ) (5.78)

The last step is the evaluation the function of singular points in the strict sense g̃.
This scalar function specializes now in

R2 0 E he · g1e
1
1

g̃(ξ, θ, φ) = 2 2 0 (Xp + R1 cos ξ)2 G he · g2e (5.79)
R1 (Xp + R1 cos ξ)
f˜,ξ f˜,θ f˜,φ

5.3.3 Numerical example


The proposed approach is employed to perform a curvature analysis of a spiral
bevel pinion for aerospace application shown in Fig. 5.2.
The main data of the transmission, the geometric parameters of the convex
side (inside blade) of the grinding wheel and the machine settings employed, are
given in Tables 5.1, 5.2 and 5.3. As a result of the proposed approach, the principal
radii of curvature R1g = 1/k1g and R2g = 1/k2g along the profile, computed employing
Eqs. (5.56), are plotted in Fig. 5.3 and 5.4.
In particular, Fig. 5.3 shows that the minimum radius of curvature R1g increases
in absolute value both travelling from the fillet to the top of the tooth and from the
front cone to the back cone. Fig. 5.4 shows that the maximum radius of curvature
R2g decreases in absolute value along the tooth profile, while increasing from the
front cone to the back cone in the lead direction. Finally, it is worth noting that
on a Pentium Centrino 1.5 GHz, with 1 GByte of RAM, a Mathematica program
implementing the proposed formulas, runs 200 times faster than the correspondent
using the classical differential geometry approach.
86 5. Curvature analysis of gear surfaces

5.4 Conclusions
In this chapter explicit formulas for the determination of principal relative normal
curvatures and directions of conjugate surfaces and principal normal curvatures
and directions of the generated tooth surface have been provided within the same
framework of the invariant approach. The ease of use, compactness and computa-
tional efficiency of the proposed approach have been illustrated with a numerical
calculation of the principal curvatures of a real aerospace spiral bevel pinion.

Acknowledgement
The support of Avio S.p.A. is gratefully acknowledged.

Parameter name Symbol Value


Pinion tooth number N1 27
Gear tooth number N2 38
Module m 4.950 mm
Mean spiral angle β 35.0 deg
Pinion Hand LH –
Face width Fw 32.0 mm
Outer cone distance A0 97.8706 mm
Pinion face angle γa 1 46.712 deg

Table 5.1: Main input data of the transmission modelled.

Parameter name Symbol Value


Spherical radius R1 173.7800 mm
Spherical center x-offset Xp -80.5632 mm
Spherical center z-offset Zp 60.8096 mm
Blade angle αp 20.4826 deg
Fillet radius ρf 0.7620 mm

Table 5.2: Geometric parameters of the inside blade (convex side).

Parameter name Symbol Value


Radial setting Sr 1 76.5559 mm
Blank offset ∆EM1 -1.7272 mm
Root angle γm 1 40.629 deg
Mach. center to back ∆XD1 3.521050 mm
Sliding base ∆XB1 -2.284300 mm
Cradle angle q1 57.9919 deg
Ratio of roll m0 1.512830
M.R. 1st coeff. 2C 0.004968
M.R. 2nd coeff. 6D -0.056380
M.R. 3rd coeff. 24E 0.000000
M.R. 4th coeff. 120F 0.000000

Table 5.3: Machine settings employed for grinding the pinion convex side.
5.4 Conclusions 87

-20

-25

Front cone
Radius R1, mm

-30
g

-35

-40

Back cone
-45

2 4 6 8 10
Fillet Point number on profile Top

Figure 5.3: Principal normal radius of curvature R1g of the pinion surface.

-87

-88

-89 Front cone


Radius R 2, mm

-90
g

-91

-92
Back cone

-93

-94

2 4 6 8 10
Fillet Point number on profile Top

Figure 5.4: Principal normal radius of curvature R2g of the pinion surface.
88 5. Curvature analysis of gear surfaces

5.5 Appendix
Proof of eq. (5.45)
From Eq. (5.32), cross multiplying by pe,ξ and then dot multiplying by mue (and
similarly with pe,θ ) we obtain

  gφ
eξ − gξ = mue he pe,θ
me
  gφ
eθ − gθ = − mue he pe,ξ
me

where me = me · mue . Accordingly, we have

eξ (eθ − gθ ) − eθ (eξ − gξ )
  gφ   gφ
= −eξ mue he pe,ξ − eθ mue he pe,θ
me me
 u
 gφ  u
 gφ
= he me pe,θ eθ + he me pe,ξ eξ
me me
 u
 gφ
= he me te
me

Moreover

mue ,θ ×pe,ξ − mue ,ξ ×pe,θ


 
= − k1e (pe ,θ ·g e1 ) g e1 + k2e (pe ,θ ·g e2 ) g e2 × (pe ,ξ ·g e1 ) g e1 + (pe ,ξ ·g e2 ) g e2
 
+ k1e (pe ,ξ ·g e1 ) g e1 + k2e (pe ,ξ ·g e2 ) g e2 × (pe ,θ ·g e1 ) g e1 + (pe ,θ ·g e2 ) g e2

= −k1e (pe ,θ ·g e1 )(pe ,ξ ·g e2 ) + k2e (pe ,θ ·g e2 )(pe ,ξ ·g e1 ) mue

+ k1e (pe ,θ ·g e2 )(pe ,ξ ·g e1 ) − k2e (pe ,θ ·g e1 )(pe ,ξ ·g e2 ) mue

= (k1e + k2e ) me · mue mue = (k1e + k2e ) me mue

Combining the last results


 u  
τ eg = me ,se he mue − te · we gφ
 
+ (mue ,θ ×pe,ξ − mue ,ξ ×pe,θ ) eξ (eθ − gθ ) − eθ (eξ − gξ ) · mue
  
= he mue mue ,se − te · we gφ
  gφ
+ (k1e + k2e ) me he mue te
me
  
= he mue (mue ,se +(k1e + k2e )te ) − te · we gφ

which is the next to last expression in (5.45).


5.5 Appendix 89

Furthermore, we can also include the vector k. First let us observe that
  
τ eg = he mue (mue ,se +(k1e + k2e )te ) − te · we gφ
  
= (g e1 · he )g e1 + (g e2 · he )g e2 mue (mue ,se +(k1e + k2e )te ) − te · we gφ
 

= −(g e1 · he )g e2 + (g e2 · he )g e1 · (mue ,se +(k1e + k2e )te ) − te · we gφ

where

mue ,se +(k1e + k2e )te = − k1e cos β g e1 + k2e sin β g e2 + (k1e + k2e )(cos β g e1 + sin β g e2 )

= k1e sin β g e2 + k2e cos β g e1

And hence
   
τ eg = −(g e1 · he )g e2 + (g e2 · he )g e1 · mue ,se +(k1e + k2e )te − te · we gφ
   
= −(g e1 · he )g e2 + (g e2 · he )g e1 · k1e sin β g e2 + k2e cos β g e1 − te · we gφ

= −k1e sin β(g e1 · he ) + k2e cos β(g e2 · he ) − te · we gφ

= −k1e sin β(g e1 · he ) + k2e cos β(g e2 · he ) − (g e1 · we ) cos β − (g e2 · we ) sin β gφ
   
= − k1e (g e1 · he ) + (g e2 · we ) sin β + k2e (g e2 · he ) − (g e1 · we ) cos β gφ

= −λ sin β + µ cos β gφ

= k · (sin βg e1 − cos βg e2 ) gφ

= k · te × me gφ

since

te × me = (cos β g e1 + sin β g e2 ) × me = − cos βg e2 + sin βg e1


Chapter 6

Curvature analysis: different


methods within the same
framework

6.1 Introduction
Curvature analysis of a surface can be carried out quite easily by means of well
established results of classical differential geometry, ([33] and [13]). However, their
application to the investigation of the surfaces of a gear pair is still a difficult task,
mainly because it requires solving the equation of meshing with respect to one
of the involved parameters [48]. Another difficulty is the extreme complexity of
the equations of the generated surface and its derivatives. Therefore, alternative
approaches have been sought by many researchers working on the theory of gearing.
They can be represented mainly by three different methods:
• the indirect meshing method proposed by Litvin [51];
• the indirect generating method proposed by Chen [9];
• the direct generating method proposed by Di Puccio et al. [12] and by Wu
and Luo [84].
In the indirect meshing method [51], relationships between the curvatures of
the gear pair are obtained starting from the analysis of the motion of the contact
point. That is why it is defined here as indirect. Since the two gears are not
necessarily obtained by means of an enveloping process, this method differs also
for considering meshing and not generating conditions. The method has been
proposed by Litvin employing kinematic relationships and scalar components of
the involved vectors. An extension to a generalized motion has been proposed in
[8].
92 6. Curvature analysis: different methods within the same framework

In the indirect generating method [9], Chen follows Litvin’s approach for the
case of enveloping process between the gears. He obtains some fundamental ex-
pressions, again in kinematic terms, for the case of fixed axes and constant gear
ratio.
The direct generating method has been recently proposed by the authors in
[12] and [11], and is similar to Wu and Luo’s approach [84]. The equations for
the normal curvature and geodesic torsion of the surface are obtained simply by
applying their definitions, therefore in a direct way, for the case of the enveloping
process.
Among other curvature analyses, two at least are worth mentioning: one was
proposed by Dooner in [14] and the other by Ito and Takahashi in [30]. In the
first one the author employs the screw algebra to derive the so called “third law of
gearing” and formulates the limiting relationship between the radii of curvature,
which is only valid, though, for the reference pitch surfaces. In the second one the
curvatures in hypoid gears are investigated starting from a classical differential
geometry point of view, but then kinematic relationships are introduced.
In this chapter all the three main approaches are extended to the case of trans-
lating axes and variable gear ratio, and re-written using a purely geometric ap-
proach [11], that does not require reference systems. A coordinate-free approach,
based on geometric algebra, has also been proposed by Miller in [69], but it only
deals with the generating process and curvature analysis is not addressed.
The new feature of this chapter is also the introduction of the tensor of curva-
ture which plays a key role in the basic equations.

6.2 Theoretical background: curvatures of a surface


Let us consider, in the Euclidean space E3 , a regular surface Σ whose generic
point will be denoted by P (ξ, θ), where (ξ, θ) are the parametric coordinates of
the surface. Once chosen a fixed point O of E3 , the position vector p of P (ξ, θ) is
p(ξ, θ) = P (ξ, θ) − O (6.1)
By definition, the normal vector m to Σ is given by
∂p(ξ, θ) ∂p(ξ, θ)
m(ξ, θ) = × = p,ξ ×p,θ (6.2)
∂ξ ∂θ
and, due to regularity of the surface, it is also m 6= 0. The normal unit vector to
the surface is
m(ξ, θ)
mu (ξ, θ) = (6.3)
|m(ξ, θ)|

6.2.1 Normal curvature and geodesic torsion of a surface


In differential geometry the second order properties of a surface, as curvature and
torsion, are conveniently derived from their evaluation for curves belonging to the
6.2 Theoretical background: curvatures of a surface 93

surface [33, p. 121]. Therefore, let us consider an arbitrary regular curve c on the
surface Σ defined by the position vector

c(s) = p(ξ(s), θ(s)) (6.4)

where ξ(s) and θ(s) are analytic functions and s is a generic parameter. In the
following we will conveniently assume s to be the arc length of c, so that the unit
tangent vector to c is given by
dc
t(s) = = p,ξ ξ,s +p,θ θ,s = p,s (6.5)
ds
and the derivative along c of the unit normal vector is
dmu
= m,uξ ξ,s +m,uθ θ,s = m,us (6.6)
ds
If s is not the arc length, the vector in (6.5) is still tangent to the curve but has
not unit magnitude.
The normal curvature kn and the geodesic torsion τg of Σ along t at a point P
can be easily computed by means of the following equations:

kn = −m,us ·t (6.7)
 
τg = mu m,us t (6.8)

and they are features of the surface. It is well known, [33, p. 128], that the normal
curvature assumes its extreme values k1 and k2 along two principal (orthogonal)
directions of unit tangent vectors t1 and t2 , having also null geodesic torsion. The
normal curvature and the geodesic torsion along any other direction at P are not
independent and, according to Euler’s and Bertrand’s formulas, [33, p. 132] and
[33, p. 159], are given by the following equations

kn = k1 cos2 β + k2 sin2 β (6.9)


τg = (k2 − k1 ) sin β cos β (6.10)

the unit tangent vector being

t = cos β t1 + sin β t2 (6.11)

Therefore to determine kn for any direction at P , the principal curvatures k1 and


k2 of the surface and the principal directions are required. If the normal curvatures
and geodesic torsions are given along two orthogonal but non-principal directions
of unit tangent vectors ta and tb , the generalized Euler’s and Bertrand’s formulas
can be employed

kn = ka cos2 β + kb sin2 β + τa sin2β (6.12)


τg = τa cos2β + (kb − ka ) sin β cos β (6.13)
94 6. Curvature analysis: different methods within the same framework

where in this case


t = cos β ta + sin β tb (6.14)
In [9], a more general form for non orthogonal direction is given in Eq.(45).
Let us observe that all unit tangent vectors lie in the tangent plane to the surface
in P , therefore
mu = t1 × t2 = ta × tb

6.2.2 Tensor of curvature


The curvature at a point of the surface Σ can be simply described using the tensor
of curvature K, [13] and [80]. We will denote with [K] the matrix form of the
tensor, i.e. its elements, that expressed in principal directions is
 
k 0
[K] = 1 (6.15)
0 k2

while in non principal (but orthogonal) directions is


   
k −τb k τ
[K] = a = a a (6.16)
τa kb τa kb

being τb = −τa . Therefore in an orthogonal reference frame [K] is symmetric while


in a non orthogonal one [K] is more complex, not symmetric and its elements do
not correspond to normal curvature and torsion, as detailed in Appendix 6.10.
Once K is expressed in an orthogonal reference frame S(P, i, j), either principal
or not, its components in another orthogonal system S 0 (P, i0 , j 0 ) can be obtained
by using the rotation matrix T between them
   
i · i0 i · j 0 cos ϕ − sin ϕ
[T ] = = (6.17)
j · i0 j · j 0 sin ϕ cos ϕ

where ϕ is the angle between i and i0 , and applying the tensor law of transforma-
tion
[K]S 0 = [T ]T [K]S [T ] (6.18)
In the following we will assume to express K in an orthogonal reference frame,
so that [K] will be a symmetric matrix.

6.2.3 Rodrigues’ formula


We recall that the inner product between a tensor and a vector is the vector whose
components, in an orthogonal reference frame S(P, i, j), are given by the product
of K and t components

[K · t] = [K] [t] (6.19)


6.2 Theoretical background: curvatures of a surface 95

For example, when i = t1 (and j = t2 ) employing Eqs. (6.11) and (6.15) we obtain

K · t = k1 cos β t1 + k2 sin β t2 (6.20)

and when i = ta (and j = tb ), for Eqs. (6.14) and (6.16), it becomes

K · t = (ka cos β + τa sin β)ta + (τa cos β + kb sin β)tb (6.21)

If we choose i = t (and j = mu × t) from Eq. (6.19) a general useful expression


can be obtained
  
k τg 1
[K · t] = n
τg kn⊥ 0
K · t = kn t + τg (mu × t) (6.22)

kn⊥ being the normal curvature along j; this equation holds even for a non unit
vector v parallel to t
K · v = kn v + τg (mu × v) (6.23)

Due to Eq. (6.22), Euler’s and Bertrand’s equations become

kn = (K · t) · t (6.24)
τg = (K · t) · (mu × t) (6.25)

which correspond to Eqs. (6.9) and (6.10) when K is written in the principal
directions as in (6.15), or alternatively to the generalized version (6.12) and (6.13)
if we employ (6.16). It may be interesting to note that Eqs. (6.24) and (6.25) hold
in any reference system, orthogonal or not. The previous relationships become
very useful when applied in Rodrigues’ formula, which relates the derivative of the
normal unit vector to the curvature. The most common version of this formula
employs scalar components in a principal reference frame, as in (6.20)
 
k1 0
[m,us ] = − [t] (6.26)
0 k2

but a more general expression can be used as well, exploiting Eq.(6.22)

m,us = −K · t = −kn t − τg (mu × t) (6.27)

The previous equation holds even if s is not the arc length, with p,s instead of t

m,us = −K · p,s = −kn p,s −τg (mu × p,s ) (6.28)


96 6. Curvature analysis: different methods within the same framework

6.3 Rotation operator


In this study, three different techniques in curvature analysis are compared after
being re-elaborated by means of a new approach presented in [11], which is based
on the use of the rotation operator R to describe the kinematics of the gear pair.
In this chapter only the basic concepts of this new approach are given, referring
to Chapter 2 and to the paper [11] for a deeper insight.
In order to express the rotation of a point P around an axis a of an angle α,
we first take a point O on the axis to define the position vector p = P − O and
then apply Euler’s relation corresponding to the compact notation R(p, a, α)

p̂(α) = P̂ − O = R p, a, α
= (p · a)a + [p − (p · a)a] cos(α) + a × p sin(α) (6.29)

where P̂ is the image of P after the rotation and a is the unit vector that marks
the axis direction.
Since Eq.(6.29) is defined on vectors, this approach has the main advantage
of involving vector as such and not their components, required for example by
homogeneous transformations, frequently employed in gear generation.
Some relevant properties of rotating vectors are described in Appendix 6.10.

6.4 Surfaces of the gear pair


The aim of the curvature analysis is the definition of a relationship between the
tensors of curvature of a gear pair, that can be a pinion and a gear or a generating
tool and the generated gear. In this section the definitions of Sec.6.2 are applied
to the surfaces of the pair.

6.4.1 Surface of the first gear


The first gear, pinion or generating tool, is assumed to be a (fixed) regular surface
Σe in the Euclidean space E3e . Given a fixed point Oe conveniently taken on the
axis of the pinion, it is possible to associate a position vector pe of R3 to each
point Pe (ξ, θ) of the surface

pe (ξ, θ) = Pe (ξ, θ) − Oe (6.30)

As in Eq.(6.2), the normal vector me to Σe is given by

me (ξ, θ) = pe,ξ × pe,θ (6.31)

and me 6= 0 for the regularity of the surface; the unit normal vector mue is
me (ξ, θ) me (ξ, θ) me (ξ, θ)
mue (ξ, θ) = =√ = (6.32)
|me (ξ, θ)| EG − F 2 D(ξ, θ)
6.4 Surfaces of the gear pair 97

where E, F, G are the coefficients of the first fundamental form of Σe [33, p. 82],
that is
E = pe,ξ · pe,ξ F = pe,ξ · pe,θ G = pe,θ · pe,θ (6.33)

and D = EG − F 2 .
It is useful for the development of the analysis to introduce a curve ce on the
surface Σe , that according to Eq.(6.4), is defined as

ce (se ) = pe (ξe (se ), θe (se )) (6.34)

where se is the arc length but a generic parameter could do as well. The unit
tangent and normal vectors to the surface along the curve are respectively

te (se ) = ce,se = pe,ξ ξe,se + pe,θ θe,se (6.35)


muce (se ) = mue (ξe (se ), θe (se )) (6.36)

We can assume that the tensor of curvature K e at each point of Σe is known, in


particular its principal curvatures k1e and k2e and the corresponding directions of
unit vectors e1 and e2 .

6.4.2 Surface of the second gear


In a similar way, we introduce another surface Γg in the Euclidean space E3g . Two
different cases are considered for defining this surface: in the first one (meshing)
Litvin assumes it to be a generic surface, while in the other case (generating), [11]
and [9], it is the envelope surface.

6.4.2.1 Meshing case


In Litvin’s approach, a second gear is considered whose curvature features are
investigated, in order to make it mesh with the pinion Σe under known relative
motion. Although we do not actually have its equations, we assume that the gear
surface is expressed using the parametric coordinates (χ, ν) and that its position
vector pl is
pl (χ, ν) = Pl (χ, ν) − Og (6.37)
The normal and unit normal vectors to the surface are ml (χ, ν) and mul (χ, ν),
evaluated according to Eqs. (6.2) and (6.3). Also in this gear we introduce a curve
cl defined as in Eq. (6.4)

cl (sl ) = pl (χ(sl ), ν(sl )) (6.38)

The normal unit vector along the curve is

mucl (sl ) = mul (χ(sl ), ν(sl )) (6.39)


98 6. Curvature analysis: different methods within the same framework

6.4.2.2 Generating case


In the second case, the generated surface is related to pe (ξ, θ) by the enveloping
process, involving also the relative motion between the pair. As better described
in Sec. 6.5.2, it is possible to define the position vector of Γg as
(
pg = pg (ξ, θ, φ)
f (ξ, θ, φ) = 0
where f (ξ, θ, φ) = 0 is the equation of meshing, [51]. We will denote with
mg (ξ, θ, φ) the normal vector to Γg and mug (ξ, θ, φ) will be the unit normal vector
in the point where the equation of meshing is satisfied.
A curve cg on Γg is defined, as done for ce and cl , taking into account the special
definition of points of the surface Γg
(
cg (sg ) = pg (ξg (sg ), θg (sg ), φ(sg ))
(6.40)
f (ξg (sg ), θg (sg ), φ(sg )) = 0
The unit tangent and normal vector along the curve are
tg (sg ) = cg,sg = pg,ξ ξg,sg + pg,θ θg,sg + pg,φ φg,sg (6.41)
mucg (sg ) = mug (ξg (sg ), θg (sg ), φ(sg )) (6.42)
again where the equation of meshing holds.
Actually in both the meshing and generating case, the explicit relation between
position vector and parametric coordinates is considered to be too complex and not
convenient for investigating the features of the surface, as described in Sec. 6.2, [83].
Therefore in all the approaches compared in this study, the aim is to determine
the tensor of curvature K g of surface Γg , in terms of tensor of curvature of the
pinion K e for a given (known) motion of the gears.

6.5 General definition of the meshing/generating


process
Let us consider a third fixed Euclidean space E3f where the enveloping or meshing
process takes place. The first gear (pinion) rotates of an angle ψ about a translating
axis a, identified by means of one of its (moving) points Oa and of a fixed unit
vector a. In a similar way, the second generated gear rotates of an angle ϕ about
the translating axis b defined by Ob and b. Both the rotation angles ψ and ϕ and
the points Oa and Ob depend on the parameter of motion φ.
The surface Σ̂e (φ) of the pinion, isomorphic to Σe , rigidly rotates around the
first axis a while moving with the transfer motion of Oa . Denoting by P̂ (ξ, θ, φ)
its generic point, the correspondent position vector p̂a with respect to Oa is

p̂a (ξ, θ, φ) = P̂ (ξ, θ, φ) − Oa (φ) = R pe (ξ, θ), a, ψ(φ) (6.43)
6.5 General definition of the meshing/generating process 99

S (f)

Oa(f) y(f)

Gg (f) pa
a

mu
d (f)

pb

j (f)
Ob(f)

Figure 6.1: Mating surfaces in the fixed space E3f .

employing notation in Eq. (6.29). Similarly, with respect to the moving point Ob ,
the position vector p̂b (ξ, θ, φ) of P̂ (ξ, θ, φ) is

p̂b (ξ, θ, φ) = P̂ (ξ, θ, φ) − Ob (φ) = p̂a (ξ, θ, φ) − d(φ) (6.44)

where (Fig. 6.1)


d(φ) = Ob (φ) − Oa (φ) (6.45)
3
Both position vectors p̂a (ξ, θ, φ) and p̂b (ξ, θ, φ) of R describe the same family of
surfaces Φf of E3f .
The unit normal vector m̂ue to each regular surface Σ̂e (φ) of the family (hence
for fixed φ) is by definition given by

m̂ue (ξ, θ, φ) = R(mue (ξ, θ), a, ψ(φ)) 6= 0 (6.46)

where Eqs. (6.141), (6.143) and (6.31) were employed.


