Cad 01

Computer Aided Design
CAD course
Master in
Aerospace and Mechanical Engineering
1
Service...
Eric Béchet (it's me !)

Engineering studies in Nancy (Fr.)

Ph.D in Montréal (Can.)

Academic career in Nantes as Post-doc then ass. Prof. in
Metz (Fr.)
then Liège...
Christophe Leblanc

Assistant ( Ph.D. student ) at our university
Website http://www.cgeo.ulg.ac.be/CAO
Contacts :
cleblanc@ulg.ac.be
2
Procedures
st

12 Ex-Cathedra lectures (~2h except for the 1 )

Practical sessions

On the NX CAD software (5 4h-sessions)

Programming (8 2h-sessions)

Practical assessment : practical works (pw) are
rated.

Final test about theory (tt) – Jan./Sept.

The practical sessions, the project and the final
test are all mandatory, with over 8/20 to pass.
The final mark m=pw/3 + 2tt/3, over 10 to pass &
get the credits for the course. 3
Procedures

Lectures are given on Tuesday, 2 PM – Lecture
room is O1 ( B37 bldg.)

Begins today (exceptionally lasts 4h !), then

Sept. 26, Oct. 3, 10, 17, 24; Nov. 7, 14, 21, 28,
Dec. 5, 14, 19

Practical work begins Sept. 26th after the course
(check website for more dates & info)

It is organized by groups of around 20 students
formed in accordance with availabilities.

Any question : I am available mostly Fridays,
appointment via email or phone : 04-366-9165
Office : Bldg B52 +2/438 4
Outline (1st few classes)

Introduction

Brief history

Some basics about solid modeling

Parametric forms

Curves

Surfaces

Implicit forms

Plane curves et surfaces

Implicit volume modeling
5
Outline (whole course)

Interpolation and polynomial approximation

Cubic splines

Bézier curves

B-Splines

Curves and rational surfaces

In-depth B-REP models
6
Introduction
CAD/CAM – Computer Aided Design / Computer
Aided Manufacturing

Development since the 1950s-1960s

Iso Schoenberg (at U-Wisconsin) – splines (40's)

First « interactive » CAD graphic engine at the MIT
(I. Sutherland, Sketchpad) – 60's

James Ferguson (at Boeing) – splines (60's)

Paul De Casteljau (at Citroën) – Bézier curves (50's)

Pierre Bézier (at Renault) – '' ''

Carl De Boor (at GM / U-Wisconsin) – B-Splines (70's)

Catia, Unigraphics (now Siemens NX) date back to the mid-70’s,
emergence of 3d modeling kernels on mainframes.

Huge expansion in the 80’s with cheap personal
computers becoming ubiquitous. 7
Introduction

Sketchpad (Ivan Sutherland) – 1963
8
Introduction

Old mathematical and theoretical bases

Practical framework has been developed
because of industrial needs.

In return, it has contributed at the theoretical
level.
9
Basics
10
Solid modeling

Many types of CAD systems

First systems (such as Sutherland's) were aides for
drafting in 2D

With the evolution of computers, 3D capabilities
were added

Wireframe representation

Surface modeling

Then, real 3D modelers came over.

Today, most CAD system incorporate all
capabilities.

One can start with surface design, from which a 3D
model can be built, which yields at the end the 2D
technical drawings that may be printed on paper.
11
Solid modeling

More generally, what are the geometric queries
that a solid modeler is expected to respond to ?
Eg.

« Is a given point inside or outside a volume, on a
face or an edge, or even on an existing vertex ? »

« Generate a set of points located on the boundary
of the solid »

« Compute the union of two solids »

« Build one solid from a set of intersecting
surfaces »
...
12
Solid modeling

3D modelers, many types of representation for
volumes

Two families

Volumetric models

By construction

Half spaces (level-sets)

Composition of primitive objects (CSG)

By decomposition

Exhaustive enumeration
and structured meshes

Models defined by boundaries

Definition of the volume by
the enumeration of its boundaries
B-Rep
13
Solid modelling

Fundamental properties of solid models

Power of expression
Degree in the capability to represent all real solids, without
approximation.

Validity
Among the possible representations within the model, are some
representations invalid ?
14
Solid modelling


Non-ambiguity and uniqueness
Does a representation in the model correspond to one or several
real solids ? / Can a real solid be represented in different ways ?
The wireframe
representation
is ambiguous

Concision
To represent a particular solid, do we need many primitives or
operations ?
15
Solid modelling


Is the set of operations closed or open with respect to the
model ?
During the application of an operation onto a model, can we
obtain a non-valid solid, (are every operations valid ?) or a solid
that is non representable ?

Calculation cost and applicability
Is the construction of a solid within the selected model costly ?
Does the selected model address the issue raised by the user ?
16
CSG model
Description of a solid with the help of simpler
primitives and elementary operations is
ubiquitous in CAD modeling :

CSG model (Constructive Solid Geometry)

Internal representation
is in the form of a binary tree

Leaves are geometric primitives

Nodes are operations
« union », « difference »,
Wikipedia
« intersection »

Possible to :
« move », « complement »,
« change scale » etc...
17
Solid modeling

Characteristics of the modeling by half-spaces /
CSG

Power of expression: depends on the set of available
half-spaces (planes, quadratic surfaces, etc.) or
primitives
It is not possible to represent a single surface limited by 4
curves (a patch) , for example.

Validity : The half spaces are infinite (limited by planes
for example). A combination can give an infinite volume
i.e. not acceptable under the assumption that solids are
finite and “closed” volumes.

Ambiguity/uniqueness : each valid combination of half-
spaces determines an unique solid → non ambiguous.
A given solid can be represented by several different
combination of half-spaces (non uniqueness)
18
Solid modeling

Concision : the model by half-spaces is relatively
concise.

Closedness of the set of operations : we focus here on
boolean operations of union and intersection, these
always give a valid description → closed set of
operations

The operation of modeling itself is not costly (update of a
tree). The extraction of useful information can be much
more difficult e.g. calculation of surfaces... or very simple
such as collision detection (inside / outside test for a
point)
19
Solid modeling

CSG models

User-friendliness

Widely used in the interface shown by the CAD software
to the user

In general many CAD software
integrate some sort of
history of the construction
of the model
(cf model tree in Catia)
20
Solid modeling

Modeling by spatial decomposition

Division in regular cells

Exhaustive enumeration of non empty cells
M. Mäntylä
21
Solid modeling

Modeling by spatial decomposition

Cells of variable size

Recursive division of a cell containing the whole
solid in smaller cells located along the boundary of
the solid, until a prescribed error is reached.
22
Solid modeling

Modelling by spatial decomposition

Quadtree (2D)

Each node has 4 children.