At this point we have to distinguish between a meshing process considered by
Litvin and a generating process as done in [11] and [9], but in both cases the aim is
the investigation of the relationship between the tensors of curvature of the gears,
once their motion is given.
100 6. Curvature analysis: different methods within the same framework

6.5.1 Meshing process


In Litvin’s approach a second gear is considered in the fixed space, that is a
surface Γ̂g (φ) isomorphic to Γg introduced in Sec.6.4.2.1. Denoting by P̂l (χ, ν, φ)
the generic point of Γ̂g (φ), its position vector p̂l (χ, ν, φ) with respect to the moving
point Ob is given by the following expression

p̂l (χ, ν, φ) = P̂l (χ, ν, φ) − Ob (φ) = R pl (χ, ν), b, ϕ(φ) (6.47)

The normal vector m̂ul to each regular surface Γ̂g (φ) (hence for fixed φ) is given
by
m̂ul (χ, ν, φ) = R(mul (χ, ν), b, ϕ(φ)) (6.48)
In curvature analysis we assume that Γ̂g (φ) is in meshing contact with the surface
Σ̂e (φ), therefore three relationships are supposed to be satisfied: same position
vector and same normal vector (for fixed φ) and the equation of meshing (in
kinematic terms v(1) · n(1) = v(2) · n(2) [51]), respectively

p̂b (ξ, θ, φ) = p̂l (χ, ν, φ) (6.49)


m̂ue (ξ, θ, φ) = m̂ul (χ, ν, φ) (6.50)
 
0
ψ a× p̂a (ξ, θ, φ) − d0 · m̂ue = ϕ0 b × p̂l (χ, ν, φ) · m̂ul (6.51)

Strictly speaking Eq.(6.50) should be m̂ue (ξ, θ, φ) = ±m̂ul (χ, ν, φ), but since at the
end the sign does not affect results, we will assume it to be positive. Taking into
account Eqs. (6.49) and (6.50), the equation of meshing (6.51) becomes
 
ψ 0 a × p̂a (ξ, θ, φ) − ϕ0 b × p̂l (χ, ν, φ) − d0 · m̂ue

= (ψ 0 a − ϕ0 b) × p̂a (ξ, θ, φ) + ϕ0 b × d(φ) − d0 · m̂ue

= w × p̂a (ξ, θ, φ) + ϕ0 b × d(φ) − d0 · m̂ue = 0 (6.52)

where w(φ) = ψ 0 a − ϕ0 b is directed as the screw axis of the relative motion


between the gears. Taking advantage of the properties of the rotation operator,
we can develop further the last expression

R w × p̂a (ξ, θ, φ) + ϕ0 b × d(φ) − d0 , a, −ψ(φ) · mue (ξ, θ)

= we (φ) × pe (ξ, θ) + q e (φ) · mue (ξ, θ) = 0
f˜(ξ, θ, φ) = he (ξ, θ, φ) · mu (ξ, θ) = 0
e (6.53)

already obtained in [11], where three vectors have been introduced



we (φ) = R w(φ), a, −ψ(φ) (6.54)

q e (φ) = R ϕ0 b × d(φ) − d0 , a, −ψ(φ) (6.55)
he (ξ, θ, φ) = we (φ) × pe (ξ, θ) + q e (φ) (6.56)
6.5 General definition of the meshing/generating process 101

It is interesting to note that he (ξ, θ, φ) corresponds to the rotated relative velocity


between the gears, denoted by v(12) in [51].
As in [84], we use a different notation when the equation of meshing is written
using the normal vector or the unit normal vector

f (ξ, θ, φ) = he (ξ, θ, φ) · me (ξ, θ) = 0


f˜(ξ, θ, φ) = he (ξ, θ, φ) · mu (ξ, θ) = 0
e

It can be interesting to underline that in the meshing case the gears can be in
point or in line contact, depending on the solution of Eq.(6.49). The important
point is that in curvature analysis we are not interested actually in solving the
three conditions stated above (as in a Tooth Contact Analysis case), but we simply
assume that they are satisfied (at least at one point). This means that, although
we do not know it explicitly, because of the complexity of the solution of the TCA,
there exists a relationship between (χ, ν) and (ξ, θ) for each value of the parameter
of motion φ.

6.5.1.1 Conjugate curves in the meshing process


A curve ce of Σe and a curve cl of Γg , defined respectively in (6.34) and (6.38), are
said to be conjugate if, for each φ, they are in contact at one point. This means
that during meshing their arc lengths vary with the parameter of motion so that

ĉe (se (φ)) − d(φ) = ĉl (sl (φ)) (6.57)

or simply
ĉe (φ) − d(φ) = ĉl (φ) (6.58)
where

ĉe (φ) = R(ce (φ), a, ψ(φ)) = R(pe (ξe (φ), θe (φ)), a, ψ(φ)) = Ĉe (φ) − Oa (φ)
(6.59)
ĉl (φ) = R(cl (φ), b, ϕ(φ)) = R(pl (χ(φ), ν(φ)), b, ϕ(φ)) = Ĉl (φ) − Ob (φ)
(6.60)

The unit normal vectors along conjugate curves, defined in Eqs.(6.36) and (6.39),
are related as follows

m̂uce (φ) = R(muce (φ), a, ψ(φ)) = R(mucl (φ), b, ϕ(φ)) = m̂ucl (φ) (6.61)

It may be worth noting that in case of point contact between the gears, there
is only one couple of conjugate curves, corresponding to the trajectories of the
contact point over the two surfaces Σ̂e and Γ̂g . In case of line contact there are
infinitely many couples of conjugate curves, but each curve ce corresponds to only
one curve cl .
102 6. Curvature analysis: different methods within the same framework

6.5.2 Generating process


When Γg is the generated gear, it is obtained from Σe by an enveloping process
and E3g is the moving space where the gear surface Γg will be obtained as a fixed
one, [11]. In this space we first introduce a family of surfaces Φg whose envelope
Γg we are interested in; each surface of Φg is isomorphic to Σe . If Pg denotes the
generic point of Φg , the corresponding position vector pg = Pg − Og is obtained
by rotating p̂b around b by an angle −ϕ(φ) (minus sign to take the gear back in
a fixed position)

pg (ξ, θ, φ) = R p̂b (ξ, θ, φ), b, −ϕ(φ)

= R p̂a (ξ, θ, φ) − d(φ), b, −ϕ(φ)

= R R(pe (ξ, θ), a, ψ(φ)) − d(φ), b, −ϕ(φ) (6.62)
According to Eqs. (6.31) and (6.46), the normal vector mg to each regular surface
Σg (φ) of the family Φg (hence for fixed φ) is given by

mg (ξ, θ, φ) = pg,ξ × pg,θ = R m̂e (ξ, θ, φ), b, −ϕ(φ)

= R R(me (ξ, θ), a, ψ(φ)), b, −ϕ(φ) (6.63)
and similarly

mug (ξ, θ, φ) = R R(mue (ξ, θ), a, ψ(φ)), b, −ϕ(φ) (6.64)
The generated gear is the envelope Γg of the family of surfaces Φg in the Euclidean
space E3g given by the points of the family satisfying the equation of meshing, which
in this case is also the triple product of the partial derivatives of the position
vectors pg (ξ, θ, φ)
 
pg,ξ pg,θ pg,φ = mg · pg,φ = f (ξ, θ, φ) = 0
corresponding to Eq. (6.53) with (ξ, θ, φ), as detailed in [11].
As already mentioned in Sec. 6.4.2.2, position vectors of points of the envelope
surface are given by the system
(
pg = pg (ξ, θ, φ)
(6.65)
f (ξ, θ, φ) = 0

In the generating case, Eqs. (6.49) and (6.50) correspond to


(
R(pg (ξ, θ, φ), b, ϕ(φ)) = p̂b (ξ, θ, φ) = R(pe (ξ, θ), a, ψ(φ)) − d(φ)
(6.66)
f (ξ, θ, φ) = 0

and  
R mue (ξ, θ), a, ψ(φ) = R mug (ξ, θ, φ), b, ϕ(φ (6.67)
which are automatically satisfied by the definitions of pg and mug .
6.6 Litvin’s approach to curvature: indirect meshing method 103

6.5.2.1 Conjugate curves in the generating process


We can repeat exactly the same steps of the meshing case, for obtaining the con-
dition for ce on Σe and cg on Γg to be conjugate curves, that is, for a given φ,

ĉe (se (φ)) − d(φ) = ĉg (sg (φ)) (6.68)


or
ĉe (φ) − d(φ) = ĉg (φ) (6.69)
where ĉe (φ) has already been introduced in (6.59) and

ĉg (φ) = R(cg (sg (φ)), b, ϕ(φ)) = R(pg (ξg (φ), θg (φ), φ), b, ϕ(φ)) = Ĉg (φ) − Ob (φ)
(6.70)
In a similar way, for the unit normal vectors along conjugate curves, defined
in Eqs.(6.36) and (6.39), we have

m̂uce (φ) = R(muce (φ), a, ψ(φ)) = R(mucg (φ), b, ϕ(φ)) = m̂ucg (φ) (6.71)

It is interesting to note that in this case conjugate curves share the same
relations between the parametric coordinates and the arc lengths, that is

ξe (se (φ)) = ξg (sg (φ)) = ξ(φ)


θe (se (φ)) = θg (sg (φ)) = θ(φ)
f (ξ(φ), θ(φ), φ) = 0

6.6 Litvin’s approach to curvature: indirect mesh-


ing method
The classical curvature analysis proposed by Litvin moves from the determination
of the velocity of the contact points to relate the curvature features of the gear Γg
meshing with a pinion Σe . In [51], Litvin first considers a generic contact point P̌
defined by r(1) = r(2) + d and n(1) = n(2) , which in the present approach become
ˇ θ̌, φ̌) satisfying also the
Eqs. (6.49) and (6.50) for a given set of parameters (χ̌, ν̌, ξ,
equation of meshing, that is
ˇ θ̌, φ̌) = p̂ (χ̌, ν̌, φ̌)
p̂b (ξ, l
u ˇ
m̂ (ξ, θ̌, φ̌) = m̂u (χ̌, ν̌, φ̌)
e l (6.72)
ˇ θ̌, φ̌) = 0
f (ξ,

Then Litvin investigates the motion of the contact point, requiring that there
is still contact between the gears after an infinitesimal time increment. In this
context time does not play any role, so we express an infinitesimal time increment
as an infinitesimal increment of the motion parameter φ. This small rotation
104 6. Curvature analysis: different methods within the same framework

causes the contact point to move to an unknown point Q̌, in the neighborhood of
the previous P̌ .
The trajectories of the contact point over the two rotating surfaces Σ̂e (φ) and
Γ̂g (φ) are the rotating conjugate curves ĉe and ĉl , for which Eqs. (6.58) and (6.61)
hold. Accordingly the first two equations in (6.72) can also be written as

ĉe (φ̌) − d(φ̌) = ĉl (φ̌) (6.73)


m̂uce (φ̌) = m̂ucl (φ̌) (6.74)

and for the new contact point Q̌ we have

ĉe (φ̌) + ĉe,φ dφ − d(φ̌) − d0 dφ = ĉl (φ̌) + ĉl,φ dφ

that, for (6.73), can be simplified as

ĉe,φ − d0 = ĉl,φ (6.75)

while the normal unit vector is

m̂uce (φ̌) + m̂uce,φ dφ = m̂ucl (φ̌) + m̂ucl,φ dφ

or, for (6.74), simply


m̂uce,φ = m̂ucl,φ (6.76)
(1) (2)
Equations (6.75) and (6.76) correspond to Litvin’s conditions ṙ = ṙ and
ṅ(1) = ṅ(2) .
By employing the properties of the rotation operator and taking into account
Eqs. (6.59), (6.60), (6.73) and (6.74) , the last two results can be further developed.
In particular Eq. (6.75) becomes

R(ce (φ), a, ψ(φ)),φ −d0 = R(cl (φ), b, ϕ(φ)),φ


 
ψ 0 a × ĉe + R ce,φ , a, ψ(φ) − d0 = ϕ0 b × ĉl + R cl,φ , b, ϕ(φ)
 
w × ĉe (φ) + ϕ0 b × d(φ) − d0 + R ce,φ , a, ψ(φ) = R cl,φ , b, ϕ(φ)
  
R w × ĉe (φ) + ϕ0 b × d(φ) − d0 , a, −ψ + ce,φ = R R cl,φ , b, ϕ(φ) , a, −ψ
he + ce,φ = cle,φ (6.77)

where he has been defined in (6.56). In a similar way, Eq. (6.76) becomes
 
R muce (φ), a, ψ(φ) ,φ = R mucl (φ), b, ϕ(φ) ,φ
 
ψ 0 a × m̂uce + R muce,φ , a, ψ(φ) = ϕ0 b × m̂ucl + R mucl,φ , b, ϕ(φ)
 
w × m̂uce + R muce,φ , a, ψ(φ) = R mucl,φ , b, ϕ(φ)
  
R w × m̂uce (ξ, θ, φ), a, −ψ(φ) + muce,φ = R R mucl,φ , b, ϕ(φ) , a, −ψ(φ)
we × mue + muce,φ = mucle,φ (6.78)
6.6 Litvin’s approach to curvature: indirect meshing method 105

It is worth observing that the derivatives ce,φ , cl,φ , mce,φ and mucl,φ , as well
as cle,φ and mucle,φ are meaningful only along the curves and should be calculated
as
ce,φ = pe,ξ ξ,φ +pe,θ θ,φ = pe,φ
muce,φ = mue,ξ ξ,φ + mue,θ θ,φ = mue,φ
(6.79)
cl,φ = pl,χ χ,φ +p,lν ν,φ = pl,φ
mucl,φ = mul,χ χ,φ +mul,ν ν,φ = mul,φ

Therefore to determine the motion of the contact point we should know the values
at P̌ of the four scalar functions ξ,φ , θ,φ , χ,φ and ν,φ . However, as explained in
Sec. 6.5.1, such functions are very difficult to obtain in an explicit form so they are
practically unknown. Instead of trying to solve them, it is equivalent to assume as
unknowns all the derivatives vectors pe,φ , pl,φ , mue,φ and mul,φ , which correspond
(1) (2) (1) (2)
respectively (and rotated) to the unknowns vr , vr , nr and nr in [51].
All these derivatives are related through Rodrigues’ formula (6.28)

mue,φ = −K e · pe,φ (6.80)


mul,φ = −K g · pl,φ (6.81)

Applying rigid rotations to the second equation, surface features as K g do not


change, therefore we have
 
R R(mul,φ , b, ϕ), a, −ψ = −R R(K g · pl,φ , b, ϕ), a, −ψ
 
R R(mul,φ , b, ϕ), a, −ψ = −K g · R R(pl,φ , b, ϕ), a, −ψ
mule,φ = −K g · ple,φ (6.82)

Substituting in Eq. (6.78) we find

we × mue − K e · pe,φ = −K g · ple,φ


we × mue − K e · (ple,φ − he ) = −K g · ple,φ
(K e − K g ) · ple,φ = K e · he + we × mue
K eg · ple,φ = K e · he + we × mue (6.83)

K eg being the relative tensor of curvature. The unknown vectors must also satisfy
the derivative of the equation of meshing

df
= he,φ · mue + he · mue,φ

= (w0e × pe + we × pe,φ + q 0e ) · mue + he · mue,φ = 0 (6.84)
106 6. Curvature analysis: different methods within the same framework

that inserting Eqs. (6.77) and (6.80) becomes

df
= (w0e × pe + we × pe,φ + q 0e ) · mue + he · (−K e · pe,φ )

= (w0e × pe + q 0e ) · mue + (−K e · he + mue × we ) · pe,φ
= (w0e × pe + q 0e ) · mue − (K e · he + we × mue ) · (ple,φ − he ) = 0

and finally

(K e · he + we × mue ) · ple,φ = (w0e × pe + q 0e ) · mue + (K e · he + we × mue ) · he


(6.85)

Let us observe that the hypothesis of expressing K e in an orthogonal reference


frame has been applied for commutating the dot product, i.e. (K e · he ) · pe,φ =
(K e · pe,φ ) · he .
In the following we will use a new vector k defined as

k = K e · he + we × mue (6.86)

normal to the contact line as shown in Sec.6.7, that will help us to have more
compact expressions. Employing k, the solving equations (6.83) and (6.85) in the
unknown ple,φ can be written as

k · ple,φ = (w0e × pe + q 0e ) · mue + k · he (6.87)


K eg · ple,φ = k (6.88)

which represent a system of three scalar equations in two scalar unknowns, the
components of ple,φ in the tangent plane at P̂ , corresponding to (8.4.42) in [51].
Curvature features are determined requiring infinitely many solutions for the case
of line contact between the gears. Litvin discusses this issue by writing the matrix
of the linear system obtained after introducing a reference frame and expressing
vectors and tensor in their scalar components.
It can be interesting to follow another way, employing a different form for
Eq. (6.87); from the dot product

k · ple,φ = (w0e × pe + q 0e ) · mue + k · he = c (6.89)

it follows also that the component of ple,φ along k is determined, since c is a


known scalar quantity; therefore the vector ple,φ has an unknown component y in
the direction normal to k in the tangent plane. The solving equations become
c
ple,φ = k + (mue × k) y (6.90)
k·k
K eg · ple,φ = k (6.91)
6.6 Litvin’s approach to curvature: indirect meshing method 107

Having infinitely many solutions of ple,φ means that the above equations must be
satisfied for every value of y; so from y = 0 we obtain
 c 
K eg · k = k (6.92)
k·k
and therefore
K eg · (mue × k) = 0 (6.93)
Compared to Eq. (6.23), Eq. (6.93) means that both the normal curvature and
the geodesic torsion are null (as expected) in the direction mue × k, parallel to
the contact line, which is a principal direction for the relative curvature. From
Eqs. (6.92) and (6.93) we have
k1eg = 0 along mue × k (6.94)
k·k
k2eg = along k (6.95)
c
therefore the relative tensor of curvature is completely determined and to have the
tensor of curvature K g we only need to remind
K g = K e − K eg (6.96)
In scalar components in the reference system S(P, e1 , e2 ), the tensor of curvature
is
 e   T   
k1 0 cos σ − sin σ 0 0 cos σ − sin σ
[K g ] = − (6.97)
0 k2e sin σ cos σ 0 k2eg sin σ cos σ
where σ is the angle between mue ×k and e1 , positive if a counterclockwise rotation
of σ around mue moves mue × k over e1 .
We just observe that in case of point contact K eg has both principal values
different from zero and there is only one value of y and one ple,φ ; inserting Eq.(6.90)
in (6.91) we have  c 
K eg · k + (mue × k)y = k
k·k
from which we have
 c 
k− K eg · k = yK eg · (mue × k)
k·k
that corresponds to
  c  
k− K eg · k × (K eg · (mue × k)) = 0
k·k
Since the previous equation is a cross product of vectors in the tangent plane, it
may be written as the following scalar equation
  c  
k− K eg · k × (K eg · (mue × k)) · mue = 0 (6.98)
k·k
that corresponds to the null determinant required by Litvin in the case of point
contact, i.e. Eq.(8.4.52) in [51].
108 6. Curvature analysis: different methods within the same framework

6.7 Chen’s approach to curvature: indirect gen-


erating method
All the previous procedure results in part simplified when the second gear is the
envelope gear, generated according to equations in Sec.(6.5.2). As already men-
tioned, Eq. (6.72) is automatically satisfied from the definition of pg and mug in
(6.62) and (6.64). In the initial contact point P̌ we have

pe = pe (ξ, ˇ θ̌)
ˇ θ̌, φ̌) = 0
f (ξ,
ˇ θ̌)
mue = mue (ξ,

As in Sec. 6.6, we follow the contact point after an infinitesimal increment of


the motion parameter φ and its trajectories over the two rotating surfaces Σ̂e (φ)
and Γ̂g (φ), which again correspond to the rotating conjugate curves ĉe and ĉg .
Requiring the contact in Q̌, leading to Eqs.(6.75) and (6.76), we simply obtain in
the generating case already known properties of the rotation operator

pe,ξ = R R(pg,ξ , b, ϕ(φ)), a, −ψ(φ)

pe,θ = R R(pg,θ , b, ϕ(φ)), a, −ψ(φ)

he = R R(pg,φ , b, ϕ(φ)), a, −ψ(φ)

mue,ξ = R R(mug,ξ , b, ϕ(φ)), a, −ψ(φ)

mue,θ = R R(mug,θ , b, ϕ(φ)), a, −ψ(φ)

that directly give relationships equivalent to (6.77) and (6.78)

ce,φ + he = cge,φ (6.99)


muce,φ + we × mue = mucge,φ (6.100)

where

cge,φ = R R(cg,φ , b, ϕ(φ)), a, −ψ(φ)

mucge,φ = R R(mug,φ , b, ϕ(φ)), a, −ψ(φ)

In this case we need to take care of the notation of the derivatives of pg with
respect to φ, in particular

∂pg (ξ, θ, φ)
pg,φ = (6.101)
∂φ
dpg (ξ(φ), θ(φ), φ)
cg,φ = (6.102)

6.7 Chen’s approach to curvature: indirect generating method 109

The expressions of the derivative vectors are

ce,φ = pe,ξ ξ,φ + pe,θ θ,φ = pe,φ


muce,φ = mue,ξ ξ,φ + mue,θ θ,φ = mue,φ
(6.103)
cg,φ = pg,ξ ξ,φ + pg,θ θ,φ + pg,φ
mucge,φ = mug,ξ ξ,φ + mug,θ θ,φ + mug,φ

where the unknowns are the values at P̌ of the derivatives ξ,φ and θ,φ , which, from
the first equation, Chen writes as
  1
ξ,φ = − pe,θ pe,φ mue (6.104)
me
u 1
 
θ,φ = pe,ξ pe,φ me (6.105)
me
with me = mue · me . In this way instead of having two scalar unknowns ξ,φ and
θ,φ , we have the unknown vector pe,φ that lies in the tangent plane.
Rodrigues’ formula applied to Γg and rotated, as in Eq. (6.82), becomes

mug,φ = −K g · cg,φ
 
R R(mug,φ , b, ϕ), a, −ψ = −R R(K g · cg,φ , b, ϕ), a, −ψ
mucge,φ = −K g · cge,φ (6.106)

that inserted in Eq. (6.100) with (6.80) and (6.99) gives

we × mue − K e · pe,φ = −K g · cge,φ


we × mue + K e · he = K eg · cge,φ

It is also

K eg · (pe,ξ ξ,φ +pe,θ θ,φ +he ) = we × mue + Ke · he = k (6.107)

in the unknowns ξ,φ and θ,φ .