Each children has a different representation in function of
the “filling” of the area
1 2
3 4
23
Solid modeling


Quadtree (2D)


Each children has a different representation in function of
the “filling” of the area
2.1 2.2
1
2.3 2.4
3.1 3.2
4
3.3 3.4
24
Solid modeling


Boolean operations : union, complement,
difference, intersection
\ =
_
U
U = = \ = =
_
U
U = = \ = =
_
U
U = = \ = =
U
U = = \ =
U
U = = \ =
U
U = = \ =
\ =
Union Intersection difference 25

complement
Solid modeling


Octree (3D) – analogy : voxels


Often used in medical images as means of efficient
storage

Effective means of geometric
location in 3D space
26
Solid modeling

Characteristics of the modelling by spatial
decomposition

Power of expression: By definition, it's an approximation
as soon as the boundaries of the object are not aligned
with the cells

Validity : always valid

Ambiguity/uniqueness

Every description corresponds to an unique solid → non
ambiguous.

A given solid is represented by a single way for the same root
cell (uniqueness)
→ However, since the representation is approximative, we can
consider that many real solids can lead to the same
representation.
27
Solid modeling

Concision : For a precise representation, the model can
become very heavy thus it is not very concise

Closedness of the set of operations : we speak here
about boolean operations of union and intersection, that
systematically give a valid description → closed set of
operations

Such models are easy to manipulate but with a limited
applicability due to high calculation costs (large quantity
of data). However some operations are efficient , collision
(inside/outside) test for instance.
28
Solid modeling

Description by meshes

It is a variant of the exhaustive decomposition that
is respecting to a certain extent the boundaries of
the object

Cells are not regularly organized anymore

Used mainly for scientific computations
29
Solid modeling

B-Rep model

« Boundary representation »

Model based on the representation of surfaces

Model of exchange (STEP format) and definition

The “natural” set of operators is richer than for CSG

Extrusion, chamfer etc ...

Does not inherently carry the history of construction
of the model (whereas CSG usually does)
30
Solid modeling

B-Rep model

Consists of two type of information

Geometric
Geometric information is used for defining the spatial position,
the curvatures, etc...
That's what we will see in the next couple of lectures – Nurbs
curves and surfaces

Topological
This allows to make links between geometrical entities.

Two types of entities

Geometric entities: surface, curve, point

Topological entities : volume, face, edge, vertex

A topological entity “lies on” a geometric entity,
which is its geometrical support 31
Solid modeling

B-Rep model

Simplified hierarchical model vertex-edge-face-
volume
v e v
v f
f
32
Solid modeling

B-Rep model

Simplified hierarchic model vertex-edge-face-volume
e1 v 1 v2
Composed of e2 v 2 v3
v7 v6 e3 v 3 v4
v8 v5 f1 e 1 e2 e3 e4
e4 v 4 v1
f2 … … … ...
...
f3 … … … ...
...
v4 e3 Composed of
e4 v3
f1
e2 Composed of
v1 a1 v2 v1 x 1 y1 z 1
v2 x 2 y2 z 2
Volume 1
v3 x 3 y3 z 3
...
33
Solid modeling

B-Rep model

Complete hierarchical structure found in most solid
modelers
Volume
1:n
« shells »
1:n Boundary
Face
1:{1 if manifold , 2 otherwise }
1:n 1:n Boundary

face « loops »
edge
1:{2 if manifold, n otherwise }
« Composed of » 1:2 Boundary

edge vertex
1:n
« is related to »
If it exists, the geometric

vertex support (G.S.) G.S. Expressed in
is expressed in cartesian parametric coordinates 34
coordinates x,y,z u,v or t
Solid modeling

Characteristics of the B-REP modelling

Power of expression: Allows to represent any solid if we
dispose of appropriate surfaces/curves (NURBS)

Validity : Every B-REP is not necessarily valid

Topological validity (see the Euler Poincaré relations in the
sequel)

Geometric validity

Ambiguity/uniqueness : Every valid B-REP represents an
unique solid → non ambiguous.
There are many different ways to represent a given solid
(non uniqueness)
35
Solid modeling

Concision : Generally concise

Closedness of the set of operations : In general,
operations made on B-REPs do not necessarily yield
regular models of real solids (in particular for boolean
operations)
Euler operators do maintain the topological validity, but
not necessary the geometric validity

boolean operators are costly, but other modelling
operators may be easier to implement using a B-REP
representation. In general, modelling operators are more
rich than for CSG
36
Solid modeling

Validity of the model

The boundary of a face is made up of edges that are not
allowed to intersect each other
invalid
37
Solid modeling

Validity of the model

The faces of a model can only intersect in common
edges or vertices.
Invalid
Pyramid on base
38
Solid modeling

How to represent individual entities ?

CSG

Primitives (Leaves of the CSG tree) → equations ???

Operators (Nodes " " " ) → How ?

B-REP

Vertices → trivial (coordinates)

Edges → straight line or ???

Faces → Planar or ???

Volumes → Indirectly

Need to define a geometric support.
39
Point

Point

Defined in an euclidean space (2D or 3D)

Notion of distance between two points is trivial.

Sometimes defined in a non euclidean space (e.g.
parametric space, in 1D = curve or 2D=surface )

Ex : points of a curve localized on a parametric surface.

Notion of distance between two points is not trivial on a
curve or a surface.

Sometimes, use of homogeneous coordinates (4th
coordinate)

4 dimensions + one equivalence relation → 3D

Perspective division to come back to euclidean 3D space.

Rational curves and surfaces. 40
Point

Homogeneous coordinates and class of
equivalence

Coordinates x, y, z and w (weight)

All points (4D) located on a line passing through the
origin O(0,0,0,0) are equivalent.

Particularly, a point located on the plane w=1.

Change from homogeneous coordinates to
euclidean coordinates by perspective division:
x x/w
()( )(
y ≡ y /w ≡
z
w
z/w
1
x/w
y /w
z/w )
41
Curve

Curve : 1D manifold in a space of higher
dimension

Plane curves (2D)

Position, tangent, curvature

Skew curves (3D)

Position, tangent, curvature, torsion
42
Curve
Meaning of curvature k

Deviation rate of the curve in comparison with a
straight line.

Radius of curvature is the inverse of the curvature

We can define a vector n that is normal at the
tangent and point to the center of curvature : it is
the normal vector
t 43
Curve
Meaning of torsion t

Deviation of the curve in comparison with a plane
formed by the tangent and the normal vector...

For a plane curve, torsion is zero.

We can define a vector normal to normal vector and
also to the tangent : it is the binormal vector b

b points to the torsion center
t
b n
44
Continuity
Geometrical continuity : Given two
« independent » curves :

They are discontinuous
 of continuity G0 (position p)
 of continuity G1 (p+tangent t)
 of continuity G2 (p+t+curvatures)
 ... G
45
Surfaces

2D manifold in 3D space

Tangent plane (defined by two linearly independent
vectors tangent to the surface)

Normal vector

Several curvatures may be defined

Gaussian (or total) curvature and average curvature

Principal curvatures
46
Different representations of curves and

surfaces
47
Modes of representation

There are many

Parametric form

Explicit form

Implicit form

Etc. (eg. solution of partial differential equations –
PDEs … )
48
Parametric representation
49
Use of scalar parameters to “walk” on the
curve/surface

The number of those scalars determine the dimension
of the entity and the mapping from one space (the
parametric space to the other (usually cartesian)
space.

0 – Point , 1 (u) – Curve , 2 (u,v) – Surface
3 (u,v,w) – Volume ( transformation used for finite elements)

We associate to those scalars some function for each
of the space coordinates ( x,y in the plane; x,y,z in 3D)
x= f (u) x= f (u , v)
P
{
⃗ (u)= y=g (u)
z=h(u)
or P
{
⃗ (u , v)= y=g (u , v)
z=h(u , v) 50
Parametric curves
51
For a curve, the representation takes the
following form: 8 y
x= f u 0
{
7 6
P u= y=g u
 1 5
z=hu 2 3 4 x
u is a real parameter.

We obtain all the points of the curve by varying u.

The parametrization is not unique !

The parameter u can be limited (limits of the curve)
52
Continuity of the parametrization
 A parametric curve is said to be Ck if the
parametrization P(u) is Ck (i.e. the kth derivative is
continuous)
 However : a parametric curve of class C may have an
angular point... (i.e. some continuity of the
parametrization does not necessarily involve the same
geometric continuity of the resulting curve !)
y
3
x=u
 u=
P {
y=u
2
x
0
 '
P u=0=
0 { 53
Regularity of the parametrization

A parametric curve P(u) is regular if the first
derivative P'(u) does not vanish at any place on all
the interval of definition.

On the contrary, points where P'(u) vanishes, are
called singular points.

A given curve can admit two parametric forms such
that one is regular and the other one isn't...

Example : curve y=x2 with two parametric forms P(u) and
Q(v)
3
 v= x=v 6
P u=
 x=u
{
y=u
2
 ' u=0= 1
P {
0
Q {
y=v
 ' v=0= 0
Q {
0
54
Regularity and continuity of the parametrization
 A curve of geometric continuity Gk can be described by
a parametrization that is non- Ck
 A curve of parametrization Ck may not be Gk
 A curve admitting singular points can be Gk and/or Ck.
 A regular curve is not necessary Ck and/or Gk.
To conclude :

Parametric continuity, geometric continuity, and
regularity are all relatively independent notions!
55
Ck -equivalent parametric forms

Two parametric forms P(u) and Q(v) of the same
curve are Ck - equivalents if there is a bijective
function j of class Ck whose reciprocal is also of
class Ck and such that:
v= u P   u
 u= Q
56
Length of a regular curve

It is a measure that is independent of the
parametrization (provided that those are C1-
equivalent)
b d = b
L=∫∣P
 ' u∣du= ∣  v∣dv
'
∫ Q
a c= a
57
Natural parametrization

It is a parametrization s such that : B y
A
s=0.1
L  AB=s  B−s A x

s is also named curvilinear abscissa along the
curve.

It can be built from any regular parametrization u of
the curve :
u
 t ∣dt
su=∫∣P
'
A
58

Parametrization of a circle in the plane :
x=r cos u , u∈ [ 0, 2 [
{ y=r sin u
r

Natural parametrization of the same circle...
s
{ x=r cos
y=r sin
r
s
r
, s∈ [ 0, 2  r [ s=r u
59
Differential geometry for parametric curves
Position P : x u
 

' dP d2 P
''
P u= y u P= P= 2
du du
z u
y

Tangent vector T :
'
dP dP du P
T u= = ⋅ = ' N T
ds du ds ∣P ∣
x

Normal vector N :
z B
Norm(u) '' ''
N (u)= with Norm(u)= P −( P ⋅T ) T
∣ Norm(u)∣ (It is not defined when Norm(u) = 0 !)

Binormal vector B :
B u=T u× N u
60
Frenet's frame and equations B s
d T  s
[ ]
T  s
ds = 0  s T s

2D
d N s [
− s 0
⋅
][ ]
N  s N  s
ds
d T  s

3D
[]ds
d N s
ds
d Bs
ds
[
0
= − s
0
 s
0
− s
0 T  s
][ ]
  s ⋅ N s
0 Bs
s
Works only with s : the curvilinear abscissa ! 61


Curvature k : |P ' × P ''|
dT
'
T u⋅N u
k=
| |
| |
ds
=
|P |'3
 u=
∣P ' u∣
Curvature vector

Torsion t: = N u u
d 2 T dT
 u=
'
N u⋅B u
=
 ds ds
2
, ,T
 =
 P ''' , P '' , P ' 
'
∣P u∣ 2 '
∣P × P ∣ '' 2
62
Parametric surfaces
63
A parametric surface is represented as :
u=cst
x= f u , v

{
P u , v = y=g u , v
z=hu , v
u,v
u,v are two real parameters

v=cst

All the points of the surface can be y z
obtained by varying u and v.
x

Again, the parametrization is not unique !

The interval of definition is limited (limits the
surface) 64
A curve can be defined on the surface
Parametric space « Ambient » space
u=cst
x= f u , v 
 uv
 t :
u=u t 
{
v=v t  {
P u , v : y= g u , v

z=hu , v
t
x= f u t  , v t 


{
 t : y=g u t  , v t 
z=hu t  , v t
The equations seen for parametric
y z
v=cst
curves are valid in this particular case. x
65
Regularity and continuity of the parametrization
 A parametric surface is of class Ck if the mapping
P(u,v) onto R3 is of class Ck.

A parametrization is regular if and only if (iff)
∂
P 
∂P
u 0 , v0 × 0 ∀ u 0 , v 0 ∈ D⊂ℝ 2
u 0 , v 0 ≠ 
∂u ∂v

The points on the parametric space for which it is
not the case are called singular points.
 Ck–equivalent parametric forms follow the same
definitions as for curves
66
Differential geometry for parametric surfaces
x u , v


Position P : P
 u , v= y u , v 
z u , v
Unit tangent vectors T u and T v :
[ ] P
u
 = ∂
∂u
P

P
 =uv∂ P

∂u∂v
⋯
2
−1 u −1 v
∂P ∂P P ∂P ∂P P
u
T u , v=
 ⋅
∂u ∂u ∣ ∣ = u
∣P ∣
v
T u , v =
 ⋅
∂v ∂v ∣ ∣ = v
∣P ∣

These vectors are, in general, not orthogonal !