The scalar unknowns must also satisfy the derivative of the equation of meshing

df˜
= f,ξ ξ,φ +f,θ θ,φ +f,φ (6.108)

The derivatives of the function f (ξ, θ, φ) are

f˜,ξ = mue ,ξ ·he + mue · he,ξ = mue ,ξ ·he + mue · (ce (φ) × pe,ξ (ξ, θ)) (6.109)
f˜,θ = mue ,θ ·he + mue · he,θ = mue ,θ ·he + mue · (ce (φ) × pe,θ (ξ, θ)) (6.110)
f˜,φ = mu · he,φ = mu · (c0 (φ) × p (ξ, θ) + q 0 (φ))
e e e e e (6.111)
110 6. Curvature analysis: different methods within the same framework

which, although fairly long, are easily computable functions requiring the knowl-
edge of the generating surface equations and the relative motion. Putting all
together in (6.108) we obtain a scalar equation in the unknown quantities ξ,φ and
θ,φ . Substituting in (6.108) we have
df 1
= (−f,ξ pe,θ + f,θ pe,ξ ) · (pe,φ × mue ) + f,φ = 0
dφ me
df  1
= me ce − (mue ,ξ ·he )pe,θ + (mue ,θ ·he )pe,ξ · (pe,φ × mue ) + f,φ = 0
dφ me
corresponding to Eq.(10) in [9]. Employing Rodrigues’ formula, some simplifica-
tions can be done leading to an expression already obtained in Litvin’s approach
(cf. Eq. (6.87))
 pe,φ × mue
me ce − ((−K e · pe,ξ ) · he )pe,θ + ((−K e · pe,θ ) · he )pe,ξ · + f,φ
me
 pe,φ × mue
= me ce − ((−K e · he ) · pe,ξ )pe,θ + ((−K e · he ) · pe,θ )pe,ξ · + f,φ
me
 pe,φ × mue
= me ce − me (K e · he ) × mue · + f,φ
me

= ce − (K e · he ) × mue · (pe,φ × mue ) + f,φ
= −k · pe,φ + f,φ = 0

= −k · pe,φ + c0e (φ) × pe (ξ, θ) + q 0e (φ) · mue = 0 (6.112)
The solving equations are therefore (6.107) and (6.112)
K eg · (pe,ξ ξ,φ +pe,θ θ,φ +he ) = k

c0e (φ) × pe (ξ, θ) + q 0e (φ) · mue = k · (pe,ξ ξ,φ +pe,θ θ,φ +he )
which can be discussed as already done for the meshing case. At this point Chen,
assuming again infinitely many solutions for pe,φ and also assuming the line of
contact to be a principal direction, chooses to express Eqs. (6.107) and (6.112)
along lines at constant ξ or θ on the generating tool, and along the direction k.
For example, on a curve at constant θ we have
pe,φ = pe,ξ ξ,φ
and from Eq. (6.107) the relative curvature in the direction of pe,ξ is
k · (he + pe,ξ ξ,φ )
kξeg =
(he + pe,ξ ξ,φ )2
where, from Eq. (6.112)

c0e (φ) × pe (ξ, θ) + q 0e (φ) · mue
ξ,φ =
he + pe,ξ
6.8 New approach to curvature: direct method 111

In this approach, although not proven by Chen, it is possible to recognize that


k is orthogonal to the contact line deriving the equation of meshing with respect
to the arc length of a curve ce
df
= f,ξ ξ,se +f,θ θ,se +f,s φ,se = 0 (6.113)
dse
It is rather easy to repeat the steps in (6.112) and to obtain

df
= −k · pe,se + f,φ φ,se = 0 (6.114)
ds
If ce is a contact line, being φ,se = 0, gives

k · pe,se = k · te = 0 (6.115)

confirming that k is orthogonal to the contact line.


One of the main points in Chen’s paper is the description of the curvature
analysis in case of non orthogonal reference system; in this comparison this is not
taken into account, but can be developed employing the relations in Appendix
6.10.

6.8 New approach to curvature: direct method


In the direct method [12] and [11], (see also [21] for a preliminary explanation),
we want to determine K g moving directly from the definitions of curvature and
torsion in (6.7) and (6.8)

kng = −mug ,sg ·tg (6.116)


τgg = [mug mug ,sg tg ] (6.117)

Again we want to evaluate the normal curvature at a specific point P̌ , correspond-


ing to čg = pg (ξg (šg ), θg (šg ), φg (šg )), and along a given direction ťg = tg (šg ). The
point and the direction are selected by means of the numbers

ξˇ = ξg (šg ), θ̌ = θg (šg ), φ̌ = φg (šg ) (6.118)

and
gξ = ξg,sg (šg ), gθ = θg,sg (šg ), gφ = φg,sg (šg ) (6.119)
The parametric coordinates in (6.118) are not completely independent but must
satisfy the equation of meshing f˜(ξ, ˇ θ̌, φ̌) = 0. In addition, the components
(gξ , gθ , gφ ) in (6.41) must be such tg (šg ) · tg (šg ) = 1 and the total derivative
of the equation of meshing vanishes, that is

f˜,ξ gξ + f˜,θ gθ + f˜,φ gφ = 0 (6.120)


112 6. Curvature analysis: different methods within the same framework

We could also choose the direction for evaluating kng and τgg on the tool surface Σe
by means of the unit vector

ˇ θ̌, φ̌), b, ϕ(φ̌)), a, −ψ(φ̌)
te = R R(tg (ξ, (6.121)

which, when the tool and the gear get in contact at (ξ, ˇ θ̌, φ̌) marks the same
direction of tg . It is important to stress that in this case we do use conjugate
curves as in the other approaches, since they do not satisfy Eq.(6.121). The
tangent vector te lies in the tangent plane therefore can also be written as linear
ˇ θ̌) and p (ξ,
combination of pe,ξ (ξ, ˇ θ̌)
e,θ

te = pe,ξ eξ + pe,θ eθ (6.122)

with suitable scalars eξ and eθ , satisfying also (see (6.33))

te · te = E e2ξ + 2F eξ eθ + G e2θ = 1 (6.123)

Although eξ and eθ have been introduced as numbers, they could also be seen as
ˇ θ̌) of the derivatives of functions like in (6.35).
the values at (ξ,
Inserting Eqs. (6.41) and (6.122) in (6.121), employing (6.99) and (6.103), we
get
pe,ξ eξ + pe,θ eθ = pe,ξ gξ + pe,θ gθ + he gφ (6.124)
The dot products of Eq. (6.124) with pe,ξ and pe,θ together with the derivative
of the equation of meshing with respect to sg give the following system of three
linear equations

 E gξ + F gθ + he · pe,ξ gφ = E eξ + F eθ


F gξ + G gθ + he · pe,θ gφ = F eξ + G eθ (6.125)


 f˜, g + f˜, g + f˜, g =0
ξ ξ θ θ φ φ

which is solvable at non singular points on Γg [11, eqn. (60)] since



E F he · pe ,ξ

1
g̃(ξ, θ, φ) = 2 F G he · pe ,θ 6= 0 (6.126)
D
f˜,ξ f˜,θ f˜,φ

For instance, gφ is given by



E F E eξ + F eθ E F 0
1 1
gφ = 2 F G F eξ + G eθ = − 2 F
G 0

D g̃ D g̃
f˜,ξ f˜,θ 0 f˜,ξ ˜
f ,θ ˜ ˜
f ,ξ eξ + f ,θ eθ
1 f˜,φ
= − (f˜,ξ eξ + f˜,θ eθ ) = φ,se (6.127)
g̃ g̃
6.8 New approach to curvature: direct method 113

The normal curvature of Γg defined in (6.116), owing to equations (6.99),


(6.103) and (6.100), along with property (6.140) can be written as
 
kng = − mue ,ξ gξ + mue ,θ gθ + we × mue gφ · pe,ξ gξ + pe,θ gθ + he gφ (6.128)
and for (6.121)

kng = − mue ,ξ gξ + mue ,θ gθ + we × mue gφ · (pe,ξ eξ + pe,θ eθ ) (6.129)

This is a computable expression where everything is evaluated at (ξ, ˇ θ̌, φ̌) and
expressed in terms of the tool shape and the generating relative motion.
A simpler equation can be obtained for the relative normal curvature kneg be-
tween surfaces Σe and Γg at contact points [48, p. 286] along the same direction;
by definition, it is
kneg = kne − kng = −(mue ,se ·te − mug ,sg ·tg ) (6.130)
provided equation (6.121) holds. Employing equations (6.86), (6.99), (6.103),
(6.122), (6.124) and (6.129) we obtain

kneg = − mue ,ξ eξ + mue ,θ eθ − (mue ,ξ gξ + mue ,θ gθ ) − we × mue gφ · te

= − mue ,ξ (eξ − gξ ) + mue ,θ (eθ − gθ ) − we × mue gφ · te
= −(mue ,ξ eξ + mue ,θ eθ ) · (pe,ξ (eξ − gξ ) + pe,θ (eθ − gθ )) + te · we × mue gφ
= −((mue ,ξ eξ + mue ,θ eθ ) · he + mue · we × te )gφ
= −(mue ,se ·he + mue · we × te )gφ
= −((−K e · te ) · he + mue · we × te )gφ
= −((−K e · he ) · te + mue × we · te )gφ
= (k · te )gφ (6.131)
that, taking into account (6.114) and (6.127), becomes
1
kneg = k · te )2 (6.132)

This result confirms that k is a principal direction and that the relative normal
curvature along the contact line, orthogonal to k, is null. Compared to (6.94) and
(6.95), Eq. (6.132) underlines that c = (k · k)g̃.
Without reporting all the steps that can be found in [12], the geodesic torsion
of Γg along cg defined in (6.117) can be written as
  
τgg = mue mue ,ξ gξ + mue ,θ gθ + we × mue gφ te (6.133)
while the relative geodesic torsion τgeg at contact points of the surfaces Σe and Γg
is given by the following equation
  1 
τgeg = − k te mue gφ = [k mue te k · te ) (6.134)

114 6. Curvature analysis: different methods within the same framework

6.9 Concluding remarks


Three different approaches in curvature analysis have been compared with a new
method, which is based on geometric instead of kinematic relationships.
The indirect approach obtains curvatures starting from the analysis of the
motion of contact points from the following conditions:

– contact in a point P̃ :

1. corresponding position vectors of the gear surfaces in P̃ in the fixed


space
2. equal normal unit vectors to the gear surfaces in P̃ in the fixed space
3. equation of meshing satisfied in P̃

– contact in a point Q̃, after dφ

1. equal derivatives of position vectors of the gear surfaces with respect to


φ in P̃ in the fixed space
2. equal derivatives of the normal vectors to the gear surfaces in P̃ with
respect to φ in the fixed space
3. derivative of equation of meshing with respect to φ satisfied in P̃

– Rodrigues’ formula

In Litvin’s approach the two gear surfaces are generic and previous conditions are
used to relate the curvature of one gear to the curvature of the other one knowing
relative motion. Since the contact point P̃ is supposed to exists, the first three
conditions are not really employed; the fundamental relations are obtained from
the last four requirements.
Chen considers a generating case where one gear is obtained from the other
by an enveloping process; in this case the first three conditions are automatically
guaranteed, and also conditions on the derivatives of the position and normal
vectors are satisfied thanks to properties of the rotation operator. We could simply
apply the derivative of the equation of meshing and Rodrigues’ formula.
It does not seem to be a relevant advantage to restrict the analysis to the
generating case instead of the general meshing study considered by Litvin. In
both these indirect approaches, although not done in the original version, we have
introduced conjugate curves as trajectories of the contact points over the two
surfaces which should make some steps clearer.
In the direct approach we do not involve conjugate curves but we use rotation
operator properties to relate the generating tool characteristic vectors, as tangent
and normal vectors, to those of the generated gear. In particular we consider curves
sharing the same tangent vector in the contact point during meshing. Employing
a relationship between the derivatives of the unit normal vectors along such curves
6.9 Concluding remarks 115

and the derivative of the equation of meshing, we obtain the very same results of
the previous approaches but in a direct way.
The present comparison employs the tensor of curvature K, not explicitly
considered in other papers, which simplifies many steps of the analysis leading to
very compact and simple expressions without requiring reference systems.
116 6. Curvature analysis: different methods within the same framework

6.10 Appendix
Tensor of curvature in a non orthogonal reference frame
The tensor of curvature can be expressed in a non orthogonal reference frame
S(P, i, j) but its elements do not correspond to the curvature and torsions along
the axes as in (6.16). In order to obtain a more general form for [K] it is convenient,
first of all, to extend the rotation in Eq.(6.17) to a more general transformation
case, from S(P, i, j) to S 0 (P, i0 , j 0 ) both with generic directions
   
i · i0 i · j 0 cos α cos β
[T ] = 0 0 = (6.135)
j·i j·j sin α sin β

where α and β are the angles between i0 and j 0 and i respectively; instead of
(6.18) in this case we have

[K]S 0 = [T ]−1 [K]S [T ] (6.136)

To give an idea of a possible form of [K] in a non orthogonal reference frame, the
previous equations are applied to transform [K] expressed in principal directions
S(P, i = t1 , j = t2 ) as in (6.15), to a generic S 0 (P, i0 , j 0 )
 
−1 k1 0
[K]S = [T ]
0 [T ]
0 k2
 
k1 cos α sin β − k2 cos β sin α k1 − k2
sin(2β)
 sin(β − α) 2 sin(β − α) 
= 
 k1 − k2 k2 cos α sin β − k1 cos β sin α 
sin(2α)
2 sin(β − α) sin(β − α)

As a general rule, writing [K] in a reference system S(P, i, j), that can be orthog-
onal or not, as
 
k k12
[K] = 11
k21 k22

the normal curvature and geodesic torsion along the axes directions i and j can
be obtained from the elements of the matrix as

kn(i) = (K · i) · i = k11 + k21 i · j


τg(i) = (K · i) · (mue × i) = k21 (mue × i) · j
kn(j) = (K · j) · j = k22 + k12 i · j
τg(j) = (K · j) · (mue × j) = k12 (mue × j) · i
6.10 Appendix 117

Properties of rotating vectors


The rotation operator has many important and useful properties, that are em-
ployed in this approach.
Given the rotated vectors

û = R(u, a, α) v̂ = R(v, a, α) ŵ = R(w, a, α) (6.137)

we immediately have the following algebraic properties

u = R(R(u, a, α), a, −α) (6.138)


û + v̂ = R(u + v, a, α) (6.139)
û · v̂ = u · v (6.140)
v̂ × ŵ = R(v × w, a, α) (6.141)
 
û · (v̂ × ŵ) = u · (v × w) = u v w (6.142)

As far as differential properties are concerned, we have



p̂,ξ (ξ, θ, φ) = R p,ξ (ξ, θ), a, α(φ)

p̂,θ (ξ, θ, φ) = R p,θ (ξ, θ), a, α(φ) (6.143)

whereas the derivative with respect to φ that controls the rigid rotation is

p̂,φ (ξ, θ, φ) = R p(ξ, θ), a, α(φ) ,φ

= α0 a × R p(ξ, θ), a, α(φ) = α0 a × p̂(ξ, θ, φ), (6.144)

where α0 (φ) = dα(φ)/dφ. In a more general case, like in

p̂(ξ, θ, φ) = R(p(ξ, θ, φ), a, α(φ)), (6.145)

a combination of the previous results is required



p̂,φ (ξ, θ, φ) = R p,φ (ξ, θ, φ), a, α(φ) + α0 a × p̂(ξ, θ, φ) (6.146)
Chapter 7

Third order analysis of gear


surfaces

A rough estimate of contact stresses and contact deformations between gear sur-
faces in direct contact is usually performed by employing the results of the Hertz
theory. Hertz’s solution for the three-dimensional contact between two elastic
bodies is based on the assumptions that the relative normal curvatures of the sur-
faces is uniform in the contact region and the extension of contact region itself is
small as compared to the radii of relative curvature. Under these hypotheses the
gap surface, representing the relative distance between the undeformed bodies in
geometrical contact, can be represented by a quadric surface.
However, in several practical cases the condition of constant relative curvatures
in the whole contact region is questionable. For instance, in a heavy duty spiral
bevel gear transmission for aerospace applications, the contact region can become
as large as 50–80% the extension of the tooth lead. In these conditions, a non-
negligible variation of relative curvature can be expected within the contact region.
In the present chapter a detailed investigation on the differential properties of
the generated gear and of the gap surface between the tool and the generated gear
itself is performed. The procedure is based on the evaluation of terms up to the
third order which appear in the Taylor’s series approximation of the surface in a
neighborhood of a selected contact point.
The information obtained can be employed to evaluate the distribution of the
contact pressures more accurately as reported in [7].
As suggested also in [84], the proposed analysis can be employed to examine
more closely the relationship between the conjugate surfaces in the neighborhood
of a flat contact point, which occur, e.g. [11], when an envelope to the contact lines
on the tool surface exists, that is when the derivative of the equation of meshing
with respect to the parameter of motion vanishes.
The sought for expression of curvature and its derivatives are obtained in a
120 7. Third order analysis of gear surfaces

direct way, simply employing their definitions and some fundamental properties
due to the enveloping process.
A former investigation on this topic can be found in [84], although this approach
is very different in many aspect with the respect to the proposed one. Here time
is not involved in the treatment, nor we need to introduce the concept of relative
differentiation.

7.1 Third order differential properties of the gen-


erated gear
In this section a purely geometric approach is developed for the evaluation of
local differential properties of Γg up to the third order of differentiation, in the
neighborhood of any of its regular points, in terms of the features of the tool Σe
and the enveloping rigid-body motion, that is pe (ξ, θ), a, b, ψ(φ), ϕ(φ) and d(φ).
Let us take a curve cg on the gear surface Γg , with arc length sg . Exactly like
in (3.64) at page 45, we assume that three functions ξg (sg ), θg (sg ) and φg (sg ) exist
such that
f˜(ξg (sg ), θg (sg ), φg (sg )) = 0 (7.1)
Therefore, the position vectors cg (sg ) of the curve cg are given by

cg (sg ) = pg (ξg (sg ), θg (sg ), φg (sg )) (7.2)

the unit tangent vector to cg is


dcg
tg (sg ) = = pg,ξ ξg,sg + pg,θ θg,sg + pg,φ φg,sg (7.3)
dsg
and the derivative along cg of the unit normal vector to Γg is
dmug
= mug ,ξ ξg,sg + mug ,θ θg,sg + mug ,φ φg,sg = mug ,sg (7.4)
dsg
Typically we want to evaluate the differential properties of Γg at some selected
points, say P̌g = čg + Ǒg , where

čg = pg (ξg (šg ), θg (šg ), φg (šg )), (7.5)

and Ǒg is the chosen origin for R3 . The evaluation is performed along a given
direction ťg = tg (šg ). Therefore, we select the point and the direction, respectively,
by means of the numbers

ξˇ = ξg (šg ), θ̌ = θg (šg ), φ̌ = φg (šg ) (7.6)

and
gξ = ξg,sg (šg ), gθ = θg,sg (šg ), gφ = φg,sg (šg ) (7.7)
7.1 Third order differential properties of the generated gear 121

Of course they are not completely independent as (ξ, ˇ θ̌, φ̌) must satisfy the equation
of meshing f˜(ξ,
ˇ θ̌, φ̌) = 0, while the components (gξ , gθ , gφ ) in (7.3) must be such
that tg be of unit length (tg (šg ) · tg (šg ) = 1) and the total derivative (3.65) (see
page 45) of the equation of meshing vanishes, that is
f˜,ξ gξ + f˜,θ gθ + f˜,φ gφ = 0 (7.8)
By expanding in Taylor series the curve cg with respect to the arc length sg , in
the neighborhood of Pˇg , we obtain
3
X ∆sνg d ν cg (šg )
cg (sg ) = čg + ν
+ o(∆s3g ) (7.9)
ν=1
ν! d s g

where ∆sg = sg − šg . Neglecting terms of order higher than three, the increment
∆cg (∆sg ) = cg (sg ) − čg can be approximated by the following expression


∆cg (∆sg ) = ε(∆sg ) tg + δ(∆sg ) mug + γ(∆sg ) (mug × tg ) (7.10)
šg šg šg

where the components along the directions tg , mug


and mug
× tg , are respectively
computed in terms of more familiar quantities, as follows
1  g 2 
ε(∆sg ) = ∆sg − (kn ) + (kgg )2 ∆s3g
6
1 g 1  g 
δ(∆sg ) = kn ∆s2g + kn ,sg + kgg τgg ∆s3g
2 6
1 g 2 1  g 
γ(∆sg ) = kg ∆sg + kg ,sg − kng τgg ∆s3g (7.11)
2 6
The calculations to obtain the expressions in (7.11) are reported in the Appendix.
In Eq. (7.11), the scalar quantities kng and τgg are the normal curvature and the
geodesic torsion of Γg , evaluated at P̌g along tg (see, e.g., [33], p.122 and p.159).
The geodesic curvature kgg is instead a property of both the curve cg and the
surface Γg where it is immersed ([33], p.155). In order to actually evaluate the
increment ∆cg (∆sg ) up to the third order, it is evident from (7.10) and (7.11)
that both derivatives kng ,sg and kgg ,sg ought to be evaluated at P̌g .
A major simplification comes from the hypothesis that the curve cg considered
is a normal section at P̌g . This curve is obtained as the intersection between
surface Γg and the osculating plane ΠP̌g , that is cg : Γg ∩ ΠP̌g . By construction
this is locally a geodesic on Γg , thus kgg = 0. Moreover it belongs to ΠP̌g , therefore
γ(∆sg ) = 0 and the components (7.11) simplify in
1 g 2
ε(∆sg ) = ∆sg − (k ) ∆s3g (7.12)
6 n
1 1
δ(∆sg ) = kng ∆s2g + kng ,sg ∆s3g (7.13)
2 6
γ(∆sg ) = 0 (7.14)
122 7. Third order analysis of gear surfaces

The curve cg is not identically a geodesic but only locally therefore in general,
kgg ,sg 6= 0, nevertheless being a normal section through P̌g , at P̌g we must have

kgg ,sg −kng τgg = 0 ⇒ kgg ,sg = kng τgg (7.15)

which provides a way to compute the derivative kgg ,sg along a normal section.
Our goal is to assess, up to the third order, the shape of ∆cg which lies on
the osculating plane ΠP̌g through P̌g . We will therefore not take into account the
component of displacement γ(∆sg ) orthogonal to ΠP̌g , since it vanishes identically.
From (7.12) and (7.13) it is evident that the proposed analysis can be performed
just evaluating the normal curvature kng and its derivative kng ,sg along cg , which are
the only features of Γg appearing in the tangent and normal components ε(∆sg )
and δ(∆sg ).
In the following we will seek an expression for the component δ of displacement
of ∆cg along mug , as a function of the displacement ε along tg . This relation can
be obtained by performing a series reversion of function ε(∆sg ) in (7.12). This
leads to
1
∆sg (ε) = ε + (kng )2 ε3 (7.16)
6
Substituting the result in (7.13) we get
1 g 2 1 g
δ(ε) = k ε + kn ,sg ε3 , (7.17)
2 n 6
where, again, only terms up to the third order have been considered. The normal
section cg of Γg with ΠP̌g is therefore approximated as a third-order polynomial
expression, being the second and third order coefficients related to the normal
curvature kng and to its derivative kng ,sg with respect to the arc length sg of cg .
The goal is then to obtain the curvature kng and its derivative kng ,sg directly from
the properties of the enveloping tool Σe and the geometry of the relative motion.
Similar expressions, through a different route though, were obtained in [84] at page
110.