Tangent plane (parametric form)
u v
Pt u

0 , v  a ,b= P u 0 , v 0 a⋅T u 0 , v 0 b⋅T u 0 , v 0 
0
  
or a , b∈ℝ
2
u v
Pt (u , v ) (a , b)= P (u 0 , v0 )+a⋅P (u 0 , v 0 )+b⋅P (u 0 , v0 )
⃗
0 0
⃗ ⃗ ⃗
67

Normal vector N :
Norm(u , v) u v u v
N (u , v)= with Norm(u , v)=T ×T or P ×P
|Norm(u , v)|
u=cst
Tv
Tu
u ,v
N
v=cst
68
dS

Area of a surface
A=∬ dS dv P v
Ω
u v u v
dS=∣du⋅P ×dv⋅P ∣=∣P ×P ∣dudv du P u
1st fundamental form


Other notation of the area of a surface
2 2
∣a ×b∣ =
a⋅a ⋅ b⋅b−a⋅b Lagrange's identity
dS =√(e g− f 2 )dudv with e= P u⋅P u , f =P u⋅P v , g= P v⋅P v

2
A=∬  eg− f dudv
D
69

Length of a curve on a surface
∂ P u , v u
x= f u , v  P =
P u , v : y= g u , v

z=hu , v { v
P =
∂u
∂ P u , v
∂v
x= f (u(t ) , v (t ))
⃗ uv
Γ (t ):
u=u(t ) ⃗
{
v=v (t ) {
Γ (t ): y= g (u(t ) , v (t ))
z=h(u(t) , v (t ))
'
Γ=
dP (u(t ) , v (t ))
dt
' ∂ P (u(t ) , v (t )) du ∂ P (u(t) , v (t )) dv

Γ= +
Derivative of the curve expressed
∂u dt ∂v dt
in the parametric space of the surface Derivative of the surface (in 3D)
' ' u ' v

 t =u t  P ut  , v t v t  P ut  , v t  70
b b 2
 t ∣dt=
L=∫∣
a
'
∫
a
∣ t ∣ dt
'
' ' u ' v

 t =u t  P ut  , v t v t  P ut  , v t 
∣ ' t ∣2=e u' t 2 2 f u ' t v ' t  g v ' t 2

u u u v v v
with e= P ⋅P , f = P ⋅P , g= P ⋅P
71

If we set
ds= √ e u ' (t )2 + 2 f u ' (t ) v ' (t )+ g v ' (t )2 dt
that is equivalent to : ds= √ e du 2 + 2 f dudv+ g dv 2
s b
( we have L= ∫ ds=sb−sa )
s a 
, we get a quadratic form :

'
' 2 ' ' ' 2
e u t  2 f u t  v t  g v t  = u t  v t  
e
f
' '
  
f u t 
g v ' t 
b

'
L=∫
a
 u' t    
v ' t   e
f
f u t  dt
g v ' t 
72

Angle between two curves ...
'
Γ (t 1 )⋅Γ (t 2 )=∣Γ (t 1 )∣∣Γ (t 2 )∣cos α=( u (t 1) v (t 1) ) e f u (t 2 )
'
1
'
2
'
1
'
2
f
'
1
'
1 ( )( 2
g v (t 2 )
'
2
)
'
e f u 2 (t 2 )
u '
( 1 1 1 1) f
(t ) v '
(t ) ( )(
g v '2 (t 2 ) )
cos α=
√ '
1
'
(e
( u (t 1 ) v (t 1)) f
1
f
)( u 1' (t 1 )
g v 1 (t 1 )
' ( '
) '
u 2 (t 2 ) v 2 (t 2 ) ) e
f ( f
)( u '2 (t 2 )
g v '2 (t 2 ) )
73

The first fundamental form is the application :
e f du 2 du 2
uv
1
uv
j1 (d Γ , d Γ )= ( du 1 dv 1 )
2 ( f )( )
g dv 2
=( du 1 dv 1 ) M1
dv 2 ( )
with e=P u⋅P u , f = P u⋅P v , g=P v⋅P v

It is a symmetric bilinear form that allows to measure real
distances from variations in the parametric space...
 The matrix M1 is a representation of the metric tensor,
often noted g ij , i , j=u , v ∂ x/∂ u ∂ x /∂ v
 M1 also is related to the Jacobian matrix
(
J = ∂ y /∂ u ∂ y /∂ v
of the transformation (u,v) → (x,y,z) (it is JTJ ).
∂ z /∂ u ∂ z /∂ v )
b
uv uv j(d Γ uv
1 , d Γ uv
2 )
L=∫ √ j(d Γ , d Γ ) dt cos α= uv uv uv uv
a √ j(d Γ 1 , d Γ1 )j(d Γ 2 , d Γ 2 )
A=∬ √ det M 1 dudv =∬ det J dudv under some conditions

D ( D ) 74

Curvatures...
Geodesic curvature
or relative curvature
w/r to the surface :
kg
Simple or total curvature

of the 3D curve : k
Normal curvature :
kn
75

Curvature vectors N
dT
k= =k⋅n=k n + k g (e.g. from Frenet's
ds relations) P(u,v)
k n =k n⋅N k g =k g⋅G T
G =N ×T
kn
n

Necessarily, we do
have : N⋅T =0

Differentiating yields : kg
k
dT dN
⋅N +T⋅ =0
ds ds
dT ds=
dT dN
kn=
ds
⋅N =−T⋅
ds (
kg=
ds )(
⋅G √ e du 2 +2 f dudv + g dv 2 )
d P (u , v) d N d P⋅d N l du2 +2 m dudv +n dv 2
=− ⋅ =− = Center o
ds ds ds
2 2
e du + 2 f dudv+ g dv
2
with curvatur
dP d N 1 dP d N dP d N dP d N
l=− ⋅ m=− ( ⋅ + ⋅ ) n=− ⋅
du du 2 du dv dv du dv dv 76
dP d N 1 dP d N dP d N dP d N
l=− ⋅ m=− ( ⋅ + ⋅ ) n=− ⋅
du du 2 du dv dv du dv dv

This can be rewritten by noting that
dP and dP
⋅N =0 ⋅N =0
du dv
, and integrating by parts UV =∫ U dV +∫ V dU
where e.g. for l : U = dP dN

dV = ⋅du
du du
d2 P d2 P d2 P
l= 2 ⋅N m= ⋅N n= 2 ⋅N
du du dv dv
77
Curvatures and the second fundamental form

The second fundamental form is the application
l m du 2 du2
uv uv
j2 (d Γ , d Γ )= ( du1 dv 1 )
1 2 ( )( )
m n dv 2
=( du 1 dv 1 ) M 2
dv 2 ( )
uu vv uv
with l= N⋅P , n=N⋅P , m= N⋅P

Normal curvature of a curve on a surface :
uv uv
uv
j2 (d Γ , d Γ ) uv
j2 (d Γ , d Γ )
kn= kn= uv uv
ds
2
j1 (d Γ , d Γ )

Geodesic curvature (curvature in the tangent plane)

Measures the deviation of the curve in comparison with a
geodesic
78
Definition of the geodesic

It is a curve with a minimal length on a surface.

In the plane, those are lines

On a sphere, great circles

By definition, the geodesic has no geodesic curvature.
Relations between « simple » curvature, geodesic
curvature and normal curvature is obtained from the
curvature vectors
k=k n + k g k n =k n⋅N k=k⋅n k g =k g⋅( N ×T )=k g⋅G
k 2 =k 2n + k 2g
The geodesic curvature is therefore readily obtained from
the others (except for the sign, which is same as d T ⋅G )
2 2 2 du
k g =k −k n
79

A circle of radius R in the plane
Curvature k = 1/R
Geodesic curvature kg= 1/R
Normal curvature kn= 0

A big circle on a sphere or radius R
Curvature k = 1/R
Geodesic curvature kg= 0
Normal curvature kn= 1/R

In the previous slides, k is the curvature of the
curve in 3D (not in the parametric space !!!)
80
uv uv
j 2 (d Γ , d Γ )

Curvatures of the surface k n=
j1 (d Γ uv , d Γ uv )
 The normal curvature kn at a point only depends on the
local orientation of Γ on the surface. If we sweep through
all the orientations, this curvature passes by a minimum
kmin then a maximum kmax for two perpendicular directions.
 These directions and the particular values of kn are
respectively called principal directions and principal
curvatures of the surface. They do not depend on the
parametrization of the surface...

They are the solution of an eigenvalue problem.

The Shape operator (also called Weingarten Map)
plays a key role here, see what follows.
81

Shape operator
W p (w )=−∇ w⋅N w=w u⋅P u +w v⋅P v

Here, N is the normal to the surface, p is a point on the
surface, w is a tangent vector to the surface at p. ∇ w is a
directional derivative along a unit vector w (in the surface) :
∂ +w ∂
it is equal to w u v .
∂u ∂v
This operator gives back a vector in the tangent plane.

If one works in an orthonormal basis, eigenvectors are easy
to compute as the operator is symmetric...

Let's build such an orthonormal basis from the natural one :
u ∂P v ∂P P u× P v
P = P = N = N (u , v )= u
∂u ∂v v v ‖ P × P v
‖
u
P P −( P ⋅t 1 )t 1
t 1= u t 2 =N ×t 1=
‖P ‖ ‖P v −( P v⋅t 1 )t 1‖ 82

It is easy to show that one can switch from one
basis to the other using coefficients of the 1 st
fundamental form …
Pu Pu eP −f P
v u v
eP −f P
u
t 1= u = t 2= =
‖P ‖ √ e √ e √ eg − f 2
√e √ h
t1 1 √h 0 Pu
[] =
[
t 2 √ e √h − f ][ ]
e Pv

One can also switch the other way back, working on
coefficients rather than on vectors e.g. for any
vector w u
⃗ =w 1 t 1 +w 2 t 2 =w u P + w v P
v
wu = 1 w 1 =Φ w 1
One gets
[ ]
wv √e √ h 0 [
√h − f
e ][ ] [ ]
w2 w2
83

Now back to the shape operator : we will start with
a vector expressed in the orthonormal frame
(w1,w2) , convert it to the parametric basis (wu,wv),
take the directional derivative w u ∂ +w v ∂ of the
∂u ∂v
normal n and then project back to the orthonormal
frame.
W p (w )=− t 1 [ nu nv ]Φ w 1
[] [ ]
t2 w2
u
P w1
=−Φ
P [ ] [ ]
T
v
[n u n v ]Φ
w2
u
P

Here, [ ]
− v [n u nv ]is precisely the matrix of the
P
coefficients of the second fundamental form...
84

In the orthonormal basis
1 √h
W p (w)=ΦT l m Φ w 1 = 0 l m √ h − f w1
[ ][ ] [
m n w2 e h − f ][ ] [
e m n 0 e ][ ]
w2
1 hl (e m− f l) √ h
=
[
eh (e m− f l ) √ h 2 2
][ ]
w1
f l +e n−2 e f m w 2

In the original basis
T l m
W p (w)=Φ W p (w)Φ wu =Φ(Φ Φ)Φ w u
~ −1
wv [ ] m n
−1
[ ] [ ]
wv
−1
1 g − f l m wu
=Φ ΦT l m w u = = e f ⋅l m w
[ ][ ] [
m n wv h − f e m n wv ] [ ][ ] [ g ] [ m n ] [ w ]
f
u
(Weingarten Equation)
85

Eigenvalues and eigenvectors of the shape
operator

The shape operator gives the normal curvature for a
given orientation (i.e. the derivative of the unit normal
vector with respect to a small increment along that
orientation on the surface)

One can work either in the original frame (non
orthonormal)
du 0 ~
( L−k I ) ( )()
dv
=
0
−1
with L= M 1 M 2 =W p (w ) det (L−k I )=0

Or the orthonormal frame, for which the matrix is
symmetric and the 2D eigenvectors are orthogonal
dt 1 0
(W p (w)−k I )
( )()
dt 2
=
0
det (W p (w)−k I )=0
86
du 0 l −k e m−k f du 0
( M 2−k M 1 )
dv ( )()
=
0 ( m−k f n−k g )( ) ( )
dv
=
0

The eigenvectors are obtained using any of the relations:
du l−k e du m−k f
( )(
⋅
dv m−k f
=0 ) ( ) (
dv
=α⋅
k e−l ) , α>0: du2 +dv 2=1

Just use the other relation if the basis contains a null
vector for one of the values of k :
du m−k f du l−k g
( )(
⋅
dv l−k g
=0 ) ( ) (
dv
=α⋅
k f −m ) , α>0: du +dv =1
2 2
87

Example u
 
P u , v= v
cos u
1 0
u
P u , v= 0
 
−sin u
v

P u , v= 1
0
0 0
uu
P u , v= 0
 
−cos u
uv

P u , v= 0
0
0 P u ×P v −sin u
vv
P u , v = 0
0  N (u , v)= u
∣P ×P ∣ ( )
v
= 0 ⋅
1
1
2
√ 1+sin u 88
du 0
( L−k I ) ( )()
dv
=
0
−1
with L= M 1 M 2
u u u v 2

P ⋅P
M 1= u v
P ⋅P
P ⋅P
P v⋅P v
=
1sin u 0
0 1 
−cos u
( )
uu uv
N⋅P N⋅P 0
M2= (
N⋅P uv N⋅P vv )
= √ 1+sin 2 u
0 0
89

Numerical application : u=0 and v=0
2
M 1=  1sin u 0
0 1
=
1 0
0 1   c2
1 0 c1
−1
M =
1  
0 1
o
−cos u
M2=
( √ 1+sin
0
2
u
0
0
)(
=
−1 0
0 0 )
c 1= 1
L=M −1
1 M 2 = −1 0
 
 1=−1
 0 C 3D
1 =( P u , P v ()
)⋅ 1
0
0 0
c 2= 0
 2 =0
1 C 3D
2 =( P u , P v ()
)⋅ 0
1 90
Umbilical points
 Points whose normal curvature kn is independent of
the orientation of Γ .

Locally, the surface looks like a plane or a sphere.