7.1.1 Normal curvature


The normal curvature kng of Γg along cg is defined, like in (5.10) at page 74, as

kng = −mug ,sg ·tg (7.18)

However, we wish to manipulate this formula in order to obtain a more convenient


expression as done for (5.27). Owing to equations (3.43), (3.50), (3.53) and (3.54)
(see page 42 and thereafter), along with property (3.5) we obtain that
 
kng = − mue ,ξ gξ + mue ,θ gθ + we × mue gφ · pe,ξ gξ + pe,θ gθ + he gφ (7.19)
7.1 Third order differential properties of the generated gear 123

This is a computable expression where everything is evaluated at (ξ, ˇ θ̌, φ̌) and it
is in terms of the tool shape and the generating relative motion.
However, in many cases it is perhaps simpler to choose the direction for the
evaluation of kng on the tool surface Σe . Therefore we need the unit vector

ˇ θ̌, φ̌), b, ϕ(φ̌)), a, −ψ(φ̌)
te = R R(tg (ξ, (7.20)

ˇ θ̌, φ̌), the same direction


to mark, when the tool and the gear come in contact at (ξ,
of tg .
By construction, te is tangent to the tool surface Σe at pe (ξ,ˇ θ̌). Accordingly
it can be written as
te = pe,ξ eξ + pe,θ eθ (7.21)
ˇ θ̌) and p (ξ,
that is as a linear combination of pe,ξ (ξ, ˇ θ̌) with suitable scalars eξ
e,θ
and eθ . We remark that in this framework they are numbers, although they could
also be seen as derivatives of functions like in (5.8) at page 73. For te to be of unit
length we require (see (5.5))

te · te = E e2ξ + 2F eξ eθ + G e2θ = 1 (7.22)

Of course it is simpler to choose the direction for the evaluation of kng by employing
ˇ θ̌), p (ξ,
eξ and eθ , since the natural basis (pe,ξ (ξ, ˇ θ̌)) has known directions on
e,θ
the generating tool Σe .
On the other hand, if the direction te of inspection is given and we are required
to calculate the correspondent coefficients eξ and eθ , it suffices to take the cross
product of (7.21) by pe ,θ and pe ,ξ to obtain

1
eξ = [te pe ,θ mue ]
D
1
eθ = [te mue pe ,ξ ] (7.23)
D
Combining Eqs. (7.3), (7.20) and (7.21), we obtain that

pe,ξ eξ + pe,θ eθ = pe,ξ gξ + pe,θ gθ + he gφ (7.24)

that is
pe,ξ (eξ − gξ ) + pe,θ (eθ − gθ ) = he gφ (7.25)
If we take the dot product of equation (7.24) with pe ,ξ and pe ,θ and we make
use of (3.65) (page 45), we get the following system of three linear equations
    
E F he · pe,ξ gξ E eξ + F eθ
F G he · pe,θ     
   gθ  = F eξ + G eθ  (7.26)
˜ ˜
f ,ξ f ,θ ˜
f ,φ g φ 0
124 7. Third order analysis of gear surfaces

which is solvable at non singular points on Γg [11, eqn. (60)] since



E F he · pe ,ξ

1
g̃(ξ, θ, φ) = 2 F G he · pe ,θ 6= 0 (7.27)
D
f˜,ξ f˜,θ f˜,φ
For instance, gφ is given by

E F E eξ + F eθ E F 0
1 1
gφ = 2 F G F eξ + G eθ = − 2 F
G 0

D g̃ D g̃
f˜,ξ f˜,θ 0 f˜,ξ ˜
f ,θ ˜ ˜
f ,ξ eξ + f ,θ eθ
1 f˜,φ
= − (f˜,ξ eξ + f˜,θ eθ ) = φ,se (7.28)
g̃ g̃
Therefore for the computation of kng we can employ the following formula once
we have obtained the three quantities gξ , gθ and gφ from eξ and eθ

kng = − mue ,ξ gξ + mue ,θ gθ + we × mue gφ · (pe,ξ eξ + pe,θ eθ ) (7.29)
In order to obtain a more compact expression for the normal curvature of Γg
the definition of the relative normal curvature between Σe and Γg can be used.
The relative normal curvature kneg between surfaces Σe and Γg at contact points
[48, p. 286] is, by definition, given by the following equation (cf. (5.10) and (7.18))
kneg = kne − kng = −(mue ,se ·te − mug ,sg ·tg ) (7.30)
provided equation (7.20) holds.
Employing equations (5.6), (5.9), (3.54), (3.70), (7.21), (7.25) and (7.29) we
obtain

kneg = − mue ,ξ eξ + mue ,θ eθ − (mue ,ξ gξ + mue ,θ gθ ) − we × mue gφ · te

= − mue ,ξ (eξ − gξ ) + mue ,θ (eθ − gθ ) − we × mue gφ · te
= −(mue ,ξ eξ + mue ,θ eθ ) · (pe,ξ (eξ − gξ ) + pe,θ (eθ − gθ )) + te · we × mue gφ
= −((mue ,ξ eξ + mue ,θ eθ ) · he + mue · we × te )gφ
= −(mue ,se ·he + mue · we × te )gφ
= −(k · te )gφ (7.31)
which in the last lines provides a fairly compact and computable expressions that
fully exploits the enveloping process.
The normal curvature of Γg can therefore be evaluated as
kng = (k gφ − mue ,se ) · te (7.32)
= L e2ξ + 2 M eξ eθ + N e2θ + gφ k · (pe ,ξ eξ + pe ,θ eθ ) (7.33)
where only the coefficient gφ has to be computed according to (7.28).
7.1 Third order differential properties of the generated gear 125

7.1.2 Derivative of the normal curvature


The derivative kng ,sg of the normal curvature kng of Γg at P̌g with respect to the
parameter of motion sg can be calculated directly from the definition of kng as
follows
dkng d (mug ,sg ·tg )
kng ,sg = =− = −mug ,sg sg ·tg + kgg τ g (7.34)
dsg dsg

Since in our analysis cg is locally a geodesic at P̌g , it follows that kgg = 0, and (7.34)
simplifies in

kng ,sg = −mug ,sgsg ·tg (7.35)

Owing to equations (3.43), (3.50), (3.53) and (3.54), along with property (3.5) we
obtain

kng ,sg = −(mug ,sgsg )e · te , (7.36)

where we defined

(mug ,sgsg )e = R(R(mug ,sgsg , b, ϕ(φ)), a, −ψ(φ))


= mue ,ξξ gξ2 + mue ,θθ gθ2 + (mug ,φφ )e gφ2
 
+ 2 mue ,ξθ gξ gθ + (mug ,θφ )e gθ gφ + (mug ,ξφ )e gξ gφ
+ mue ,ξ gξξ + mue ,θ gθθ + mue ,φ gφφ (7.37)

and the expressions of the known vectors (mug ,ij )e with i = ξ, θ, φ, j = φ are given
in (7.83) in the Appendix. It is important here to point out that in the generic
expression (mug ,ij )e the derivatives are intended to be performed first and then
result is applied a double rotation from E3g to E3e . According to (3.9) (page 36) it
is evident that for the derivatives involving only the surface parameters ξ and θ
the order of derivation and derivation can be interchanged, as applied in (7.37).
It is worth noting also that gφφ , appearing in Eq. (7.37), are the values of
the second derivatives of functions ξg (sg ), θg (sg ) and φg (sg ) with respect to sg
evaluated at šg , that is

gξξ = ξg ,sgsg (šg ) gθθ = θg ,sgsg (šg ) gφφ = φg ,sgsg (šg ) (7.38)

Till now the values of gξξ , gθθ and gφφ are not known and our goal is to find a
procedure to calculate these numbers.
First of all, we wish here exploit the fact that the second derivative vector
cg ,sgsg is known. In fact, since cg is a normal section at P̌g , kgg = 0, and the
following expression holds

cg ,sgsg = kng mug (7.39)


126 7. Third order analysis of gear surfaces

Moreover, it is also possible to express it as a function of the first and second


derivatives gξ , gθ , gφ and gξξ , gθθ and gφφ as follows

cg ,sgsg = pg ,ξξ gξ2 + pg ,θθ gθ2 + pg ,φφ gφ2


+ 2 pg ,ξθ gξ gθ + 2 pg ,θφ gθ gφ + 2 pg ,ξφ gξ gφ
+ pg ,ξ gξξ + pg ,θ gθθ + pg ,φ gφφ (7.40)
= z + pg ,ξ gξξ + pg ,θ gθθ + pg ,φ gφφ , (7.41)

where we defined the known vector z as the first two lines of (7.40). It is worth
stressing that gξ , gθ and gφ are already known quantities since we made use of the
solution provided by the sistem of equations (7.26) for their determination. By
comparing Eq. (7.39) and (7.41) we obtain

kng mug = z + pg ,ξ gξξ + pg ,θ gθθ + pg ,φ gφφ (7.42)

By applying a double rotation R(R(. . . , b, ϕ(φ̌)), a, −ψ(φ̌)) to both members of


Eq. (7.42) we get

kng mue = z e + pe ,ξ gξξ + pe ,θ gθθ + he gφφ (7.43)

In (7.43) we defined vector z e as

z e = pe ,ξξ gξ2 + pe ,θθ gθ2 + (pg ,φφ )e gφ2


+ 2 pe ,ξθ gξ gθ + 2 (pg ,θφ )e gθ gφ + 2 (pg ,ξφ )e gξ gφ (7.44)

Vectors (pg ,ij )e , with i = ξ, θ, φ, j = φ, are known at P̌g and their expressions
are given in (7.84) in the Appendix.
By taking the scalar product of (7.43) with (mue ×pe ,θ )/D and (mue ×pe ,ξ )/D
we get the two important equations

[mue he pe ,θ ] [mue pe ,θ z e ]
gξξ + gφφ =
D D
[mue pe ,ξ he ] [mue z e pe ,ξ ]
gθθ + gφφ = (7.45)
D D
Another condition that can be exploited is that also the second derivative of the
function of meshing f˜ with respect to the arc length sg must vanish, that is

d 2 f˜
=0 (7.46)
d s2g

The previous equation expands to

f˜,ξ gξξ + f˜,θ gθθ + f˜,φ gφφ = c, (7.47)


7.2 Third order analysis of the gap surface 127

where the scalar c is a known quantity given by the following expression



c = − f˜,ξξ gξ2 + f˜,θθ gθ2 + f˜,φφ gφ2

+ 2(f˜,ξθ gξ gθ + f˜,θφ gθ gφ + f˜,ξφ gξ gφ ) (7.48)

It is worth stressing that the second derivatives f˜ij with i, j = ξ, θ, φ of the


equation of meshing in the strict sense f˜ are known functions of the known vectors
mue ,ij (with i, j = ξ, θ), which depend on the geometry of Σe only, and on the
derivatives he ,ij , with i, j = ξ, θ, φ, which depend on both Σe and the relative
motion. Explicit expressions for f˜,ij and he ,ij are given in the Appendix, see (7.86)
and (7.85).
Equations (7.45) and (7.47) can be cast in the following linear system
    u 
[mue he pe ,θ ] gξξ [me pe ,θ z e ]
1 0

 D   
   D 

u u

 0 [me pe ,ξ he ]   gθθ  =  [me z e pe ,ξ ] 
   
, (7.49)
 1    
 D    D 
f˜,ξ f˜,θ f˜,φ gφφ c
which, again, it is solvable at non singular points on Γg , that is where the coefficient
matrix B of (7.49) verifies det(B) 6= 0. Coefficients gξξ , gθθ , gφφ are now known
quantities and we can compute kng ,sg by employing Eq. (7.36).

7.2 Third order analysis of the gap surface


In order to investigate the geometry of the gap curve obtained as the intersection
between the gap surface and the osculating plane in the neighbourhood of the
contact point P̌g on Γg ( which corresponds to P̌e on Σe ), is it possible to employ
equation (7.17)
1 i i 2 1 i
k (ε ) + kn ,si (εi )3 ,
δ i (εi ) = (7.50)
2 n 6
where i = e, g are referred to the tool and the generated gear, respectively. Under
the assumption that (7.20) still holds the tangential and normal components of
the gap surface can be written as
εe = εg = ε
1 1 e 
δ eg (ε) = (kne − kng ) ε2 + kn ,se −kng ,sg ε3 (7.51)
2 6
In the analysis to the third order of the gap function it is therefore necessary to
evaluate the relative normal curvature kneg = kne − kng given in (7.31), which is here
recalled
kneg = kne − kng = −(k · te )gφ , (7.52)
128 7. Third order analysis of gear surfaces

and the quantity



k̇neg = kne ,se −kng ,sg = muge ,sgsg −mue ,sese · te (7.53)

The second term in the first member of (7.53) has been the objective of Sec. 7.1.2
and, once the coefficients gξ , gθ , gφ and gξξ , gθθ , gφφ have been computed ac-
cording to (7.26) and (7.49), its value is easily computable by employing (7.36).
The last step to take is to find a computable expression for mue ,sese which
expands to

mue ,sese = mue ,ξξ e2ξ + 2 mue ,ξθ eξ eθ + mue ,θθ e2θ
+ mue ,ξ eξξ + mue ,θ eθθ (7.54)

In this expression the coefficients eξξ and eθθ are still unknown. A procedure for
their evaluation is outlined in the following.
Let us consider the second derivative vector

ce ,sese = pe ,ξξ e2ξ + 2 pe ,ξθ eξ eθ + pe ,θθ e2θ + pe ,ξ eξξ + pe ,θ eθθ (7.55)

Under the assumptions that ce is locally a geodesic at P̌e , we are allowed to write

kne mue = pe ,ξξ e2ξ + 2 pe ,ξθ eξ eθ + pe ,θθ e2θ + pe ,ξ eξξ + pe ,θ eθθ (7.56)

thus taking the scalar product of the previous expression by


mue × pe ,θ mue × pe ,ξ
and (7.57)
D D
we easily obtain
mue × pe ,θ 
eξξ = · pe ,ξξ e2ξ + 2 pe ,ξθ eξ eθ + pe ,θθ e2θ
D
pe ,ξ ×mue 
eθθ = · pe ,ξξ e2ξ + 2 pe ,ξθ eξ eθ + pe ,θθ e2θ (7.58)
D
Vector (7.54) has now a known expression and can be employed in (7.53) to eval-
uate the difference of the curvature derivatives, essential for the proposed third
order investigation of the gap surface according to (7.51).

7.3 Application to spiral bevel gears


The proposed analysis is further illustrated by a numerical example of a face-milled
spiral bevel pinion generated on a Gleason’s CNC machine. The pinion modelled
is a component of an actual aerospace drive. The methodology described here is
an extension of the procedure presented in Chapter 5.
7.3 Application to spiral bevel gears 129

7.3.1 Preliminary definitions


We adopt the notation already introduced in Section 5.3, where we defined as a
the axis of the machine cradle and as b the axis of the pinion blank. As well
known in gear literature [77], during generation of a spiral bevel pinion axes a and
b are skew and can experience a relative translating motion. Even if UMC-like
motions, where also the orientation of the axes can change during the generation
process, cannot be studied by the proposed approach, the range of applicability
of the present theory is quite broad and embrace the majority of the cases of
industrial interest. Here we follow the choice suggested in [11], where points Oa
and Ob (φ) are not taken on the line of shortest distance, but according to common
practice they are placed in the spindle mounting point and in the crossing point.
A schematic representation of the geometric set up during generation is given in
Fig. 7.1.
Point Oa , which is a fixed point on the fixed axis of the machine, is moved
along a by a variable parameter called sliding base ∆XB1 (φ), while point Ob (φ) is
moved along b by the machine center to back ∆XD1 , which has a constant value.
Quantity ∆EM1 (φ), called blank offset, is indeed the shortest distance between a
and b and is a function of the motion parameter φ.

head-cutter
machine
frame

cutter
axis

y
Sr


q 

y(f)
cradle Oa blank
axis a z axis
j(f)
D E M (f)


g 

D XB (f)


Ob (f)
D XD 

line of shortest
distance

Figure 7.1: Geometric set up for the generation of the spiral bevel pinion.
130 7. Third order analysis of gear surfaces

On Gleason’s CNC machines, blank offset ∆EM1 (φ) and sliding base ∆XB1 (φ)
are polynomial functions of the motion parameters of the following types
1 1
∆EM1 (φ) = ∆EM10 + V1 φ + V2 φ2 + V3 φ3 (7.59)
2 3
1 1
∆XB1 (φ) = ∆XB10 + H1 φ + H2 φ2 + H3 φ3 (7.60)
2 3
where ∆EM10 and ∆XB10 are constant values and Vi and Hi (with i = 1, 2, 3) are
called, respectively, vertical and helical motion coefficients.
Functions ψ(φ) and ϕ(φ) controlling the rotation of the cradle and the blank,
respectively, specialize now in

ψ(φ) = φ, (7.61)
 
2C 2 6D 3 24E 4 120F 5
ϕ(φ) = m0 φ − φ − φ − φ − φ , (7.62)
2 6 24 120

where the last expression is generally referred to as modified roll function and the
coefficients within as modified roll coefficients [77].
To perform the computations with the proposed approach we need to define,
in the linear space R3 , a single (fixed) reference frame S = (O; x, y, z), with unit
vectors (i, j, k). It is worth noting that all computations will be performed using
just this unique reference system.
We may select k = a and

1 a×b
j= a×b= , i=j×k (7.63)
sin γ |a × b|

where γ is the angle between a and b. Angle γ is generally expressed with respect
to the machine root angle γm1 = π/2 − γ.
In the selected reference system S the components of a and b are, respectively,

a = (0, 0, 1), b = (cos γm1 , 0, sin γm1 ) (7.64)

and the components of the variable vector d(φ) and its derivatives d0 (φ) and d00 (φ)
become

d(φ) = (∆XD1 cos γm1 , −∆EM1 (φ), ∆XB1 (φ) + ∆XD1 sin γm1 )
d0 (φ) = (0, −∆EM
0
1
0
(φ), ∆XB 1
(φ))
d00 (φ) = (0, −∆EM
00
1
00
(φ), ∆XB 1
(φ)) (7.65)

It is worth observing that the components of the vectors just defined are now
functions of the parameter of motion φ, since the relative displacement between
axes a and b during generation must be accounted for.
7.3 Application to spiral bevel gears 131

In this application we consider a generating tool Σe with a circular blade profile


and a circular fillet at the top. The geometric parameter will be labelled in the
(a)
same manner as in [4]. The parametric equations in S of the flanks pe (ξ, θ) of
the tool and their unit normal vectors mue (a) (ξ, θ) are
   
(Xp ± R1 cos ξ) cos θ + Sr1 cos q1 cos ξ cos θ
pe(a) (ξ, θ) =  (Xp ± R1 cos ξ) sin θ + Sr1 cos q1  , mue (a)
   
=  cos ξ sin θ  , (7.66)
Zp − R1 sin ξ ∓ sin ξ

¯
where ξ ∈ [ρf (1−sin αp )/(cos αp ), ξ(θ)], θ ∈ [0, 2 π), and the upper and lower signs
refer to the convex (inside blade) and concave side (outside blade), respectively.
(b)
The parametric equations of the fillets at the top of the tool pe (ξ, θ) and their
u (b)
unit normal vectors me (ξ, θ), again in S, are
   
(Xf ∓ ρf sin ξ) cos θ + Sr1 cos q1 sin ξ cos θ
p(b) mue (b)
   
e (ξ, θ) =  (Xf ∓ ρf sin ξ) sin θ + Sr1 cos q1  , =  sin ξ sin θ  , (7.67)
−ρf (1 − cos ξ) ∓ cos ξ

where ξ ∈ [0, π/2 − αp ], θ ∈ [0, 2 π). Again the upper and lower signs must be
taken when considering the convex and concave side, respectively.

7.3.2 Preliminary computations


Our goal is twofold: approximate as a cubic polynomial

1. the shape of the curve obtained as the normal section of Γg with the os-
culating (inspection) plane in the neighborhood of a selected contact point
P̌g ;

2. the shape of the section of the gap surface with the osculating (inspection)
plane through P̌g .

We have therefore to calculate the quantities kng and its derivative kng ,sg and kneg
and its derivative k̇neg . According to the procedure proposed, we have first to
evaluate certain quantities essential in the rest of the development.
First of all, we evaluate the principal normal curvatures k1e and k2e and the
components g1e and g2e in S of the principal directions g e1 and g e2 of the tool surface
Σe , along with its first coefficients E, F, G. Explicit formulas and numerical results
will be provided only for the convex active flank (conjugate to the inside blade of
the tool).
Since Σe is a surface of revolution, it is easy to verify that the coordinate lines
(dθ = 0, dξ = 0) mark the principal normal directions on Σe , and therefore we the
132 7. Third order analysis of gear surfaces

following simplified formulas hold


pe ,ξ (ξ, θ) · mue (ξ, θ) 1
k1e = − =− (7.68)
E R1
pe ,θ (ξ, θ) · mue (ξ, θ) cos ξ
k2e (ξ) = − =− , (7.69)
G(ξ) Xp + R1 cos ξ
where explicit expressions of the first coefficients of Σe are

E = R12 , F = 0, G(ξ) = (Xp + R1 cos ξ)2 , (7.70)

and the components of the principal directions in S become


   
− cos θ sin ξ − sin θ
pe ,ξ (ξ, θ)  pe ,θ (ξ, θ) 
g1e (ξ, θ) = =  − sin θ sin ξ  , g2e (θ) = p
 
√ =  cos θ (7.71)
E G(ξ)
− cos ξ 0

The components w(φ) in S of the vector w(φ), marking the direction of the screw
axis, are readily computed as follows
 
6D 2 24E 3 120F 4
w(φ) = a − ϕ0 (φ) b = a − m0 1 − 2Cφ − φ − φ − φ b (7.72)
2 6 24

along with its derivatives w0 (φ) and w00 (φ), which are omitted for brevity.
The components he (ξ, θ, φ) and its derivative he ,φ (ξ, θ, φ) are computed by
employing definitions (3.47) and (3.63), which give

he (ξ, θ, φ) = R(w(φ), a, −φ) × pe (ξ, θ) + R(ϕ0 (φ)b × d(φ) − d0 (φ), a, −φ)


= we (φ) × pe (ξ, θ) + qe (φ) (7.73)

and

he,φ = R(w0 − a × w, a, −φ) × pe



+ R ϕ00 b × d − a × (ϕ0 b × d) + (a + ϕ0 b) × d0 − d00 , a, −φ (7.74)

We are now ready to evaluate the equation of meshing in the strict sense by
means of the following expression

f˜(ξ, θ, φ) = mue (ξ, θ) · he (ξ, θ, φ) = 0 (7.75)

Then, the important scalar functions λ(ξ, θ, φ) and µ(ξ, θ, φ) are computed taking
advantage of Eqs. (3.69) at page 45, which specialize in
" # " # " #
λ(ξ, θ, φ) k1e he (ξ, θ, φ) we (φ) g1e (ξ, θ)
= · (7.76)
µ(ξ, θ, φ) −we (φ) k2e (ξ) he (ξ, θ, φ) g2e (θ)
7.3 Application to spiral bevel gears 133

Equations (3.58), (3.60) allow us to determine f˜,ξ and f˜,θ , which become

f˜,ξ (ξ, θ, φ) = − E λ(ξ, θ, φ) = −R1 λ(ξ, θ, φ) (7.77)
p
f˜,θ (ξ, θ, φ) = − G(ξ) λ(ξ, θ, φ) = −(Xp + R1 cos ξ) µ(ξ, θ, φ), (7.78)

while derivative f˜,φ is evaluated by employing (3.62) as follows

f˜,φ (ξ, θ, φ) = mue (ξ, θ) · he ,φ (ξ, θ, φ) (7.79)

The evaluation of the function of singular points in the strict sense g̃ is performed
by means of

R2 0 E he · g1e
1
1

g̃(ξ, θ, φ) = 2 0 (Xp + R1 cos ξ)2 G he · g2e (7.80)
R1 (Xp + R1 cos ξ)2
f˜,ξ f˜,θ f˜,φ

If we now choose the direction te in (3.27) of the osculating (inspection) plane


by means of the angle β, we easily compute eξ and eθ by means of (7.23). Sys-
tem (7.26) provides then the scalars gξ , gθ and gφ .
The evaluation of the normal vector to the contact line k by employing (3.68)
along with the second coefficients L, M and N in (3.18) (see page 37) allows the
determination of the curvature kng and the relative curvature kneg in the selected
direction te according to (7.33) and (7.31), respectively.
We now outline the procedure to evaluate the derivatives kng ,sg and k̇neg . By
employing equations (7.83), (7.84), (7.85) and (7.86) it is possible to evaluate the
scalar c from (7.48) and vector z e from (7.44) and to solve system (7.49). This
allows to obtain as known quantities the scalars gξξ , gθθ and gφφ and therefore to
know vector (mug ,sg sg )e from (7.37) and finally evaluate kng ,sg according to (7.36).
For the evaluation of k̇neg we have to take the further step of calculating eξξ
and eθθ from (7.58), employ (7.54) to evaluate vector mue ,se se and finally compute
k̇neg according to (7.53).

7.3.3 Numerical example


The proposed methodology is applied to investigate the shape of a spiral bevel
pinion for aerospace application, shown in Fig. 7.2, in the neighborhood of its
mean contact point M , which is in the middle of the profile and the lead.
The main data of the transmission, the geometric parameters of the convex
side (inside blade) of the grinding wheel and the machine settings employed, are
given in Tables 7.1, 7.2 and 7.3.
We define α as the counterclockwise angle between the direction τ parallel to
the contact line trough M and the tangent direction te of inspection. The plane
used to intersect both the pinion surface Γg and the gap surface is spanned by mue
and te .
134 7. Third order analysis of gear surfaces

The resulting intersections are the curves cg and ceg which are approximated
in the neighborhood of M by
1 g 1
δ g (ε) = k (ε)2 + kng ,sg (ε)3 , (7.81)
2 n 6
1 1
δ eg (ε) = kneg (ε)2 + k̇neg , (ε)3 (7.82)
2 6
As results of the proposed approach, the curvature kng (α) and its derivative kng ,sg (α)
and kneg (α) and its derivative k̇neg (α) along te , for α ∈ [0, 2π) are plotted in polar
form in Fig. 7.3-7.4.
In Fig. 7.3 (A) the distance of the generic point on the curve from the origin
represents the absolute value of kng (α) while the angle between τ and the radius
vector is the local value of α. The closed curve is not symmetric with respect
to the selected axes since we employed (τ, σ) as reference directions and not the
principal directions at M on Γg .
In Fig. 7.3 (B) the polar plot of kng ,sg (α) is shown. The shape of the curve
denotes an interesting behavior of the third order properties of Γg as the direction
of inspection varies on the tangent plane. The points on the curve are symmetric
with respect to the origin showing that |kng ,sg (α)| = |kng ,sg (α + π)|, while if we
consider the sign we have indeed kng ,sg (α) = −kng ,sg (α + π). It is worth observing
that since there two values, say α1 and α2 = α1 + π where the sign of kng ,sg
changes along the curve, the curve itself must pass two times through the origin
and therefore there must be a direction where kng ,sg vanishes.

Figure 7.2: Aerospace pinion modelled.


7.3 Application to spiral bevel gears 135

Fig. 7.4 (A) represents kneg (α). Obviously the resulting closed curve is tangent
in the origin to the direction τ , tangent to the local contact line, along which the
principal relative curvature vanishes. The curve is symmetric with respect to both
τ and σ since they are parallel to the principal relative directions.
The shape of k̇neg in Fig. 7.4 (B), like that of kng ,sg , is instead not so obvious
and predictable, since it is not symmetric with respect to the selected axes. Again
it is true that k̇neg (α) = −k̇neg (α + π) and therefore the curve is symmetric with

(A) Absolute value of kng (α), [ 10


−4
mm
]
π
2 400
2 π
3
π 3
300
5 π
6
π 200
6
100

π 0

7 11
6
π 6
π

4 5
3
π 3
π
3
2
π
(B) Absolute value of kng ,sg (α), [ 10
−4
mm2
]
π
2 3
2 π
3
π 3
2
5 π
6
π 6
1

π 0

7 11
6
π 6
π

4 5
3
π 3
π
3
2
π

Figure 7.3: Polar plots. (A): Absolute value of the normal curvature kng (α) of Γg
for α ∈ [0, 2π). (B): Absolute value of the derivative kng ,sg (α) of the the normal
curvature of Γg for α ∈ [0, 2π).
136 7. Third order analysis of gear surfaces

respect to the origin. It is worth observing that since there six values, say α1 ,
α2 , α3 and α4 = α1 + π, α5 = α2 + π and α6 = α3 + π where the sign of k̇neg
changes along the curve, the curve itself must pass six times through the origin
and therefore there must be three directions where k̇neg vanishes.
Figures 7.3-7.4 reveals that the derivatives kng ,sg and k̇neg are almost two orders
of magnitude lower than kng and kneg , respectively, and can be therefore neglected
in practical calculations without loss of accuracy.

(A) Absolute value of kneg (α), [ 10


−4
mm
]
π
2 2 250 π
3
π 3
200
150
5 π
6
π 6
100
50

π 0

7 11
6
π 6
π

4 5
3
π 3
π
3
2
π
(B) Absolute value of k̇neg (α), [ 10
−4
mm2
]
π
2 3
2 π
3
π 3
2
5 π
6
π 6
1

π 0

7 11
6
π 6
π

4 5
3
π 3
π
3
2
π

Figure 7.4: Polar plots. (A): Absolute value of the normal curvature kneg (α) of the
gap surface (relative normal curvature) for α ∈ [0, 2π). (B): Absolute value of the
derivative k̇ng (α) of the the normal curvature of the gap surface for α ∈ [0, 2π).
7.4 Conclusions 137

7.4 Conclusions
Explicit formulas for the coefficients of the third-order polynomial approximation
of the gear surface and of the gap surface in the neighborhood of an arbitrary
contact point have been provided within the framework of the invariant approach.
The proposed procedure has been employed to evaluate quantitatively the third
order properties of the surface of a spiral bevel pinion in the neighborhood of the
mean contact points on its convex side.
The information obtained could be employed also to evaluate the distribution of
the contact pressures more accurately as reported in [7] and, as suggested in [84],
to examine more closely the relationship between the conjugate surfaces in the
neighborhood of a flat contact point, which occur when an envelope to the contact
lines exists on the tool surface.

Parameter name Symbol Value


Pinion tooth number N1 27
Gear tooth number N2 38
Module m 4.950 mm
Mean spiral angle β 35.0 deg
Pinion Hand LH –
Face width Fw 32.0 mm
Outer cone distance A0 97.8706 mm
Pinion face angle γa 1 46.712 deg

Table 7.1: Main input data of the transmission modelled.

Parameter name Symbol Value


Spherical radius R1 173.7800 mm
Spherical center x-offset Xp -80.5632 mm
Spherical center z-offset Zp 60.8096 mm
Blade angle αp 20.4826 deg
Fillet radius ρf 0.7620 mm

Table 7.2: Geometric parameters of the inside blade (convex side).

Parameter name Symbol Value


Radial setting Sr 1 76.5559 mm
Blank offset ∆EM1 -1.7272 mm
Root angle γm 1 40.629 deg
Mach. center to back ∆XD1 3.521050 mm
Sliding base ∆XB1 -2.284300 mm
Cradle angle q1 57.9919 deg
Ratio of roll m0 1.512830
M.R. 1st coeff. 2C 0.004968
M.R. 2nd coeff. 6D -0.056380
M.R. 3rd coeff. 24E 0.000000
M.R. 4th coeff. 120F 0.000000

Table 7.3: Machine settings employed for grinding the pinion convex side.
138 7. Third order analysis of gear surfaces

7.5 Appendix
Derivatives (mug ,ij )e (i = ξ, θ, φ, j = φ) in (7.37)

(mug ,φ )e = R(w0 (φ), a, −ψ(φ)) × mue


(mug ,ξφ )e = R(w0 (φ), a, −ψ(φ)) × mue ,ξ
(mug ,θφ )e = R(w0 (φ), a, −ψ(φ)) × mue ,θ

(mug ,φφ )e = R −ϕ0 (φ) b × (w0 (φ) × R(mue , a, ψ(φ))) + w00 (φ) × R(mue , a, ψ(φ))

+ w0 (φ) × (ψ 0 (φ) a × R(mue , a, ψ(φ))), a, −ψ(φ) (7.83)

Derivatives (pg ,ij )e (i = ξ, θ, φ, j = φ) in (7.44)

(pg ,ξφ )e = he ,ξ = we (φ) × pe ,ξ


(pg ,θφ )e = he ,θ = we (φ) × pe ,θ
(pg ,φφ )e = (ψ 0 a − ϕ0 R(b, a, −ψ(φ))) × he + w0e (φ) × pe + q 0e (φ) (7.84)

Derivatives f˜,ij (i, j = ξ, θ, φ) in (7.48)

f˜,ξξ = mue ,ξξ ·he + 2mue ,ξ ·he ,ξ +mue · he ,ξξ


f˜,θθ = mue ,θθ ·he + 2mue ,θ ·he ,θ +mue · he ,θθ
f˜,ξθ = mu ,ξθ ·he + mu ,ξ ·he ,θ +mu ,θ ·he ,ξ +mu · he ,ξθ
e e e e

f˜,ξφ = mue ,ξ ·he ,φ +mue · he ,ξφ


f˜,θφ = mue ,θ ·he ,φ +mue · he ,θφ
f˜,φφ = mu · he ,φφ
e (7.85)

Derivatives (he ,ij )e (i, j = ξ, θ, φ) in (7.48)

he ,ξξ = we × pe ,ξξ he ,ξφ = w0e × pe ,ξ +q 0e


he ,ξθ = we × pe ,ξθ he ,θφ = w0e × pe ,θ +q 0e
he ,θθ = we × pe ,θθ he ,φφ = w00e × pe ,ξ +q 00e
Chapter 8

Identification of CNC
parameters for
micro-geometry optimization

8.1 Introduction
In production of critical gear sets, like hypoid and spiral bevel gears for aerospace
applications, the contact pattern, which can be identified as the union of the bear-
ing ellipses in one cycle of meshing, must satisfy some requirements in terms of
dimension, shape and position within the boundaries of the tooth.
A contact pattern very close to the boundaries of the active flank or with a
short extension can weaken the mechanical resistence of the tooth with respect to
its potential.
In order for the gear designer to optimize the contact pattern, the ability to
modify the micro-geometry of the tooth surfaces with respect to a basic design
has to be gained. Modern Computer Numerically Controlled (CNC) machines
for cutting and grinding hypoid and spiral bevel gears, in addition to the basic
motions of the cradle style machine, are capable of supplemental motions, which
allow for accurate optimization of the tooth surface and result in a large number
of parameters to be tuned.
In gear industry, only a small number of parameters is considered in the process
of optimization of the tooth; this because it is impossible even for the most skilled
operator to gain an accurate sensitivity on the combined effects of the whole set of
the available parameters. Another reason is that the use of proprietary softwares
capable of these tasks (like the ones by Gleason and Klingelnberg) can have too
high costs for the budget of middle-sized companies.
The process of optimizing the contact pattern of a gear pair consists of two
140 8. Identification of CNC parameters for micro-geometry optimization

main tasks: the first one is to identify the ease-off surface, that represents the
difference between the optimized (final) and the basic (initial) surface of the tooth,
which ensures the desired contact pattern under load or at least a very good
approximation of it; the second one is to determine the machine-tool settings to
be employed to obtain a surface as closest as possible to the pre-designed one. The
outlined methodology could be automatized.
The solution of the previous problem is worth a dedicated thesis work itself.
The aim of this chapter is to introduce various aspects related to the second task
and present some techniques for its solution. Therefore, the hypothesis we stand
on is that, in some way, we are already given the best ease-off which allows us to
obtain the desired contact pattern.
The developed procedures are implemented via the invariant approach in a
computer code and their results for the optimization of a spiral bevel gear set for
aerospace application are presented.

8.2 Gear generation with supplemental motions


As previously outlined, modern CNC machines developed mainly by Gleason and
Klingelnberg are capable of fully general 3D motions of the tool with respect to the
gear blank. These allow for complex supplemental motions to be addressed which
ensures, to some extent, a free-form shaping of the tooth, sometimes indicated
with kinematic optimization. The set of these motions, named by Klingelnberg
modified motions, in the Gleason implementation is identified as the Universal
Motion Concept.
For the geometric description of the resulting generation process the proposed
invariant approach is employed. The use of a vectorial approach, with compact
relations, avoids the introduction of a chain of reference systems typical of the
traditional approach and gives more insight into the physical process.
To proceed with the theoretical treatment it is useful to introduce the notation
used throughout the rest of the chapter.
Let us define x ∈ Ω ⊂ Rm the vector of the machine-tool settings, where Ω
is an open set of Rm . We do not distinguish between parameters related to the
geometry of the tool surface and those related to the machine kinematics. More-
over, the terms machine-tool settings and machine parameters will be considered
as synonymous and used as such.
The vector of the basic (initial) machine-tool settings is defined as x(0) ∈ Ω ⊂
m
R and the finite variation of the updated settings with respect to the basic ones
is identified by ∆x = x − x(0) ∈ Ω ⊂ Rm .
To compact the notation, let us define ζ = (ξ, θ, φ) ∈ B ⊂ R3 , the vector of
the Gauss and motion parameters, where B is an open set of R3 . Since we will
mainly deal with a finite number of points on a grid sampled on the tooth surface,
we also define the discrete version ζ i = (ξi , θi , φi ) ∈ B ⊂ R3 . This represents the
vector of the Gauss and motion parameters for the i−th point of a grid sampled
8.2 Gear generation with supplemental motions 141

on the tooth surface with settings x, where i = 1, . . . , n and n is the total number
of points on the grid. We consider that the n points Pg (ζ i , x) on the grid include
both the tooth fillet and the active flank.
For the ζ i to actually represent a triplet on the envelope surface, the equation
of meshing must be satisfied, that is


f (ζ i , x) = [pg ,ξ pg ,θ pg ,φ ] =0 (8.1)
(ζ i ,x)

From equation (8.1), it is evident the different treatment that the ζ i ’s (i = 1, . . . , n)


and the machine parameters x deserve. While x is a set of global parameters,
the ζ i ’s vary from point to point on the surface grid. Moreover, while we can
control the machine-tool settings x by modifying the geometry of the tool and the
kinematic of the machine, we do not have direct control on the parametrization of
the hyper-family of surfaces and therefore on the ζ i ’s.
As done previously with the machine parameters, we define the i−th triplet
(0)
on the basic tooth surface as ζ i and the finite variation of the updated surface
(0)
parameters with respect to the basic one as ∆ζ i = ζ i − ζ i .
With the adopted notation, it is possible to define the hyper-enveloping family
of surfaces Φx g , whose position vector is given by pg (ζ, x). The attribute hyper-
enveloping can be explained considering that if the machine-tool parameters are
not fixed yet, still ∞m equally possible enveloping families (since x ∈ Rm ) are
(0)
available. For example, the basic enveloping family of surfaces Φg = Φg is ob-
tained as a special case by setting x = x(0) and therefore its position vector is
(0)
pg (ζ) = pg (ζ, x(0) ).
It is worth noting that, in this general framework, the basic envelope surface
(0)
Γg = Γg is defined as the solution of the following system
(
sg = pg (ζ, x(0) )
(8.2)
f (ζ, x(0) ) = 0

Analogously, the generic envelope surface Γ̃g with machine settings x = x̃ is


defined as
(
sg = pg (ζ, x̃)
(8.3)
f (ζ, x̃) = 0

Adopting the same notation used in the previous chapters, let us define mug (ζ, x)
the unit normal vector to the hyper-enveloping family as

pg ,ξ ×pg ,θ
mug (ζ, x) = , (8.4)
|pg ,ξ ×pg ,θ |
142 8. Identification of CNC parameters for micro-geometry optimization

that is the normal vector to each surface of the sequence obtained varying the
motion parameter φ, once the machine parameters x have been fixed.
The unit normal vector to each of the basic i−th point (i = 1, . . . , n) is defined
u (0)
as mgi = mug (ζ i , x(0) ).

8.3 Synthesis of the tooth micro-geometry


If we expand in Taylor series up to the first order the hyper-family of surfaces
(0) (0)
pg (ζ, x) with respect to the basic i−th point with position vector pgi = pg (ζ i , x(0) ),
(0)
that is around the point of R3 with Gaussian and motion parameters ζ i , and
also perturbing the machine settings x, we obtain
(0) (0) (0) (0)
pg (ζ i +dζ i , x(0) +dx) ' pg (ζ i , x(0) ) + pg ,ζ (ζ i , x(0) )dζ i + pg ,x (ζ i , x(0) )dx,
(8.5)

where pg ,ζ ∈ R3×3 and pg ,x ∈ R3×m are the jacobians of the hyper-enveloping


surface with respect to the surface parameters and the machine-tool settings, re-
spectively.
Considering a finite variation of the family in both the surface parameters ∆ζ i
and the machine settings ∆x we can write the following approximation
(0) (0)
pg (ζ i +∆ζ i , x(0) +∆x) ' pg (ζ i , x(0) )+
(0) (0)
+ pg ,ζ (ζ i , x(0) )∆ζ i + pg ,x (ζ i , x(0) )∆x, (8.6)

and therefore the difference can be written as


(0) (0) (0)
∆pg (∆ζ i , ∆x, ζ i , x(0) ) = pg (ζ i + ∆ζ i , x(0) + ∆x) − pg (ζ i , x(0) )
(0) (0)
' pg ,ζ (ζ i , x(0) )∆ζ i + pg ,x (ζ i , x(0) )∆x (8.7)

For fixed ∆x and regardless of its entity, equation (8.6), with i = 1, . . . , n, rep-
resents n families of planes, each one parallel to the basic tangent plane at point
(0)
Pg (ζ i , x(0) ).
We now define the desired surface “very” close to the basic one, by assigning
the position vectors of a grid of n points on it with respect to the corresponding
grid on the basic surface. The most natural way of defining this new surface is by
(0)
imposing the position vector pg (ζ i + ∆ζ i , x(0) + ∆x) of the new point P̃g (ζ i , x)
to be such that
(0) (0) (0)
pg (ζ i + ∆ζ i , x(0) + ∆x) − pg (ζ i , x(0) ) = mug (ζ i , x(0) ) hi , (8.8)

where the scalar values hi (i = 1, . . . , n) represent the amount of distance of the de-
sired surface with respect to the basic one, measured in the normal direction to each
8.3 Synthesis of the tooth micro-geometry 143

target surface
Gg

h
h


  

Gg   
hi
z   

z 
h  

  

zi
  

z  

hn
  

zn

basic surface

Figure 8.1: Ease-off definition and target surface.

(0)
point Pg (ζ i , x(0) ) on the basic surface grid. Often the vector h = (h1 , . . . , hn )
describing the whole modification is defined as ease-off.
The above mentioned situation is illustrated in Fig. 8.1.
Considering equations (8.7) and (8.8), it is possible to obtain a relationship
between the ease-off hi prescribed at each point of the grid and the corresponding
variations ∆ζ i and ∆x, as follows

(0) (0) (0)


mug (ζ i , x(0) ) hi ' pg ,ζ (ζ i , x(0) )∆ζ i + pg ,x (ζ i , x(0) )∆x (8.9)

A further step can now be pursued. If we dot multiply both sides of (8.9) by
(0)
mug (ζ i , x(0) ) we obtain

(0) (0) (0) (0)


hi ' mug (ζ i , x(0) ) · pg ,ζ (ζ i , x(0) )∆ζ i + mug (ζ i , x(0) ) · pg ,x (ζ i , x(0) )∆x
(0) (0)
= mug (ζ i , x(0) ) · pg ,x (ζ i , x(0) ) ∆x (8.10)

since, for the equation of meshing f = 0 to be fulfilled, we have


(0) (0)
mug (ζ i , x(0) ) ∈ ker(pg ,ζ (ζ i , x(0) )) (8.11)
(0)
that is mug (ζ i , x(0) ) is orthogonal to the partial derivatives with respect to the
Gaussian and motion parameters ζ of the position vector of the family pg (ζ, x).
(0)
In other words, the column vectors of the jacobian pg ,ζ (ζ i , x(0) ), which span the
144 8. Identification of CNC parameters for micro-geometry optimization

(0) (0)
tangent plane to Γg at Pg (ζ i , x(0) ), are orthogonal to the unit normal vectors
(0)
mug (ζ i , x(0) ).
By letting
(0)
sT (ζ i , x) = mug (ζ i , x(0) ) · pg ,x (ζ i , x), (8.12)

where sT ∈ R1×m equation (8.10) can be rewritten in a more compact form as


(0)
hi ' sT (ζ i , x(0) ) ∆x (8.13)

Equation (8.13) represents the relationship between the prescribed ease-off at


the i−th point and the variation in the machine-tool settings. Not surprisingly,
variations in the surface parameters ζ do not appear in (8.13) and the only impor-
tant modification occurs in the machine parameters x. This can be explained con-
sidering that at the first order of expansion, the hyper-enveloping family pg (ζ, x)
is approximated with an hyper-family of planes parallel to the tangent planes to
(0)
Γg at each point on the grid and the distance of the planes of each family to
the basic ones depends only on ∆x. Therefore, prescribed variation in the normal
direction can be obtained only for suitable modifications of x.
(k)
For further developments it is useful to define the set of the n triplets ζ i
(k) (k) (k) +
(i = 1, . . . , n) as Ξ = (ζ 1 , . . . , ζ n ), where k ∈ Z is an iterator. Besides, we
make the position sTi = sT (ζ i , x).
By collecting equation (8.13) for all the n points on the grid and we define
matrix S = (sT1 , . . . , sTn ), with S ∈ Rn×m , where its generic entry has the following
expression
  (0)
S(Ξ, x) ij = mug (ζ i , x(0) ) · pg ,xj (ζ i , x), (8.14)

we get

h ' S ∆x. (8.15)

In the previous equation we set S = S(Ξ(0) , x(0) ) for brevity.