In this case, the following relations hold for the
matrices M1 and M2 :
l m n
= = if f ≠0
e f g l
k n = =kmin =kmax = const.
e
l n Principal directions are arbitrary.
= if f =0 and m=0
e g
91

Gaussian curvature and average curvature

Invariants (do not depend on the parametrization)
det ( M 2 ) l n−m2
kT =k min k max =det ( L)= =
det ( M 1 ) e g− f 2
 min  max 1 1 e n g l−2 f m

M = = trace L=
2 2 2 e g− f
2
 L=M −1
1 M 2
92
Local nature of the surface
 kT>0 the surface is locally an ellipsoid – elliptic point
 kT<0 : hyperbolic paraboloid – saddle point –
hyperbolic point
 kT=0 and kM<>0 : parabolic cylinder – parabolic
point
 kT=0 and kM=0 : no information – locally plane
93
Christoffel symbols

Complete developments. Christoffel symbols

Expression of the first derivatives along a
curve...
∂ P (u , v) u' = du(t )
P u= dt
∂u
∂ P (u , v) ' dv (t )
P v= v=
∂v dt
Γ' (t )=u ' P u +v ' P v

Lets differentiate that expression again...
94
Christoffel symbols
 (Γ ' (t)=u ' P u +v ' P v )'
Γ" (t )=u " P u +u ' (u ' P uu +v ' P uv )+v " P v +v ' (u ' P vu +v ' P vv )
2 2
=u " P u +v " P v +u ' P uu +u ' v ' P uv +u ' v ' P vu +v ' P vv

The terms u " P u +v " P v are in the tangent plane

The terms P ij , i , j={u , v } can be expressed as
linear combinations of tangential and normal
components :
u v
Gauss formula
P ij =C P u +C P v + Lij N
ij ij
The C are the Christoffel symbols (often

k
ij
written Γij in textbooks ) and the Lij are the
k
coefficients of the second fundamental form. 95

Christoffel symbols

Using Einstein's notation (summation over
k
repeated indices ): P ij =C ij P k + Lij N , i , j , k =u , v

How to compute C and Lij ? k
ij

Multiply by P l , l=u , v
0
k
P ij⋅P l =C P k⋅P l + Lij N⋅P l
ij
Coefficients of the 1st

fundamental form = g kl = metric tensor
−1 k
P ij⋅P l⋅( g kl ) =C ij g kl⋅g kl
−1 Conventional
notation of the
k
inverse of the
−1 k lk
C = P ij⋅P l⋅( g kl )
ij C = P ij⋅P l⋅g
ij metric tensor
96
Christoffel symbols
k
P ij =C P k + L ij N
ij

Now, multiply by N
k
P ij⋅N =C ij P k⋅N + L ij N⋅N
1
0
Lij =P ij⋅N

One recovers the formula seen before:
N⋅P uu N⋅P uv
Lij ≡M 2 M2= (
N⋅P uv N⋅P vv )
97
Christoffel symbols

Use of Christoffel symbols and the second
fundamental form

Let us return to the expression
2 2
Γ" (t )=u " P u +v " P v +u ' P uu +u ' v ' P uv +u ' v ' P vu +v ' P vv

Now replace u , v by ui ,i=1,2 and use the Einstein
notation again :
i " i ' j '
Γ" (t )=(u ) P i +(u ) ⋅(u ) P ij

Now substitute P ij =C ijk P k + L ij N
k " k i ' j ' i ' j '
Γ" (t )=((u ) +C (u ) ⋅(u ) ) P k +(u ) ⋅(u ) Lij N
ij
Total tangential component Normal component 98

Christoffel symbols

Used to express curvatures (or second derivatives)
in the 3D euclidean space from curvatures (or
second derivatives) expressed in the parametric
space.

Normal curvature: based on the normal component
i ' j '
(u ) ⋅(u ) L ij N
If we have an arc-length parametrization :
i ' j ' uv uv
k n=Γ"⋅N =L ij (u ) (u ) ( =j2 (d Γ , d Γ ))
if not : need to adjust
i ' j ' i ' j '
Lij (u ) (u ) L ij (u ) (u )
k n= 2
= k ' l '
ds g kl (u ) (u ) uv uv
j 2 (d Γ , d Γ )

One recovers the formula : k N = uv uv
j1 (d Γ , d Γ ) 99
Christoffel symbols

Geodesic curvature : based on the tangential
components
k " k i ' j '
((u ) +C (u ) ⋅(u ) ) P k
ij

If the curve is parametrized by arc-length :
=0
Γ '⋅Γ ' =1 Γ '⋅Γ "=0
k " k i ' j ' i ' j '
Γ"⋅Γ ' =0=((u ) +C (u ) ⋅(u ) ) P k⋅Γ ' +(u ) ⋅(u ) Lij N⋅Γ '
ij

Therefore, the tangential component Γt " of Γ" is
orthogonal to Γ ' and to N as well. There is a
proportionality factor between Γt " and N ×Γ ' .
This factor is the geodesic curvature
100
Christoffel symbols

Γt "=k g ( N ×Γ ' ) Γt "⋅( N ×Γ ' )=k g ( N ×Γ ' )⋅( N ×Γ ' )
k g =( N ×Γ ' )⋅Γt " =( N ×Γ' )⋅Γ "
Therefore, k g =[ N , Γ' , Γ" ] , valid only if the
parametrization of Γ is normal (arc-length)

If not, need to adjust ...
[ N , Γ ' , Γ" ] [N ,Γ' ,Γ"]
kg= 3
= l ' 3/2
ds k '
( g kl (u ) (u ) )

Some developments* leads to the following identity :
2 3 1 3 2 1 2 1 2 2
√ det ( M 1)[−C 11 u ' +C 22 v ' −(2 C 12−C 11)u ' v ' +(2 C 12 −C 22 )u ' v ' +u " v '−v " u ' ]
kg = 3/ 2
u'
(( u' ( ))
v ' ) M1
v'
Again, this expression makes only use of derivatives
in the parametric space... *Surfaces
see e.g.“Modern Differential Geometry of Curves and
with Mathematica”, 2nd ed. Boca Raton, FL: CRC 101
Press, pp. 501-518, 1997.
Christoffel symbols

Equations of a geodesic
Need to set that the geodesic curvature is equal to
zero.

One obtains a set of differential equations

May be used to actually compute an approximation
of the geodesic since it is usually impossible to
solve that coupled system of ODEs algebraically.
102
Darboux frame and equations
x= f (u , v) x s= f u s , v s
P
{
⃗ (u , v): y=g (u , v)
z=h(u , v)
 uv s: u=u s
 {
v=v s

{
  s: y s= g u s , v s
z s=hu s , v s
 s= ∂ x s , ∂ y  s , ∂ z  s
 s 
G
  s
N  s
T
Darboux frame
T {∂s ∂s ∂s }
  s

 s= N
N  u s , v s
⃗ ( s)= N
G ⃗ ×T
⃗

Su , v
s is the curvilinear abscissa

103
Darboux frame and equations
 s
G Darboux frame
  s
N  s
T
  s

d T s
[]ds
d G s
ds
d N  s
ds
[−
0
= − g  s
n  s
 g  s  n s T s
 g
0
s 0 ][ ]
− g s ⋅ G s
N  s

Su , v
Geodesic torsion (or relative torsion)

104

Geodesic torsion
is identically zero iff the curve is along a direction of
principal curvature (line of curvature)

Geodesic curvature
is identically zero iff the curve is a geodesic
105

In an euclidean space:

The shortest line between two points has simple
curvature k equal to zero (= straight line)

The minimal surface carried by a curve is of
average curvature kM equal to 0 everywhere

In a non euclidean space (on a surface):

The (or one of the) shortest line between two points
on the surface has a geodesic curvature kg equal to
zero (it's a geodesic)
106

A good reference on « shape interrogation »

Shape interrogation is the ability to get information
from geometric models
N. M. Patrikalakis and T. Maekawa, Shape Interrogation

for Computer Aided Design and Manufacturing, Springer
Verlag, February 2002. ISBN 3-540-42454-7
(Available as a PDF electronic copy at the library)
107
Some results of differential geometry
Gauss-Bonnet's theorem
∫ kT dA+∫ k g ds=2 π χ (S )
S ∂S
 S  is Euler's
characteristic
of surface S
=−2
=0 108
=2 =1 =−4
Application of Gauss-Bonnet's theorem

For a polyhedron, we have (Euler's formula) :
 S = N − E F
We want to approximate a sphere with hexagons ( H ) and pentagons ( P ).