Usually, n > m, since we impose the ease-off condition on a number of points
n larger than the number m of machine parameters we are allowed to change.
This means that the system (8.14) it is hyperdetermined (too many equations)
and therefore has no solution, since in general h ∈/ Im(S).
Usually, the sought for solution to this kind of problems is the one that mini-
mizes, in some norm, the residual vector ε ∈ Rn , defined as

ε = S ∆x − h (8.16)

Depending on the metric we choose to define distances in Rn , we end up with dif-


ferent algorithms and different solutions to (8.16), as it will be discussed hereafter.
8.4 Least squares solution 145

8.4 Least squares solution


8.4.1 The full column rank case
A first approach to solve sistem (8.16) is to seek ∆x as the solution of the following
problem

min ||S ∆x − h||22 , (8.17)


∆x

where the L2 (Euclidean) norm of the residuals is minimized. Now we assume that
dim(ker(S)) = 0, that is S has full column rank. By imposing the stationarity
condition
∂  
(S ∆x − h)T (S ∆x − h) = 2 (S TS ∆x − S T h) = 0, (8.18)
∂∆x
we obtain the least square solution

∆x = (S TS)−1 S T h = S L h, (8.19)

where we defined a left inverse S L = (S TS)−1 S T ∈ Rm×n , with S L S = I m and


where the invertibility of S T S is guaranteed under the hypothesis that S has an
empty null space.
If we adopted a different metric in the vector space of the residuals, again in
the Euclidean sense, that is if we set ||e||2 = eT W e (with W ∈ Rn×n symmetric
and positive definite matrix), we would end up with the following solution

∆x = (S T W S)−1 S T W h = S L
W h, (8.20)

where we implicitly defined the weighted left inverse S L T


W = (S W S)
−1 T
S W,
L
where again S W ∈ Rm×n .
Therefore, we see from the previous considerations that the solution of the
least squares problem is readily available once we have computed the left inverse
corresponding to the chosen norm in the vector space of the residuals.

8.4.2 The column rank deficiency case


A more general way to find the left inverse of a matrix, even in the case that
dim(ker(S)) 6= 0 is to apply the Singular Value Decomposition to S and to compute
the pseudoinverse S + of S (see [78], [23]).
It can be shown that matrix S (like any other rectangular matrix, either sin-
gular or not) can be written as the following product of matrices

S = U ΣV T. (8.21)
146 8. Identification of CNC parameters for micro-geometry optimization

To understand the meaning of (8.21) we have first to consider the diagonaliza-


tion of real symmetric positive semi-definite matrices obtained from S by scalar
multiplication with its transpose S T , that is
SS T = U M U T (8.22)

ST S = V N V T , (8.23)
n×n
where U ∈ R has the columns composed by n orthonormal eigenvectors of
SS T ∈ Rn×n ordered such that in the diagonal of M ∈ Rn×n the eigenvalues of
SS T have decreasing norms; similarly, V ∈ Rm×m has the columns composed by
m orthonormal eigenvectors of S T S ∈ Rm×m ordered such that N ∈ Rm×m has
in its main diagonal the eigenvalues of S T S with decreasing norms.
In the event that S is not full column rank, that is rank(S) = r ≤ m, the last
n−r columns of U and the last m−r columns of V belong to ker(SS T ) = ker(S T )
and to ker(S T S) = ker(S), respectively, and they form an orthonormal basis of
these spaces. In fact, in this case, the elements mii and nii on the diagonal of M
and N , respectively, with i > r are zero.
It is worth noting that the first r (different from zero) values on the diagonal
of M and N are the same and are positive, that is λi > 0, i = 1, . . . , r. These
matrices have then n − r and m − r null eigenvalues on the diagonal, respectively.
This is a direct consequence of the fact that the eigenvalues of square products
of two rectangular matrices are the same except for a number of null eigenvalues
equal to the difference in the dimensions of the two square matrices.
It is therefore possible to write M = ΣΣT and N = ΣT Σ, with Σ ∈ Rn×m ,
where
p ∆
Σii = λi = σi , i = 1, . . . , r (8.24)
Σij = 0, i 6= j (8.25)
The r positive values σi are called singular values of S.
Matrices U and V considered till now represent orthonormal bases arbitrarily
chosen, except for the fact that the first columns are bases of Im(S) and S T and
the last columns are bases of ker(S T ) and ker(S). We can therefore take a further
step by choosing a relationship between the four fundamental subspaces which
will be particularly useful. It is possible to make it by requiring the following
relationship to be satisfied
S = U ΣV T , (8.26)
which represents the Singular Value Decomposition (SVD) of S. To obtain the
decomposition we can therefore choose arbitrarily an orthonormal basis of Im(S T )
in the first r columns of V and find the corresponding columns of U solving
SV = U Σ. The columns of U can be computed as follows
1
U (: , i) = S V (: , i), i = 1, . . . , r (8.27)
σi
8.5 Solution by subset selection using the SVD 147

It is easy to verify that these solutions are orthonormal columns (U (:, i)T U (:, j) =
δij , where δij is the Kronecker symbol) and that they form a basis of Im(S):
therefore, they are allowable choices for the first r columns of U . For the rest of
the columns of U and V , it will suffice take the orthogonal complements (e.g.,
by employing the Gram-Schmidt orthogonalization process). The procedure here
outlined is neither efficient nor numerically stable, but has the purpose to recall
the main ideas of the process leading to the SVD decomposition of a matrix which
is used to define its pseudoinverse. For a detailed discussion on this topic see [23].
By employing the SVD of matrix S, the problem to find the solution to the
general system S∆x = h, in the sense of finding the best approximation of min-
imum norm, has a readily available solution. By rewriting U ΣV T ∆x = h and
changing the coordinates in both domains by setting ∆x = V ∆x̄ and h = U h̄,
in the new coordinates we have Σ∆x̄ = h̄, that is


 σ1 ∆x̄1 = h̄1

 .. .
= ..




 .

σr ∆x̄r = h̄r (8.28)


 0 ∆x̄r+1 = h̄r+1




 ··· = ···


0 ∆x̄n = h̄n
The solution which minimizes L2 norm of the residual vector is therefore given
by ∆x̄1 = σ11 h̄1 , . . . , ∆x̄r = σ1r h̄r , with the remaining components of ∆x̄ which
do not influence the residual vector and therefore can be chosen arbitrarily. On
the other hand, to minimize ∆xT ∆x = ∆x̄T V T V ∆x̄ = ∆x̄T ∆x̄, the remaining
n − r − 1 components must be chosen equal to zero.
Coming back to the original coordinates we obtain the desired solution. The
pseudoinverse matrix S + which solves the problem S∆x = h in the least squares
sense is therefore
S + = V Σ+ U T , (8.29)
+ m×n
where Σ ∈ R has on the main diagonal the reciprocal of the r singular values
of S and zero elsewhere; the sough for solution in the machine-tool settings can
then be obtained as follows
∆x = S + h (8.30)
Obviously, if matrix S has full column rank, solution (8.30) is identical to that
obtained in (8.19).

8.5 Solution by subset selection using the SVD


Often, due to the similarity of the effects on the tooth surface modifications of
different machine-tool settings, the columns of S are quasi linearly dependent and
148 8. Identification of CNC parameters for micro-geometry optimization

matrix S comes close to singular (column rank deficiency). In these cases we can
take two different routes toward the solution.
The first one is to precondition matrix S by chopping the singular values which
are smaller than a threshold value, and which would make the entries blow up when
computing the pseudoinverse S + . In other words, the problem is approached by
approximating the least squares solution (8.30), which is here rewritten in an
equivalent way,
r
X 1
∆xLS = U (: , i)T h V (: , i), r = rank(A) (8.31)
σ
i=1 i

with

X 1
∆xr̃ = U (: , i)T h V (: , i), r̃ ≤ r (8.32)
i=1
σ i

where
r
X
T
S = U ΣV = σi U (: , i)V (: , i)T (8.33)
i=1

and r̃ is some numerically determined estimate of r. Note that ∆xr̃ minimizes


||S r̃ ∆x − h||2 where

X
S r̃ = σi U (: , i)V (: , i)T (8.34)
i=1

is the closest matrix to S that has rank r̃.


Replacing S by S r̃ in the least squares (LS) problem amounts to filtering the
small singular values and can make a great deal of sense in those situations where
S is derived from noisy data.
This approach has been pursued in [36] and does present some drawbacks as
reported in [37], the most evident being the lack of control on the values actually
assumed by the machine-tool settings which cannot be constrained a priori within
the feasible region Ω ⊂ Rm .
In the present application, however, rank deficiency implies redundancy among
the machine-tool settings comprised in the underlying model. Therefore, it can
be more appropriate to follow a different path to the solution. Since we are not
interested in a predictor such as S r̃ ∆xr̃ that involves all m redundant factors,
a predictor S y may be sought, instead, where y ∈ Rm has at most r̃ nonzero
components. The position of the nonzero entries determines which columns of S,
i.e., which machine-tool settings in the model, are to be used in approximating the
objective ease-off h. How to pick these columns is the problem of subset selection
which is the subject of this section. The main idea is that if r̃ is the numerical
8.5 Solution by subset selection using the SVD 149

rank of S, there is a distinguished r̃ dimensional subspace of the column space


of S which is insensitive to how it is approximated by r̃ independent columns of
S and this fact can be profitably employed in the solution of the least squares
problem.
It is worth noting that the numerical rank r̃ is estimated by starting from m
and iteratively reduce it in order for the solution y to fall within a feasible set,
that is y ∈ Ω ⊂ Rm .
The procedure outlined here is a classical one in numerical linear algebra,
see [23] and [22] for further details, but its application to the identification of
machine settings in gear design seems to be original.

8.5.1 Pseudocode of the algorithm


In this section a pseudocode of the algorithm implemented for obtaining a solution
with subset selection using the SVD is presented. The original problem can be
formulated as follows

 min ||S ∆x − h||2
∆x (8.35)

subject to ∆xmin ≤ ∆x ≤ ∆xmax

Given S ∈ Rn×m and h ∈ Rn , the following algorithm computes a permutation


matrix P ∈ Rm×m , a rank estimate r̃ and a vector y 1 ∈ Rr̃ such that the first
r̃ columns of B = S P are independent and such that ||B(: , 1 : r̃) y 1 − h||2 is
minimized while ∆x = P (y T1 , oT )T (where o ∈ Rm−r̃ ) is tried to be kept in the
range [∆xmin , ∆xmax ].

Step 1 Compute the SVD of S, that is

S = U ΣV T, i.e. U T SV = diag(σ1 , . . . , σn )
U ∈ Rn×n ; Σ ∈ Rn×m ; V ∈ Rm×m ;
save V ;

Step 2 Set counter k = 0;

Step 3 Set current number of columns deleted s = k

Step 4 Estimate the numerical rank r̃ = r(k) of S as r(k) = m − k;

Step 5 Set V r̃ = V (: , 1 : r̃)T , with V r̃ ∈ Rr̃×m and apply the QR decomposition


150 8. Identification of CNC parameters for micro-geometry optimization

with column pivoting to V r̃ , that is


 
V r̃ P = QR → QT V r̃ P = R = R11 R12 ;
|{z} |{z}
r̃ m−r̃

extract P ; set B = S P = [B 1 B 2 ],

with B 1 ∈ Rn× r̃ and B 2 ∈ Rn×(m−r̃) ;

set y = P T ∆x;

Step 6 Extract both the column and the variable subsets:

B 1 = B(: , 1 : r̃); y 1 = y(1: r̃);

Step 7 Determine y 1 = B + r̃
1 h with y 1 ∈ R such that ||B 1 y 1 − h||2 = min, by
+
employing the pseudoinverse B 1 computed with the SVD algorithm;
Step 8 Reconstruct the whole solution vector ∆x by appending the null vector
o ∈ Rm−r̃ to y 1 and reordering the components according to P as
" #
y1
∆x = P ; (8.36)
o

Step 9 Check number of iterations: if s = smax with (smax ≤ m − 1)


then EXIT (no solution found within the feasible region deleting smax columns)
else CONTINUE;
Step 10 Check constraint violation: if ∆xmin ≤ ∆x ≤ ∆xmax
then EXIT (found solution within the feasible region)
else set k=k+1 and GOTO Step 3;
It is important to note that, in general, no warranty is given that the proposed
algorithm will find a solution in the feasible region within an assigned number of
steps smax , or equivalently, by reducing of smax ≤ m − 1 the dimension of the
column space of S. On the other hand, if we require the ease off vector h to
be representative of reasonable modifications of the tooth surface and we allow
a sufficiently large number of machine-tool settings to be selected from, which
can vary in sufficiently large ranges, there are good chances that the algorithm
will return a feasible solution. Moreover, it will specify variations only on the
most independent among the parameters, thus avoiding redundancy and helping
to identify the cause-effect relationship between relevant parameters and their
associated surface modifications. A numerical example of the proposed algorithm
relative to the optimization of an aerospace pinion will be presented in section
(application).
8.6 Quadratic programming solution 151

8.6 Quadratic programming solution


A straightforward way to solve problem (8.35) is to solve the equivalent problem
of minimizing the square of the objective function. In fact, since the “square”
function is a convex function, the minimum of (8.35) will be attained in the same
point where the minimum of the following problem is located
 2
 min ||S ∆x − h||2
∆x (8.37)

subject to ∆xmin ≤ ∆x ≤ ∆xmax
Facing the problem (8.37) of finding a constrained minimum of a quadratic ob-
jective function, we can profitably employ quadratic programming (QP) algo-
rithms. The algorithms usually available in numerical computing packages (see,
e.g., ‘quadprog’ in [68]) to solve quadratic programming problems, require the
following structure of the objective function

1 

 min ∆xT H ∆x + f T ∆x
∆x 2 (8.38)


subject to ∆xmin ≤ ∆x ≤ ∆xmax
Therefore, the Hessian matrix H and the row vector f T have to be supplied to
the routine. By means of a straightforward computation
||S∆x − h||22 =(S∆x − h)T (S∆x − h)

=∆xT (S T S) ∆x − 2hT S∆x + hT h, (8.39)


where we used the property that the transpose of scalar quantity is equal to its
transpose, we obtain the sought for quantities to feed the routine, that are here
reported
H = 2 (S T S)

f = −2 S T h (8.40)

Obviously, the scalar quantity hT h is neglected in the optimization algorithm,


since it represents a constant translation and therefore has no influence on the
solution. To solve the QP problem (8.38) we employed the Optimization Toolbox
routines available through the MATLAB environment. The results obtained in
a numerical test regarding the optimization of a real spiral bevel pinion will be
presented in the last section of this chapter.

8.7 Linear programming solution


An elegant solution to problem (8.16) can be obtained if the L∞ norm (infinity
norm) or the L1 norm (Manhattan norm) of the residual vector ε is minimized.
152 8. Identification of CNC parameters for micro-geometry optimization

The reason is that by choosing either L∞ or L1 norm the problem structure fits the
standard Linear Programming (LP) problem structure for which efficient solution
algorithms exist (see [82], [27]). Moreover, within the framework of the standard
LP problem the constraints on the variations ∆x of the machine-tool parameters
can be easily taken into account and therefore practical solutions can be found.

8.7.1 Minimization of the L∞ norm


Let us consider the minimization of the L∞ norm of the residual vector ε in (8.16).
The problem can be stated as

 min ||S ∆x − h||∞
 ∆x


subject to A ∆x ≥ b (8.41)




∆xmin ≤ ∆x ≤ ∆xmax

where A ∈ Rl×m and b ∈ Rl account for linear inequality constraints among the
machine parameters and ∆xmin and ∆xmax are the lower and upper bounds for the
machine-tool setting variations. In the problem of identification of the variations
for the sake of micro-geometry optimization, there are usually no linear constraints
among the parameters but, nonetheless, we consider here the problem in its most
general form.
The equivalent form of (8.41) expressed in components is

  
 T


 min max |s i ∆x − h i |
 ∆x i = 1, . . . , n

(8.42)

 subject to aTj ∆x ≥ bj , j = 1, . . . , l



 ∆xminq ≤ ∆xq ≤ ∆xmaxq , q = 1, . . . , m

where sTi ∈ R1×m is the i-th row of matrix S, hi the i-th component of the ease-
off vector h, aTj ∈ R1×m the j-th row of linear constraint matrix A, bj the j-th
component of vector b and ∆xminq , ∆xmaxq the lower and upper bounds for the
variation of the q-th component ∆xq .
The equivalent problems (8.41) and (8.42) are still in forms which are not
easily solvable, mainly because of the discontinuity of the objective function, which
involve the function max.
In [72] it is demonstrated that problem (8.42) can be transformed into an
equivalent one, in the sense that they share the same solution. This new problem,
in its turn, can be rewritten as a standard LP problem with little effort. The
8.7 Linear programming solution 153

equivalent problem has the following structure



 min α


 ∆x


subject to |sTi ∆x − hi | ≤ α, i = 1, . . . , n

(8.43)
aTj ∆x ≥ bj , j = 1, . . . , l







∆xmin q ≤ ∆xq ≤ ∆xmax q , q = 1, . . . , m
where the difficulty in the management of the objective function has been shifted
to the constraints. Now we introduce a new variable and some new quantities
in order to rewrite problem (8.43) as an equivalent standard linear programming
one. Let us therefore introduce a new variable y ∈ Rm+1 obtained by stacking the
unknown scalar α with the machine-tool setting variations ∆x, that is
 
α
y= (8.44)
∆x
and the column vector e ∈ Rm+1 obtained appending to number 1 a vector of m
zeros, that is
 
eT = 1 | 0 · · · 0 (8.45)
such that minimizing α is indeed equivalent to minimizing
 
T
  α
e y = 1 | 0···0 =α (8.46)
∆x
Now we care about the constraints
|sTi ∆x − hi | ≤ α, i = 1, . . . , n (8.47)
These inequalities are indeed equivalent to the two sets of inequalities that follow
sTi ∆x − hi ≤ α, α − sTi ∆x ≥ −hi
⇐⇒ with i = 1, . . . , n (8.48)
sTi ∆x − hi ≥ −α, α + sTi ∆x ≥ hi ,

which can be stacked together with the original A ∆x ≥ b to obtain


       
1
  |   −S   −h 
   
 1   
      " #  
 1   
  α  
  |  
 S  
 ≥  h


 (8.49)
 1  ∆x  
       
 0   
   
 |   A    b  
0
154 8. Identification of CNC parameters for micro-geometry optimization

Now we set
" # " #
M1 m1
M= , m= (8.50)
M2 m2

where
   
1
 |   −S     

 1

 0
M1 =  
 1
 ,
 M 2 =  |   A  (8.51)

 |  
 0
S 
1

and
" #
−h
m1 = , m2 = b (8.52)
h

with M 1 ∈ R2n×(m+1) that represents the constraint matrix induced by the L∞


norm, M 2 ∈ Rl×(m+1) that accounts for possible linear inequality constraints
among the machine-tool parameters and with m1 ∈ R2n and m2 ∈ Rl .
Finally, by setting
" # " #
−∞ +∞
y min = , y max = (8.53)
∆xmin ∆xmax

we can write (8.43) as the following standard LP problem



 min eT y
 y


subject to My≥m (8.54)




y min ≤ y ≤ y max

for which efficient solution algorithms exist. Here, we will not dwell into the
mechanics of the algorithms like the revised simplex or the interior point method
suitable to attach these kinds of problems, for which excellent references, like [82]
and [27], are available. The above mentioned algorithms will be employed to solve
a real case and the results obtained will be discussed.

8.7.2 Minimization of the L1 norm


Let us now consider the minimization of the L1 norm of the residual vector ε
in (8.16). This norm, which is given by the sum of the absolute values of the
8.7 Linear programming solution 155

components of a vector, is often called Manhattan norm since it remainds the way
distances are measured between two locations in the Manhattan district, being the
components of the vector, the distances along the streets and the avenues covered
to join the two points. The problem can be stated as

 min ||S ∆x − h||1
 ∆x


subject to A ∆x ≥ b (8.55)




∆xmin ≤ ∆x ≤ ∆xmax

where A ∈ Rl×m and b ∈ Rl account for linear inequality constraints among the
machine parameters and ∆xmin and ∆xmax are the lower and upper bound vectors
for the machine-tool settings. As previously stated in § 8.7.1, in the problem of
identification of parametric variations for micro-geometry optimization, no linear
constraints are usually imposed among the parameters. Nevertheless, we shall
consider here the problem in its most general form since no particular complication
arises in consequence of this choice.
The equivalent form of (8.55) expressed in components is
 n
X 
 T


 min |si ∆x − hi |
 ∆x i=1

(8.56)


 subject to aTj ∆x ≥ bj , j = 1, . . . , l



∆xminq ≤ ∆xq ≤ ∆xmaxq , q = 1, . . . , m

where we recall that sTi ∈ R1×m is the i-th row of matrix S, hi the i-th component
of the ease-off vector h, aTj ∈ R1×m the i-th row of linear constraint matrix A, bj
the j-th component of vector b and ∆xminq , ∆xmaxq the lower and upper bounds
for the variation of the q-th component ∆xq .
Now we transform problem (8.56) into an equivalent one which is easier to put
into a standard LP form.
Firstly, we introduce a new variable y ∈ R2n obtained by stacking a vector
α ∈ Rn (where α = (α1 , . . . , αn )) with the machine-tool settings vector ∆x ∈ Rn ,
as follows
 
α
y= (8.57)
∆x

Secondly, we introduce the column vector e ∈ Rn+m which is given by n 1’s and
m zeros, as follows
 
eT = 1 · · · 1 | 0 · · · 0 (8.58)
156 8. Identification of CNC parameters for micro-geometry optimization

such that by performing the scalar product


" # n
T
  α X
e y = 1···1 | 0···0 = αi (8.59)
∆x i

one obtains the sum of the αi ’s.


If we now require the following inequalities to be satisfied

αi ≥ |sTi ∆x − hi |, for each i = 1, . . . , n (8.60)

which are indeed equivalent to the two sets of constraints

α≥ S ∆x − h
(8.61)
α ≥ −(S ∆x − h)
Pn
and we minimize the sum i αi , the L1 norm of the residuals will be minimized
as well. The demonstration of the equivalence between the previous problems is
found on many Operations Research textbooks, like [72].
By introducing the n−dimensional identity matrix I n we rewrite inequali-
ties (8.61) as

I n α − S ∆x ≥ −h (8.62)
I n α + S ∆x ≥ h (8.63)

and then in matrix form as


" #" # " #
In −S α −h
≥ (8.64)
In S ∆x h

Taking into account possible linear constraints among the parameters as well we
write
" #
  α
O l×n A ≥b (8.65)
∆x

Now, we employ a notation similar to § 8.7.1 and we set


" # " #
M1 m1
M= , m= (8.66)
M2 m2

where
" #
In −S  
M1 = , M 2 = O l×n A (8.67)
In S
8.8 Sequential refinement approach 157

and
" #
−h
m1 = , m2 = b (8.68)
h

with M 1 ∈ R2n×(n+m) is the constraint matrix induced by the L1 norm and


M 2 ∈ Rl×(n+m) accounts for possible linear inequality constraints among the
machine-tool parameters and with m1 ∈ R2n and m2 ∈ Rl . The last step is to
define the lower and upper bounds y min and y max in terms of y which result in
the following vectors
   
−∞ +∞
 |   | 
   
y min =  , y max =   (8.69)
 −∞   +∞ 
∆xmin ∆xmax

The standard form of the LP problems, which is here recalled, is formally identical
to (8.54), even if the same symbols bear now a very different meaning

 min eT y
 y


subject to My≥m (8.70)




y min ≤ y ≤ y max

Again, the solution methods available for the standard LP problem (8.70) are very
efficient and further details can be found in [82] and [27].

8.8 Sequential refinement approach


The various methods presented in the previous sections are implicitly based on the
hypothesis that the solution of the minimization problems (8.35), (8.41) and (8.55)
can be obtained in one single step. This hypothesis is acceptable and the single
step approach provides reasonable results if the difference between the maximum
and the minimum component of required ease-off is sufficiently small, say, less
than 20 µm for a face width of 33 mm. On the other hand, if the amount of
required modifications exceeds the above mentioned value, the new tooth surface,
generated with the updated machine-tool settings, may differ appreciably from the
target surface defined by the ease-off. This discrepancy is to be ascribed mainly to
the highly nonlinear relationships between surface shape and machine-tool settings
involved in the process of gear generation, which are approximated, in the above
mentioned models, by linear ones obtained by performing a Taylor series expansion
about each point on the sampling grid. On one hand, the employment of linear
models is motivated by the convenience of the efficient algorithms available for
158 8. Identification of CNC parameters for micro-geometry optimization

(0)
ζ i , basic (initial) values of the triplet (ξ, θ, φ) for the i-th point
(k)
ζ i , values of the triplet (ξ, θ, φ) for the i-th point at iteration k
(k) (k)
Ξ(k) = (ζ 1 , . . . , ζ n(k) ), set of all the n ζ i ’s at iteration k

x(0) , basic (initial) machine-tool settings

x(k) , machine-tool settings values at iteration k

x(k+1) = x(k) + ∆x(k) , iterative update of the machine-tool settings

Table 8.1: Triplets and settings.

their solution. On the other hand, as supported by many practical experiences,


very often there is the need to prescribe ease-off values in much broader ranges,
say, [−70, +70] µm, again for a gear tooth with face width equals to 33 mm.
In order to accomodate both requirements, a sequential refinement approach
is presented in this section. The basic idea is to adopt a linear model to express
the relationship between surface modifications and variable parameters, update
the tooth geometry with the new machine-tool settings and repeat the process
iteratively in order to reduce the gap between the target and the actual surface
down to a prescribed extent or, at least, to its best approximation allowed by
mathematics. This process is based on the hypothesis that the ease-off surface is
iteratively reduced as the optimization proceeds and therefore the linear approxi-
mation employed at each step gives a reasonable approximation of the relationship
between the surface shape and the machine-tool settings.
An fundamental aspect which is at the base of the previous considerations
regards the ease-off topography. In order for the algorithms to be able to locate
reasonable solutions in the machine parameters which provide, in their turn, a
modified surface complying with the prescriptions, the ease-off itself must be lay
within the reachable or, we may say, machinable space. In other words, the ease-off
topography must be smooth enough to be actually machinable with a typical tool
surface performing 3D smooth motions with respect to the gear blank. This aspect
of the choice of reasonable ease-off modifications is therefore left to the good sense
of the gear designer.
Let us first introduce the following notation, where k = 0, 1, 2, . . . is a counter
of the iterations performed and i = 1, . . . , n is the subscript relative to the i−th
point Pgi on the surface grid.
The sequential refinement approach is based on the procedure that will follow.
In Tab. 8.1-8.4 some useful definitions and recalls are given.
(0) (0)
Let us impose a certain ease-off h(0) , with i-th component hi at point Pgi
8.8 Sequential refinement approach 159

Γg = Γ(0)
g , basic (initial) tooth surface

Γ̃g , target tooth surface


(k)
Γg(k) , approximation of Γ̃g at iteration k (with x(k) and ζ i , i = 1, . . . , n)

Table 8.2: Surface names.

(0)
p(0)
gi = pg (ζ i , x
(0)
), position vector of the i-th point of the grid on surface Γ(0)
g

(0)
mugi(0) = mug (ζ i , x(0) ), unit normal vector to Γ(0)
g at the i-th point

(k)
pg(k)
i
= pg (ζ i , x(k) ), position vector of the i-th point of the grid on surface Γg(k)

Table 8.3: Surface position vectors.

h = h(0) , prescribed ease-off w.r.t. the basic (initial) surface

h(k) = actual ease-off between surface Γg(k) and surface Γ(0)


g ,

measured along the mugi(0) ’s

Table 8.4: Ease-off vectors.


160 8. Identification of CNC parameters for micro-geometry optimization

(0)
of the grid on surface Γg , by requiring the following n equations to be satisfied
(0)
p(0) u (0)
gi + hi m gi = p̃gi → ∆p(0) u (0)
gi = hi mgi , (i = 1, . . . , n) (8.71)

(0) (0) (0) u (0)


where we set ∆pgi = p̃gi − pgi . By linearizing ∆pgi , dot multiplying by mgi
and considering all the n points of the grid, as done in equations from (8.7) to (8.10)
we obtain the following system of linear equations

h(0) ' S(Ξ(0) , x(0) ) ∆x(0) (8.72)

Now we can solve the problem in ∆x(0) by employing one of the techniques pre-
sented in the previous sections. In every case we end up with a solution ∆x(0)
which we use to obtain the updated machine-tool setting vector x(1) as follows

x(1) = x(0) + ∆x(0) (8.73)

We are now ready to set up an iterative process similar to the Newton-Raphson


algorithm employed for the solution of a system of nonlinear equations. First of
(1)
all, a n-point grid is sampled on the new gear surface Γg , generated by setting
(1)
x = x . In particular, the points on the updated grid are computed intersecting
(1) u (0)
the new surface Γg with n straight lines directed like mgi and passing through
(0) (0)
the base points of Γg with position vectors pgi . Therefore, for each of the n
points one has to solve the following nonlinear system of four equations in the four
(1) (1)
unknowns ζ i and hi
( (1) (0) (1) u (0)
pg (ζ i , x(1) ) = pgi + hi mgi
(1)
(8.74)
f (ζ i , x(1) ) = 0

(1) (1)
Vector h(1) = (h1 , . . . , hn ) is the actual ease-off obtained with the new machine-
tool settings x(1) . Comparing, in some norm, the deviation between the target
h(0) and the actual ease-off h(1) it is possible to estimate if the solution is accept-
able or if more steps are necessary to obtain the desired accuracy. As previously
mentioned, if the target ease-off h(0) is such that

max(hi ) − min(hi ) ≥ 25 µm (8.75)

for a tooth with face width of 33 mm, the solution obtained in only one step is
usually not accurate. In such cases a new Taylor series expansion with respect
(1)
to points Pgi can be performed. The equation one obtains considering the 1-st
iteration is
(0) (1) (1) (1) (1)
hi − hi ' mugi(0) · pg ,ζ (ζ i , x(1) ) ∆ζ i + mugi(0) · pg ,x (ζ i , x(1) ) ∆x(1)
(8.76)
8.8 Sequential refinement approach 161

where now the first term on the right hand side does not vanish since, in general,
(1) (1) (1)
the tangent plane to Γg at Pgi , which is spanned by pg ,ζ (ζ i , x(1) ), is not
u (0)
orthogonal to mgi . Recalling the position

sT (ζ i , x) = mugi(0) · pg ,x (ζ i , x), (8.77)

where sT ∈ R1×m and setting

z T (ζ i , x) = mugi(0) · pg ,ζ (ζ i , x), (8.78)

where z T ∈ R1×3 , we can rewrite (8.76) as


(0) (1) (1) (1) (1)
hi − hi ' z T (ζ i , x(1) ) ∆ζ i + sT (ζ i , x(1) ) ∆x(1) (8.79)

From (8.79) it is easy to recognize that, at each point i, the triplet variation
(1) (1) (1) (1)
∆ζ i = (∆ξi , ∆θi , ∆φi ) has to be evaluated besides the variation ∆x(1) : the
dimensionality of the problem is therefore increased by 3n. For convenience we
define
(k) T (k) (k) T (k)
si = sT (ζ i , x(k) ) and z i = z T (ζ i , x(k) ) (8.80)

If we now consider equation (8.79) for all the n points on the grid and define
(k) T (k) T
matrix S (k) = (s1 , . . . , sn ), with S (k) ∈ Rn×m , where its generic entry has
the following expression
 (k)  (k)
S ij
= mugi(0) · pg ,xj (ζ i , x(k) ), (8.81)

(k) T (k) T
and we define matrix Z (k) = (z 1 , . . . , zn ), with Z (k) ∈ Rn×3n , where its
generic entry has expression
 (k)  (k)
Z ij
= mugi(0) · pg ,ζj (ζ i , x(k) ), (8.82)

we get, for the first step, the following system of linear equations

h(0) − h(1) ' Z (1) ∆Ξ(1) + S (1) ∆x(1) (8.83)

It worth remarking that this system is now underdetermined since we have n


equations in m + 3 n unknowns, being ∆x(1) ∈ Rm the variations of the machines
parameters and ∆Ξ ∈ R3n the variations in the triplets for all the n points of
the grid. Moreover, while at the zero step the parametrization did not enter the
solution for ∆x(0) (because of the orthogonality condition between the normal
u (0) (0)
vectors mgi and the vectors pg ,ζ (ζ i , x(0) ) spanning the tangent planes to
(0)
Γg ), now the change in the parametrization influence also ∆x(1) .
162 8. Identification of CNC parameters for micro-geometry optimization

8.8.1 A first procedure


A first way to drastically simplify the problem could be to force the correspondent
(k)
points of the successive grids on the Γg ’s to keep the values of the original triplets
as the iteration proceeds (k = 0, 1, 2, . . .). This hypothesis is indeed equivalent to
setting ∆Ξ(k) ≡ 0, ∀ k. In this case the process of resampling the points on the
updated surface can be avoided. By setting
(0) (k) (0)
h̄i = mugi(0) · pg (ζ i , x(0) ), and h̄i = mugi(0) · pg (ζ i , x(k) ) (8.84)
(k) (k) (k)
and h̄ = (h̄1 , . . . , h̄n ), h̄ = (h̄1 , . . . , h̄n ), the sequential refinement method (I)
can be formulated as the following procedure:
Step 1 Given the basic x(0) , determine the first variation ∆x(0) by solving
h(0) ' S(Ξ(0) , x(0) ) ∆x(0)

Step 2 Set k = 1;
Step 3 Set x(k) = x(k−1) + ∆x(k−1) ;
(0) (k)
Step 4 If ||h̄ + h(0) − h̄ ||∞ ≤ εh
(k)
then EXIT (found solution x(k) ; Γg converged to Γ̃g within tolerance εh )
else CONTINUE
Step 6 If k ≥ kmax
then EXIT (found solution x(k) ; upper iteration limit kmax reached)
else CONTINUE
Step 7 Determine ∆x(k) by solving
(0) (k)
h̄ + h(0) − h̄ ' S(Ξ(0) , x(k) ) ∆x(k)

Step 8 If ||∆x(k) ||∞ ≤ εx


then EXIT (found solution x(k) ; lower limit εx reached)
else CONTINUE
Step 9 Set k = k + 1 and GOTO Step 3
Some checks on the maximum number of iterations allowed kmax and on the min-
imum amount of variation εx compatible with the sensitivity of the machine are
performed to avoid infinite loops.
It is worth noting that in the previous procedure no re-sampling is performed
on the actual surface obtained with the updated machine-tool settings. Therefore,
(k)
no warranty can be given that the surface Γg will converge to the actual target
surface Γ̃g as k = 1, 2, . . ..
8.8 Sequential refinement approach 163

8.8.2 A second procedure


Another way to keep things simple but ensure a higher level of confidence that the
procedure will seek a good solution is to neglect the contribution of ∆Ξ(k) only in
the first term of the right hand side of (8.83) but take into accounts its variation
when updating the sensitivity matrix S. This second version of the sequential
refinement approach is described by the following steps:

Step 1 Given the basic x(0) , determine the first variation ∆x(0) by solving

h(0) ' S(Ξ(0) , x(0) ) ∆x(0)

Step 2 Set k = 1;

Step 3 Set x(k) = x(k−1) + ∆x(k−1) ;

(k) (k)
Step 4 For i = 1, . . . , n, determine ζ i and hi by solving
( (k) (0) (k) u (0)
pg (ζ i , x(k) ) = pgi + hi mgi
(k)
f (ζ i , x(k) ) = 0

Step 5 If ||h(0) − h(k) ||∞ ≤ εh


(k)
then EXIT (found solution x(k) ; Γg converged to Γ̃g within tolerance εh )
else CONTINUE

Step 6 If k ≥ kmax
then EXIT (found solution x(k) ; upper iteration limit kmax reached)
else CONTINUE

Step 7 Determine ∆x(k) by solving

h(0) − h(k) ' S(Ξ(k) , x(k) ) ∆x(k)

Step 8 If ||∆x(k) ||∞ ≤ εx


then EXIT (found solution x(k) ; lower limit εx reached)
else CONTINUE

Step 9 Set k = k + 1 and GOTO Step 3


164 8. Identification of CNC parameters for micro-geometry optimization

8.8.3 A third procedure


If we drop the restrictive hypothesis ∆Ξ(k) ≡ 0 and let the point triplets vary
freely, it may be reasonable to require some new conditions to balance the number
of equations and unknowns. The first one requires that the displacement from
(k) (k+1)
point Pgi to point Pgi , as a result of the variation ∆x(k) , is directed along
u (0)
mgi only. This is indeed equivalent to impose
 
(k) (k) (k) (k) (k) (k) (0)
pg ,ζ (ζ i , x ) ∆ζ i + pg ,x (ζ i , x ) ∆x · pg ,ξ (ζ i , x(0) ) = 0 (8.85)
 
(k) (k) (k) (k) (k) (k) (0)
pg ,ζ (ζ i , x ) ∆ζ i + pg ,x (ζ i , x ) ∆x · pg ,θ (ζ i , x(0) ) = 0 (8.86)

(k+1) (k)
which requires the displacement vector pgi − pgi not to have components on
(0) (0)
the tangent plane to Γg at Pgi .
(k)
Of course, for the solution ∆x(k) , ∆ζ i to correspond to a point of the envelope
(k+1)
surface Γg , the equation of meshing in the updated condition must be satisfied,
that is
(k) (k)
f (ζ i + ∆ζ i , x(k) + ∆x(k) ) = 0 (8.87)
Since we can expect small variations in the values of the triplets and the machine
settings from iteration k to k + 1, it can be reasonable to employ the linearization
of (8.87) thus requiring
(k) (k) (k) (k)
f (ζ i , x(k) ) + f,ζ (ζ i , x(k) ) ∆ζ i + f,x (ζ i , x(k) ) ∆x(k) = 0 (8.88)
If we extend the above mentioned conditions (8.85), (8.86) and (8.88) to all the n
points of the grid beside the original problem (8.83), we end up with 4n equations in
m + 3n unknowns. The whole sistem of equations if therefore still overdetermined,
with the same imbalance of n − m equations like in problem (8.72).
A procedure describing the sequential refinement approach with no direct con-
straints on the triplets is the following:
Step 1 Given the basic x(0) , determine the first variation ∆x(0) by solving
h(0) ' S(Ξ(0) , x(0) ) ∆x(0)

Step 2 Set k = 1;
Step 3 Set x(k) = x(k−1) + ∆x(k−1) ;
(k) (k)
Step 4 For i = 1, . . . , n, determine ζ i and hi by solving
( (k) (0) (k) u (0)
pg (ζ i , x(k) ) = pgi + hi mgi
(k)
f (ζ i , x(k) ) = 0
8.9 Numerical example 165

Step 5 If ||h(0) − h(k) ||∞ ≤ εh


(k)
then EXIT (found solution x(k) ; Γg converged to Γ̃g within tolerance εh )
else CONTINUE
Step 6 If k ≥ kmax
then EXIT (found solution x(k) ; upper iteration limit kmax reached)
else CONTINUE
(k)
Step 7 For i = 1, . . . , n determine ∆ζ i and ∆x(k) by solving
 (0) (k) (k) (k) (k)

hi − hi ' s(ζ i , x(k) ) ∆ζ i + z(ζ i , x(k) ) ∆x(k)



 


 (k) (k) (k) (k) (k) (k) (0)
 pg ,ζ (ζ i , x ) ∆ζ i + pg ,x (ζ i , x ) ∆x
 · pg ,ξ (ζ i , x(0) ) = 0
 
 (k) (k) (k) (k) (k) (k) (0)



 pg ,ζ (ζ i , x ) ∆ζ i + pg ,x (ζ i , x ) ∆x · pg ,θ (ζ i , x(0) ) = 0



 (k) (k) (k) (k)
f (ζ i , x(k) ) + f,ζ (ζ i , x(k) ) ∆ζ i + f,x (ζ i , x(k) ) ∆x(k) = 0

Step 8 If ||∆x(k) ||∞ ≤ εx


then EXIT (found solution x(k) ; lower limit εx reached)
else CONTINUE
Step 9 Set k = k + 1 and GOTO Step 3
Of course, to solve the overdetermined system of equations one of the methods
presented in this Chapter can be adopted. Instead, to solve the nonlinear system
of equations at Step 4 the Newton-Raphson algorithm is usually employed.

8.9 Numerical example


Some of the methods previously discussed were implemented in a computer pro-
gram written in Mathematica language and applied to the optimization of the
tooth surface of a spiral bevel pinion. The numerical results obtained with the
above mentioned program, are the subject of the present section. The pinion
optimized is a component of a gearbox for aerospace applications.
Given the basic machine-tool settings x(0) and the target ease-off modifications,
different corrective settings x, computed by the above mentioned optimization
techniques, are employed in the numerical simulation of the grinding process of
the pinion. To assess the reliability of the results, the actual surfaces obtained
by generation with the modified settings are compared with the target surface.
The results obtained are discussed and some numerical aspects of the problem are
investigated.
166 8. Identification of CNC parameters for micro-geometry optimization

8.9.1 Design variables


For a detailed description of the machine-tool parameters employed in the opti-
mization procedures, we refer to the presentation of the milling tool geometry and
the CNC machine kinematics given in Section 4.8. However, a schematic layout
of a typical spiral bevel and hypoid gear machine is recalled in Fig. 8.2

head-cutter

machine
frame
zo
O e(f)
kc
cutter
axis
Se
so S r(f)
j

jc i
q k gear
blank
f
cradle Oa blank
axis axis
a j(f)
DE 
(f)
g DX
g


(f) 

DX 
(f) s (f)


Ob (f) b (f)

line of shortest
distance

Figure 8.2: Parameters involved in the milling process.

In this section, the machine-tool parameters considered as design variables x


are in number of 10. Two of them, the cutter point radius Rp and the pressure
angle αp , influence the geometry of the tool. The other eight, which contribute
to the definition of the motion of the tool with respect to the workpiece, are the
machine center to back ∆XD , the blank offset ∆XM , the radial distance Sr , the
cradle angle q1 , the ratio of roll m0 , the first and second order coefficients of
the modified roll 2C and 6D and, finally, the tilt σ0 of the head-cutter. All the
previous parameters are treated as design variables and vector x can be represented
as follows

x = (Rp , αp , ∆XD , ∆EM , Sr , q1 , m0 , 2C, 6D, σ0 ) (8.89)


8.9 Numerical example 167

In the initial design, the numerical values of the design variables are

x(0) = (79.013, 0.3040, 2.7183, 5.0800, 72.1400,


(8.90)
1.0426, 1.7066, 0.0226, −0.0329, 0.0000)

In those procedures where constraints on the machine setting perturbations


are allowed, the boundaries where defined according to the following strategy: for
angular variables like αp , q1 and σ0 the admissible range was defined as [−0.1, 0.1]
rad; for distances and linear displacements like Rp , ∆XD , ∆EM and Sr the ad-
missible range was defined as [−10, +10] mm; for the polynomial coefficients of
the modified roll, the boundaries were defined as [−100, +100] rad/radp , where p
is the polynomial order.

8.9.2 Target ease-off


The performances of some of the proposed optimization techniques where tested
with respect to an ease-off modification carrying out a lead crowning of the concave
side of the pinion tooth surface. Therefore, the parameters described hereafter are
referred to this side of the tooth, i.e., the outside blade of the grinding wheel.
The computational grid laid on the tooth concave surface is a typical 5 × 9 grid,
i.e., 5 profilewise by 9 lengthwise points: this is normally sufficient for imposing
1st and 2nd order global surface modifications [24]. The points on the grid are
usually on the active flank of the tooth and the mid point of the grid is the tooth
measuring point (TMP), i.e., the point where the tooth thickness is measured.
However, no particular problem arises in prescribing modifications even on the
tooth fillet, especially if the whole tool surface is a single piecewise defined surface,

Tooth measuring point

Profile 1 tip Profile 9

5
4 45
3 44
2 43
1 42
41
toe heel

root

Figure 8.3: Computational grid.


168 8. Identification of CNC parameters for micro-geometry optimization

  

'

) 1

&

%
m ( ,

" ( 1


!

  


0
,

 

(
.

  
+

 


   

  

 

  

Figure 8.4: Target ease-off: lengthwise crowning.

as in (4.119). A schematic representation of the computational grid is reported in


Fig. 8.3.
The lengthwise crowning modification, used as target ease-off in the numerical
tests, is shown in Fig. 8.4. Since the unit normal vectors are directed inward the
tooth body, the prescribed ease-off represents a lengthwise crowning with material
removal at the edges of the grid toward the toe and the heel. In this test the
maximum ease-off amounts to 18 µm.

8.9.3 First results


The results obtained with six of the algorithms proposed in this chapter are listed
in Table 8.5. The methods tested are: LS, Least Squares solution computed
employing the pseudoinverse of the sensitivity matrix S obtained, in its turn, with
the SVD algorithm, where all the ten singular values σi (i=1,. . . , 10), listed in
decreasing order, are retained (see 8.30); LS (c), Least Squares solution computed
employing the pseudoinverse matrix with the SVD and chopping all the singular
values σi < σmin of S (see 8.32), where the minimum singular value was defined
by setting a tolerance t = 1 · 10−6 with respect to the maximum singular value σ1 ,
as follows

σmin = t σ1 (8.91)
8.9 Numerical example 169

  
+

*
m

- 5

'

&

, 0

! $

, 5

"


!

  


0

 


  
/

 


   

  

 

  

Figure 8.5: Target and actual ease-off with LS(c) solution.

N N K
a

`
m

c k

V ]

b f

S V

W Z

b k

S
X

P
W

O O Q

U
f

R S

@ A B

> ?

C D

C
9 < =
e

7 ;

C F G
6 7 8 9 :

A H

K L M

I J

K N O

Figure 8.6: Target and actual ease-off with QP solution.


170 8. Identification of CNC parameters for micro-geometry optimization

Par. Initial LS LS(c) QP SLS(I) L∞ L1


Rp 79.0130 -1.52e+10 79.8486 79.6302 80.0478 82.0250 82.0317
αp 0.3040 -3.34e+8 0.2782 0.2725 0.2813 0.3254 0.3261
∆XD 79.0130 2.11e+10 2.1698 2.4849 1.9362 -0.8632 -0.8527
∆EM 5.0800 -7.06e+10 4.9861 3.9475 6.0291 15.0800 15.0800
Sr 72.1400 6.25e+10 72.4216 73.3556 71.5318 63.4774 63.4799
q1 1.0426 2.66e+8 1.0715 1.0755 1.0669 1.0336 1.0333
m0 1.7066 1.16e+9 1.6892 1.7068 1.6718 1.5223 1.5227
2C 0.0226 -3.18e+8 0.0286 0.0238 0.0329 0.0741 0.0741
6D -0.0329 -3.17e+9 -0.0717 -0.1180 -0.0270 0.3821 0.3821
σ0 0.0000 -5.58e+8 -0.0286 -0.0372 -0.0220 0.0505 0.0512

Table 8.5: Updated machine-tool settings.