Each pentagon (hexagon) has 5 (6) nodes, shared by 3 faces

Each pentagon (hexagon) has 5 (6) edges, shared by 2 faces

So the total number of nodes N is ( 5P+6H )/3

Total number of edges E is ( 5P+6H )/2, total number of faces F
is P+H
 S = N −E F =5P6H/3−5P6H/ 2P H =P /6

The sphere is compact and its gaussian curvature is 1/R2
1
∫ T dA∫ G ds= R2 ∫ dA=4 =2  S 
S ∂S S
so P=12 … 12 pentagons are necessary and as many hexagons
as we want to make soccer ball.
109
110
Explicit curves and surfaces
111
Explicit forms

Curve as a function
y= f  x z= f  x , y

We can swap x, y and z according to needs
y y y
y= f  x x= f  y 
OK ! x OK ! x KO ! x
112
Explicit forms

Polar coordinates

Sometimes used. Circle is readily represented :
r =1
113
Explicit forms
Definition of an explicit surface


Depend on reference axis

No multiple definitions i.e. unique value of z for a
position (x,y)

For a given axes system, only one definition

No closed curves /surfaces in one time

Unit circle – we must split it up at least in :
2 2
y=  1− x y=− 1− x
and
114
Implicit representation
115
Implicit forms

f  x , y=0
2 2

Ex. Unit circle : x  y −1=0

Infinity of representations of a
given curve f =0

Invariance by the multiplication by a constant
k⋅f  x , y=0 ⇔ f  x , y=0 ∀ k ≠0

We can take absolute value too
∣ f  x , y∣=0 ⇔ f  x , y=0
... but the form is then not differentiable

etc...
116
Implicit forms
Advantages of an implicit form
For a point P [ x p , y p ] we know :

If it is on the curve : f  x p , y p =0

If it is not on the curve : f  x p , y p ≠0

In some cases, we may know if it is inside or
outside the volume delimited by the curve:
f ( x p , y p )<0 or f ( x p , y p )>0
2 2
If x  y −10 , P is inside the circle
p p

But : ∣x 2p y 2p−1∣0 has no meaning

Multiplication by negative k : interior exterior
117
Implicit forms
Normal forms / Contour line / level sets

Particular case of implicit forms such that :
  x , y =d , d is the signed distance to the curve

In this case, for a point P [ x p , y p ] ,   x p , y p 
is the distance from the point to the curve. d<0 means
usually that the point is inside.
2 2

Case of the unit circle :   x , y=  x  y −1
is the distance to the circle.

Unique representation !
(except the sign of the distance 0
function – but it indicates 0
where is the interior)
=0 118
Implicit forms
Normal forms

A normal form is a distance function, we have :
∣∇   x , y∣=1

For the circle :
∂ 2x
∣ ∣∣ ∣
∂x
∂y
∂z

2
2x y 2
x
2
∣∇ ∣= = 2 y = 2 2  2 2 =1

2
∂ 2x y
0
2
x y
y

x y
2
119
Implicit forms
Normal forms

Ellipse ?

We can « flatten » a circle :   x * , y * =  x *2  y *2 −1
* *
Set y =2 y ; x = x
2 2
  x , y=  x 2 y  −1
x

∣ ∣
2
x 4 y
2
2
x 16 y
∣∇ e∣= 4y = x 24 y 2 ≠1

2
 x 4 y
0
2 
2
0
=0
0
120
Implicit forms

The normal form associated with a curve (or a
surface in 3D) is usually nontrivial.
To get it nevertheless:

Use of a numerical approximation (spatial discretization)

We know the position of the curve or the surface: j=0 on Γ

The normal form is reconstructed in the considered
domain by solving the « Eikonal » equation (from greek
« image »):
|⃗
∇ j|=1 in Ω
{ j=0 on Γ
 ext
 int

121
Implicit forms
Resolution of the Eikonal equation

It is a non-linear, hyperbolic equation

« Fast Marching » method

Cf . J.A. Sethian, Fast Marching Methods, SIAM Review
(41) 2, pp 199-235, 1999

Consist in a step-by-step construction of the
distance field, from the initial front where d = 0.
J.A. Setian
122
Implicit forms

Resolution algorithm at first order on a regular
grid
J.A. Setian
Known value in a point = 0 a) Calculation of test points
123
Implicit forms

First order resolution algorithm on a regular grid
J.A. Setian
b) Among test points, we c) We fix it and (re)calculate
choose the one of lowest new test points
value 124
Implicit forms

Resolution algorithm at first order on a regular
grid
J.A. Setian
b) Among test points, we c) We fix it and we recalculate...
choose the one of lowest Here a test point has been computed
value twice : we keep the lowest value 125
Implicit forms

The former procedure is repeated until all grid
points have been fixed.

The result is a (possibly signed) distance field
known at each grid point. Interpolation inside
square patches makes it continuous.

One can extract an iso-value (e.g. for 0) and
consider it as the surface of interest

Also, one may actually modify the field (with
transport equations for instance) to change the
shape and position of the surface.
126
Implicit forms

There are algorithms of order 2 or higher

It is also possible to solve that equation efficiently
on non regular grids ( unstructured meshes)

Example : mesh
127
Implicit forms

There are algorithms of order 2 or higher

It is also possible to solve that equation efficiently
on non regular grids ( unstructured meshes)

Example : iso-distance curves
128
Implicit forms
There are high-performance algorithms to update
discretized level-sets

Resolution of partial differential equations
cf. J.A. Sethian « Level-Sets Methods and Fast Marching Methods, Evolving
interfaces in computational geometry ... », Cambridge University Press,
1999.

Alternative representation of the geometry really useful
for real objects (i.e. non perfect)

eg. Medical imaging, « historical » artifacts , etc...