QP, least squares solution computed by employing the Quadratic Programming


algorithm and imposing bound constraints to the design variables (see § 8.6);
SLS(I), first version of the sequential refinement approach where the subproblem is
solved by the algorithm of subset selection with SVD (see 8.5.1); L∞ , minimization
of the infinity-norm, which results in a Linear Programming problem (see 8.7.1);
L1 , minimization of the 1-norm, which results in a Linear Programming problem
(see 8.7.2).
From the analysis of the third column of Tab. 8.5, it is clear that a direct
approach of the least squares method (LS) without preconditioning the sensitivity
matrix S provides a unacceptable solution, which is caused by numerical insta-
bility. This because some of the design variables have a similar effect on the
modifications of the tooth surface: this results in a highly ill-conditioned sensi-
tivity matrix. This is a crucial problem though, since it is not an easy task to
distinguish linear dependencies among the machine parameters.
Encouraging results are obtained by chopping the smallest singular values of
S in the calculation of its pseudoinverse. The relative tolerance of 10−6 proved
to be a good choice. The comparison of the target ease-off and the actual ease-off
obtained with this method is illustrated in Fig. 8.5. The accordance of the results
is very good, with a maximum error of 1.42 µm, equal to the 7.8% of the maximum
modification prescribed. Nevertheless, how to set the tolerance value, which has
a key role in the success of the method, is still an open question. Unsatisfactory
results were obtained with both a tolerance of 10−5 , probably because too many
pieces of information were discarded from S, and 10−7 , maybe because, in this
case, too much numerical noise was retained in the calculation.
The solution obtained by the QP algorithm did not prove as good as the
previous one. The results of the comparison are shown in Fig. 8.6. The maximum
error is 12 µm, equal to the 66% of the maximum modification.
Probably, the algorithm ‘quadprog’ of MATLAB is not able to appropriately
8.9 Numerical example 171

  
+

*
m

)
. 5

'
- 0

&

- 5

! $
, 0


"

, 5 
!

  

 


/

  

 





   
/

  

 

  

Figure 8.7: Target and actual ease-off with SLS(I) solution.

N N K
a

`
m

e k
_

d f

V ]

d k

S V

[
c f

W Z c k

b f
S
X

P
W

O O Q

b k

U
j

R S
i

@ A B

> ?

C D

e
7

C
9 < =

7 ;

E
9

C F G
e
6 7 8 9 :

A H

K L M

I J

K N O

Figure 8.8: Target and actual ease-off with L1 solution.


172 8. Identification of CNC parameters for micro-geometry optimization

find a remedy to the ill-conditioned sensitivity matrix S. It is worth noting,


though, that the boundaries of the admissible range were not reached by the
solution.
Again, due to the high numerical noise in S, the solution with the sequential
least squares method that employed the subset selection algorithm to precondition
the matrix S proved to be not satisfactory. The comparison of the actual ease-off
with the prescribed one shows, as depicted in Fig. 8.7, sound errors that amount
to 60% in the worst case.
Regarding the method of minimizing the L1 or the L∞ norm, which results in
both cases in solving of a Linear Programming problem, the results obtained are
shown in Fig. 8.8 and 8.9. With the first method, the maximum error between
the target and the actual ease-off is 100% of the maximum modification. With the
second algorithm, errors that amount to the 50% of the maximum target value
are reported. For both methods, as clear from the boldface values in Tab. 8.5, the
upper bound for the blank offset parameter ∆XM was reached. In this case, the
bounds helped to keep the solution, that was being shot away by the numerical
instability introduced by S, within the admissible range. This approach, though,
did not prove to be satisfactory in case of redundancy among design variables, as
in the present test.

  

*
m

. 5

'

-
0

&

- 5

$


,
0


"


!

, 5
  

 

  


/

 


   

  

 

  

Figure 8.9: Target and actual ease-off with L∞ solution.


8.10 Conclusions 173

8.10 Conclusions
In this chapter different solutions to the problem of identifying practical machine-
tool settings corrections corresponding to a prescribed ease-off have been proposed.
Numerical tests revealed that trying to cure the ill-conditioned problem (that
arises when redundancy among machine parameters exists) by imposing bound
constraints to the design variables, as proposed in [37], is not adequate since does
not solve the cause of the problems. Better solutions may be found by focusing
on the sensitivity matrix (as shown by the SVD solution with preconditioning)
and, therefore, trying to discern the underlying dependencies among the machine
parameters and their redundancies with the purpose of avoiding them. The un-
derstanding of these topics is of paramount importance for the future development
of the gear theory field and will be part of our future research.
Chapter 9

Conclusions

In the present dissertation a new approach to the theory of gearing has been
formulated. The main difference with respect to the classical theory of gearing
is that there is no need of an early introduction of reference systems. In fact, as
demonstrated in the first chapters, they play no role in the theoretical treatment
and, therefore, their introduction can be postponed till the very end, when actual
calculations have to be performed. The final outcome is a more compact and,
hopefully, clearer formulation of the whole theory of gearing.
To show that the proposed methodology can be applied with success in every
facets of the theory, a fresh derivation of many classical results is provided. Also
some investigations on modern topics like curvature analysis and identification of
CNC machine settings capable of free-form surface corrections of the tooth are
presented within the framework of the invariant approach.
Still many studies have to be performed, especially on the topic of global surface
synthesis: it is believed, though, that the invariant approach can represent a new
tool to gain a better understanding and a deeper insight of the problem.
Bibliography

[1] J. Achtmann and G. Bar. Optimized bearing ellipses of hypoid gears. ASME
Journal of Mechanical Design, 125:739–745, 2003. 4
[2] Jorge Angeles. Fundamentals of robotic mechanical systems: theory, methods
and algorithms. Springer, 2nd edition, 2003. 51
[3] J. Argyris, M. De Donno, and F. L. Litvin. Computer program in Visual
Basic language for simulation of meshing and contact of gear drives and its
application for design of worm gear drive. Computer Methods in Applied
Mechanics and Engineering, 189:595–612, 2000. 3
[4] J. Argyris, A. Fuentes, and F.L. Litvin. Computerized integrated approach
for design and stress analysis of spiral bevel gears. Computer Methods in
Applied Mechanics and Engineering, 191:1057–1095, 2002. 2, 29, 30, 31, 33,
83, 131
[5] J. Argyris, F. L. Litvin, Q. Lian, and S. A. Langutin. Determination of
envelope to family of planar parametric curves and envelope singularities.
Computer Methods in Applied Mechanics and Engineering, 175:175–187, 1999.
2
[6] J. Argyris, F. L. Litvin, Aoyong Peng, and H. J. Stadtfeld. Axes of meshing
and their application in theory of gearing. Computer Methods in Applied
Mechanics and Engineering, 163:293–310, 1998. 2
[7] M. Beghini and C. Santus. Analysis of the contact between cubic profiles.
International Journal of Mechanical Sciences, 46:609–621, 2004. 119, 137
[8] C. H. Chen, S. T. Chiou, Z. H. Fong, C. K. Lee, and C. H. Chen. Mathematical
model of curvature analysis for conjugate surfaces with generalized motion in
three dimensions. Journal of Mechanical Engineering Science, 215(4):487–
502, 2001. 2, 49, 72, 91
[9] N. Chen. Curvatures and sliding ratios of conjugate surfaces. ASME Journal
of Mechanical Design, 120:126–132, 1998. 2, 38, 45, 46, 71, 74, 91, 92, 94, 97,
99, 110
178 Bibliography

[10] Hui Cheng and K. C. Gupta. An historical note on finite rotations. ASME
Journal of Applied Mechanics, 56:139–145, 1989. 8
[11] F. Di Puccio, M. Gabiccini, and M. Guiggiani. Alternative formulation of the
theory of gearing. Mechanism and Machine Theory, 40(5):613–637, 2005. 36,
39, 42, 43, 46, 49, 50, 52, 58, 59, 61, 72, 77, 79, 80, 81, 82, 92, 96, 97, 99, 100,
102, 111, 112, 119, 124, 129
[12] F. Di Puccio, M. Gabiccini, and M. Guiggiani. Generation and curvature
analysis of conjugate surfaces via a new approach. Mechanism and Machine
Theory, 41:382–404, 2006. 49, 57, 91, 92, 111, 113
[13] Manfredo Perdigao Do Carmo. Differential Geometry of Curves and Surfaces.
Prentice Hall, Englewood Cliffs, 1976. 38, 74, 91, 94
[14] David B. Dooner. On the three laws of gearing. ASME Journal of Mechanical
Design, 124:733–744, 2002. 1, 3, 72, 92
[15] David B. Dooner and Ali A. Seireg. The Kinematic Geometry of Gearing.
Wiley Interscience, New York, 1995. 1, 3, 72
[16] Leonard Euler. Nova methodus motum corporum rigidorum determinandi.
Novi Commentari Academiae Scientiarum Imperialis Petropolitanae, 20:208–
238, 1775. 1, 7, 8
[17] Qi Fan and Ron S. Dafoe. Development of bevel gear face hobbing simulation
and software. 1:485–504, 2005. 6
[18] P.-H. Feng, F. L. Litvin, D. P. Townsend, and R. F. Handschuh. Determi-
nation of principal curvatures and contact ellipse for profile crowned helical
gears. ASME Journal of Mechanical Design, 121:107–111, 1999. 71
[19] Zhang-Hua Fong. Mathematical model of universal generator with supple-
mental kinematic flank correction motions. ASME Journal of Mechanical
Design, 122:136–142, 2000. 49, 65
[20] Zhang-Hua Fong and Chung-Biau Tsay. Kinematical optimization of spiral
bevel gears. ASME Journal of Mechanical Design, 114:498–505, 1992. 4
[21] Marco Gabiccini, Massimo Guiggiani, and Francesca Di Puccio. New inves-
tigation on the geometry of the contact in gear generation. In DETC’03,
number PTG-48087, 2003. 111
[22] Gene H. Golub, V. Klema, and G. W. Stewart. Rank degeneracy and least
squares problems. Technical report tr-456, Department of Computer Science,
University of Maryland, College Park, MD, 1976. 149
[23] Gene H. Golub and Charled F. Van Loan. Matrix Computations. The Johns
Hopkins University Press, Baltimore, 1996. 145, 147, 149
Bibliography 179

[24] Claude Gosselin. Identification of the machine settings of real hypoid gear
tooth surface. ASME Journal of Mechanical Design, 120:429–440, 1998. 4,
167

[25] A. Griewank. Evaluating derivatives, principles and techniques of algorithmic


differentiation. In Frontiers in applied mathematics. SIAM, 2000. 3

[26] Joachim Grill. A generalized kinematic for calculating and evaluating surface
geometry of gears and cutting tools. In 6th International Power Transmission
and Gearing Conference. ASME, 1992. 3, 49

[27] George Hadley. Linear Programming. Addison Wesley, New York, 1962. 152,
154, 157

[28] D. Hestenes, H. Li, and A. Rockwood. New algebraic tools for classical
geometry. In G. Sommer, editor, Geometric computing with Clifford alge-
bra. Springer, 1999. 3

[29] Davis Hestenes. New foundations for classical mechanics. Kluwer Academic
Publishers, second edition, 1999. 3

[30] Norio Ito and Koichi Takahashi. Differential geometrical conditions of hypoid
gears with conjugate tooth surfaces. ASME Journal of Mechanical Design,
122:323–330, 2000. 1, 72, 92

[31] Theodore J. Krenzer. Computer aided inspection of bevel and hypoid gears.
In SAE Paper 831266. SAE, 1983. 3

[32] Theodore J. Krenzer. Computer aided corrective machine settings for man-
ufacturing bevel and hypoid gear sets. In AGMA Paper 84FTM4. AGMA,
1984. 4

[33] Erwin Kreyszig. Differential Geometry. Dover Publications, New York, 1991.
37, 38, 53, 71, 73, 75, 78, 79, 91, 93, 97, 121

[34] Màrk Lelkes, Jànos Màrialigeti, and Daniel Play. Numerical determination of
cutting parameters for the control of Klingelnberg spiral bevel gear geometry.
ASME Journal of Mechanical Design, 124:761–771, 2002. 49

[35] Q. Lian and Ron S. Dafoe. Using advanced tca theory, geometry modifica-
tions, manufacturing realization to localize bearing contact and reduce cylin-
drical gear noise. 1:737–755, 2005. 6

[36] Chung-Yunn Lin, Chung-Biau Tsay, and Zhang-Hua Fong. Computer-aided


manufacturing of spiral bevel and hypoid gears with minimum surface devia-
tion. Mechanism and Machine Theory, 33:785–803, 1998. 4, 148
180 Bibliography

[37] Chung-Yunn Lin, Chung-Biau Tsay, and Zhang-Hua Fong. Computer-aided


manufacturing of spiral bevel and hypoid gears by applying optimization tech-
niques. Journal of Materials Processing Technology, 114:22–35, 2001. 4, 148,
173

[38] F. L. Litvin. The curvature relations in toothed surfaces of a space meshing


(in german). Z. Angew. Mathematik und Mechanik, 49:685–690, 1969. 2

[39] F. L. Litvin, N. X. Chen, and J. S. Chen. Computerized determination of cur-


vature relations and contact ellipse for conjugate surfaces. Computer Methods
in Applied Mechanics and Engineering, 125:151–170, 1995. 2, 71

[40] F. L. Litvin, M. De Donno, A. Peng, A. Vorontsov, and R.F. Handschuh.


Integrated computer program for simulation of meshing and contact of gear
drives. Computer Methods in Applied Mechanics and Engineering, 181:71–85,
2000. 3

[41] F. L. Litvin, A. Egelja, J. Tan, and G. Heath. Computerized design, gener-


ation and simulation of meshing of orthogonal offset face-gear drive with a
spur involute pinion with localized bearing contact. Mechanism and Machine
Theory, 33:87–102, 1998. 2

[42] F. L. Litvin, R. N. Goldrich, J. J. Coy, and E. V. Zaretsky. Precision of


spiral-bevel gears. ASME Journal of Mechanical Design, 105:311–316, 1983.
3

[43] F. L. Litvin, C. Kuan, J.C. Wang, R.F. Handschuh, J. Masseth, and


N. Maruyama. Minimization of deviations of gear real tooth surfaces by co-
ordinate measurements. ASME Journal of Mechanical Design, 115:995–1001,
1993. 4

[44] F. L. Litvin, Jian Lu, D. P. Townsend, and M. Howkins. Computerized


simulation of meshing of conventional helical involute gears and modification
of geometry. Mechanism and Machine Theory, 34:123–147, 1999. 2

[45] F. L. Litvin, Aoyong Peng, and Anngwo Wang. Limitation of gear tooth
surfaces by envelopes to contact lines and edge of regression. Mechanism and
Machine Theory, 34:889–902, 1999. 1

[46] F. L. Litvin and I. H. Seol. Computerized determination of gear tooth surface


as envelope to two parameter family of surfaces. Computer Methods in Applied
Mechanics and Engineering, 138:213–225, 1996. 2

[47] F. L. Litvin, Y. Zhang, J. Kieffer, and R. F. Handschuh. Identification and


minimization of deviations of real gear tooth surfaces. ASME Journal of
Mechanical Design, 113:55–62, 1991. 4
Bibliography 181

[48] Faydor L. Litvin. Theory of Gearing. NASA Reference Publication 1212,


1989. 1, 2, 14, 17, 18, 26, 43, 45, 60, 71, 77, 79, 91, 113, 124

[49] Faydor L. Litvin. Gear Geometry and Applied Theory. PTR Prentice Hall,
Englewood Cliffs, 1994. 1, 2, 14, 15, 17, 44, 45, 60, 71

[50] Faydor L. Litvin. Development of Gear Technology and Theory of Gearing.


NASA Reference Publication 1406, 1997. 1, 2, 14, 19, 20, 22, 25, 28, 60

[51] Faydor L. Litvin. Gear Geometry and Applied Theory. PTR Prentice Hall,
Englewood Cliffs, 2004. 2, 91, 98, 100, 101, 103, 105, 106, 107

[52] Faydor L. Litvin, M. De Donno, Qiming Lian, and S. A. Lagutin. Alternative


approach for determination of singularity of envelope to a family of parametric
surfaces. Computer Methods in Applied Mechanics and Engineering, 167:153–
165, 1998. 2, 22

[53] Faydor L. Litvin, A. M. Egelja, and M. De Donno. Computerized deter-


mination of singularities and envelopes to families of contact lines on gear
tooth surfaces. Computer Methods in Applied Mechanics and Engineering,
158:23–34, 1998. 1, 2, 25, 26

[54] Faydor L. Litvin, Pin-Hao Feng, and Sergei A. Lagutin. Computerized genera-
tion and simulation of meshing and contact of new type of Novikov-Wildhaber
helical gears. Technical report, NASA/CR–2000–209415, 2000. 2

[55] Faydor L. Litvin, Alfonso Fuentes, J. Matthew Hawkins, and Robert F. Hand-
schuh. Design, generation and tooth contact analysis (TCA) of asymmetric
face gear drive with modified geometry. Technical report, NASA/TM-2001–
210614, 2001. 2

[56] Faydor L. Litvin, Qiming Lian, and Alexander L. Kapelevich. Asymmetric


modified spur gear drives: reduction of noise, localization of contact, simula-
tion of meshing and stress analysis. Computer Methods in Applied Mechanics
and Engineering, 188:363–390, 2000. 2

[57] Faydor L. Litvin and Jian Lu. New methods for improved double circular-arc
helical gears. Technical report, NASA/CR–4771, 1997. 2

[58] F.L. Litvin, G. Argentieri, M. De Donno, and M. Hawkins. Computerized


design, generation and simulation of meshing and contact of face worm-gear
drives. Computer Methods in Applied Mechanics and Engineering, 189:785–
801, 2000. 2

[59] F.L. Litvin, A. Demenego, and D. Vecchiato. Formation by branches of en-


velope to parametric families of surfaces and curves. Computer Methods in
Applied Mechanics and Engineering, 190:4587–4608, 2001. 2
182 Bibliography

[60] F.L. Litvin, M. De Donno, A. Peng, A. Vorontsov, and R.F. Handschuh.


Enhanced computer aided simulation of meshing and contact with application
for spiral bevel gear drives. Technical report, NASA/TM-1999–209438, 1999.
2

[61] F.L. Litvin, Qi Fan, A. Demenego, D. Vecchiato, R. F. Handschuh, and T. M.


Sep. Computerized generation and simulation of meshing of modified spur
and helical gears manufactured by shaving. Computer Methods in Applied
Mechanics and Engineering, 190:5037–5055, 2001. 2

[62] F.L. Litvin, A. Fuentes, Qi Fan, and R. F. Handschuh. Computerized design,


simulation of meshing, and contact and stress analysis of face-milled formate
spiral bevel gears. Mechanism and Machine Theory, 2002. 2

[63] F.L. Litvin, A. Fuentes, and M. Hawkins. Design, generation and TCA of
new type of asymmetric face-gear drive with modified geometry. Computer
Methods in Applied Mechanics and Engineering, 190:5837–5865, 2001. 2

[64] F.L. Litvin, A. Fuentes, C. Zanzi, M. Pontiggia, and R.F. Handschuh. Face-
gear drive with spur involute pinion: geometry, generation by a worm,
stress analysis. Computer Methods in Applied Mechanics and Engineering,
191:2785–2813, 2002. 2

[65] F.L. Litvin, G. I. Sheveleva, D. Vecchiato, I. Gonzalez-Perez, and A. Fuentes.


Modified approach for tooth contact analysis of gear drives and automatic
determination of guess values. Computer Methods in Applied Mechanics and
Engineering, 2004. 3

[66] F.L. Litvin, D. Vecchiato, A. Fuentes, and I. Gonzalez-Perez. Automatic de-


termination of guess values for simulation of meshing of gear drives. Computer
Methods in Applied Mechanics and Engineering, 193:3745–3758, 2004. 3

[67] F.L. Litvin and A.G. Wang. Local synthesis and tooth contact analysis of face-
milled, uniform tooth height spiral bevel gears. Technical report, NASA/CR-
1996–4757, 1996. 2

[68] The Mathworks. Optimization Toolbox - User’s guide. The Mathworks, Nat-
ick, MA - USA, third edition, 2005. 151

[69] Scott M. Miller. Kinematics of meshing surfaces using geometric algebra. In


DETC’03. 3, 92

[70] Giovanni C. Mimmi and Paolo E. Pennacchi. Non-undercutting conditions in


internal gears. Mechanism and Machine Theory, 35:477–490, 2000. 1

[71] David Montana. The kinematics of contact and grasp. The International
Journal of Robotics Research, 7(3):17–31, 1988. 2
Bibliography 183

[72] Massimo Pappalardo. Lezioni di Ricerca Operativa. SEU, Pisa, 2004. 152,
156

[73] Ettore Pennestrı̀. Dinamica e tecnica computazionale. Casa Editrice Am-


brosiana, Milano, 2002. 8

[74] I. H. Seol. The design, generation, and simulation of meshing of worm-gear


drive with longitudinally localized contacts. ASME Journal of Mechanical
Design, 122:201–206, 2000. 1

[75] Hermann J. Stadtfeld. Anforderungsgerechte Auslegung bogenverzahnter


Kegelradgetriebe. Ph.D. Dissertation, Aachen Techical University, 1987. 4

[76] Hermann J. Stadtfeld. Handbook of Bevel and Hypoid Gears. Rochester


Institute of Technology, Rochester, 1993. 4

[77] Hermann J. Stadtfeld. Advanced Bevel Gear Technology. The Gleason Works,
Rochester, 2000. 4, 5, 41, 49, 65, 66, 81, 82, 129, 130

[78] Gilbert Strang. Linear Algebra and Its Applications. Harcourt Brace Jo-
vanovich Publishers, San Diego, third edition, 1986. 145

[79] Xiaogen Su and Donald R. Houser. Alternative equation of meshing for worm-
gear drives and its application to determining undercutting and reverse engi-
neering. ASME Journal of Mechanical Design, 122:207–212, 2000. 1

[80] Gabriel Taubin. Estimating the tensor of curvature of a surface from a polyhe-
dral approximation. IEEE CNF, Fifth International Conference on Computer
Vision (ICCV’95), June 20-23, 1995, Massachusetts Institute of Technology,
Cambridge, Massachusetts, USA, 1995. 94

[81] J. Thomas and O. Vogel. 6m machine kinematics for bevel and hypoid gears.
1:435–451, 2005. 4, 6

[82] Chvátal Vasěk. Linear Programming. W. H. Freeman and Company, New


York, 1983. 152, 154, 157

[83] O. Vogel, A. Griewak, and G. Bär. Direct gear tooth contact analysis for hy-
poid bevel gears. Computer Methods in Applied Mechanics and Engineering,
191:3965–3982, 2002. 3, 98

[84] Da -Ren Wu and Jia -Shun Luo. A Geometric Theory of Conjugate Tooth
Surfaces. World Scientific, Singapore, 1992. 2, 3, 11, 15, 19, 20, 21, 44, 45,
72, 77, 78, 79, 80, 91, 92, 101, 119, 120, 122, 137

[85] H. S. Yan and W. T. Cheng. Curvature analysis of spatial cam-follower


mechanisms. Mechanism and Machine Theory, 34(2):319–339, 1999. 2, 72
184 Bibliography

[86] C. Zanzi and J.I. Pedrero. Application of modified geometry of face gear
drive. Computer Methods in Applied Mechanics and Engineering, 194:3047–
3066, 2005. 2

You might also like