But also evolving interfaces (fluid surface, cracks in
solids)
129
Implicit forms

Transport equation of a level-set
∂j
n =0
+V n⋅∇ j⋅⃗
∂t

We solve that using explicit time-stepping

Normal speed is given (known)

Below, the speed is depending on the density ...
N. Paragios
130
Implicit forms

Example of geometries based on « level-sets »
131
Implicit forms

Example of geometries based on « level-sets »
M.Y. Wang
132
Implicit forms
Some words on differential geometry...
133
Implicit forms
∂  x , y , z
[ ]
Normal (ext.) to the curve / surface
∂x

We use the gradient 
∇ = ∂  x , y , z
 is equal to 1 if j is ∂y
 ∇
N = a signed distance ⋮
∣
∇ ∣
( a normal form )
Tangent vector for a planar implicit curve

Use of a permutation
∂  x , y , z 
of the gradient

T =
Tan 
∣
∇ ∣

Tan =
−
[ ∂ y
∂x , y , z
∂x
]
134
Implicit forms
Curvature of a planar implicit curve

Use of the Hessian matrix =0
∂2  ∂2 
T
[∂ x2
H = 2
∂ 
∂ y∂x
∂x∂ y
∂2 
∂y 2 ] Osculating
circle
Ro
T  ∗H ∗
T  1
=− = Basis vector (along z)

∣∇ ∣ Ro
Alternative
notation =−
  T
Tan  ⋅n
Tan  ∗∇ Tan  × 3
 3
∣Tan ∣
135
Implicit forms
Demonstration of the equation of the curvature
d T s
From Frenet equations :
[ ]

ds 0   s T  s
d N  s d N  s
=
[−  s 0 ][ ]
⋅
N  s
we have : =− ⋅T s ds
ds
∂ N  s dx ∂ N s dy
and so : =− 
∂ x ds

∂ y ds
⋅T  s 
∂ Nx ∂Ny dx

now, ∇ N=
[ ] []
∂x
∂ Nx
∂y
T
∂x
∂Ny
∂y
T s =
ds
dy
ds
=−T  s ∗∇ N s∗T s

136
Implicit forms

We have :
∇  ∣∇ ∣ ∇  ∇   −∇ ∣∇ ∣ ∇ 
∇ N =∇
 
∣∇ ∣
=
∣∇ ∣
2
∇  ∇   =H  ∇ ⋅T =∣∇ ∣⋅N⋅T =0
T
T T ∗H ∗T 
=−T  s ∗∇ N  s∗T s=−
∣∇ ∣
Quod Erat Demonstrandum, QED
137
Implicit forms
Curvature of a surface

There are 2 curvatures :
 Gaussian curvature ( = total curvature ) kT
 Average curvature kM .
 These are related to principal curvatures kmin and
kmax (curvature measured along principal
directions) :
2  min  max
 min , max = M ±   − T
M M =  T = min  max
2
Along principal directions T1 and T2 , curvatures are
extrema ( kmin or kmax )
138
Implicit forms
Direct expression of principal directions

Complicated but explicit formulas exist
Cf.
Erich Hartmann, On the curvature of curves and surfaces defined by
normalforms, Computer Aided Geometric Design 16, pp. 355-376,
1999.
Numerical expression of the curvatures

Cf. sequel...
143
Implicit forms
Numerical expression of the curvatures

Conjugate transpose Hessian matrix (matrix of the
cofactors) :
∂2  ∂2  ∂2 
H =
[
∂ x2
∂2 
∂ y∂x
∂2 
∂z∂x
∂x∂ y
∂2 
∂ y2
∂2 
∂z∂ y
∂x∂z
2
∂ 
∂ y∂z
2
∂ 
∂ z2
] [
 , xx  , xy  , xz
=  , yx  , yy  , yz
 , zx  , zy  , zz ]
 , yy  , zz − , zy  , yz  , yz  , zx − , yx  , zz  , yx  , zy − , yy  , zx
*
[
H  =  , xz  , zy − , xy  , zz  xx  , zz − , xz  , zx  , xy  , zx − , xx  , zy
 , xy  , yz − , xz  , yy  , yx  , xz − , xx  , yz  , xx  , yy − , xy  , yx
144
]
Implicit forms
Expression of curvatures of an implicit surface
 T *
∇  ∗H ∗∇
T = 4
∣
∇ ∣
2
 T
∇ −∣
∇  ∗H ∗ ∇ ∣ ⋅traceH 
M = 3
2∣
∇ ∣
145
Implicit forms
Implicit curves in 3D
146
Implicit forms
Curve in 3D

Intersection of two implicit surfaces
{  x , y , z =0 } ∩ {   x , y , z =0 }
N
T
B


147
Implicit forms
Tangent
Tan  ,  =∇   x , y , z ×∇   x , y , z 
Tan  ,  =1 if we deal with two normal forms that
T  , = are “orthogonal”
∣Tan  , ∣
Curvature

For planar curves : =−
 T
Tan  ⋅n
 Tan  ∗∇ Tan ×

3
3
∣
Tan ∣

By substituting ... (we must take the norm , because
that vector is not along n3 anymore)
=
∣
 Tan ,  T∗∇ Tan  ,  ×
Tan  ,  ∣
∣Tan  , ∣3

148
Implicit forms
Torsion

The formulas are obtained from the definition of the
torsion for the parametric curves

We set :
*
T =Tan  ,  =∇   x , y , z ×∇   x , y , z 
** * *
T =T ∗∇ T
*** * * *T * * *
T =T ∗∇  ∇ T ∗T T ∗∇ T ∗∇ T
 T * ,T ** ,T *** 
=
* ** 2
∣T ×T ∣

Note : the curvature, written with the same kind of
notation, is : ∣T *×T **∣
=
*3
∣T ∣ 149
Implicit forms
Volume operations on contour lines (level-sets)

We can represent a volume as the set of points
such as
≤0

Manipulations of volumes

Union, intersection, etc...
150
Implicit forms
Compounds of level-sets
Union  u = 1∪ 2

u  x , y=min [1  x , y , 2  x , y]


Intersection  i = 1∩ 2
i  x , y=max[ 1  x , y , 2  x , y]

Complement  c =
c  x , y =−  x , y 
d =1 ∖  2 = 1∩ 2

Difference
d  x , y=max[1  x , y ,− 2  x , y]
151
Implicit forms

Disc 
Square
2 2
 d  x , y=  x  y −1 
4 half planes !
c
  1  x , y= y
1
 2  x , y =− y2

3  x , y=− x0.2
2 
4  x , y = x−1.8
3 4
Region of interest
152
Implicit forms
2 2

Use of intersections for
d  x , y=  x  y −1 the square :
 sq= 1∩ 2∩ 3∩ 4
c  sq  x , y=max[1,  2,  3,  4 ]
1

Finally, difference
between  sq and d
 f = sq ∖ d
2  f  x , y=max[ sq ,− d ]
3 4  f  x , y=max[ 1,  2,  3,  4,− d ]
Region of interest
153
Implicit forms
Disadvantges of an implicit representation

Obtaining points on the curve/surface :

Resolution of a set of non-linear equations

Not very robust, and slow

No parametrization , no « explicit » orientation

Discretization of the curve/surface(« rasterization »)

From a point, use of specific algorithms (“marching
cubes”)

Not always possible and/or robust
154

Cad 01

Încărcat de

Informații document

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Cad 01

Încărcat de

Drepturi de autor:

Formate disponibile

Computer Aided Design

Union Intersection difference 25

1:{1 if manifold , 2 otherwise }

1:n 1:n Boundary

1:{2 if manifold, n otherwise }

« Composed of » 1:2 Boundary

If it exists, the geometric

Different representations of curves and

Works only with s : the curvilinear abscissa ! 61

u,v are two real parameters

curves are valid in this particular case. x

1st fundamental form

dS =√(e g− f 2 )dudv with e= P u⋅P u , f =P u⋅P v , g= P v⋅P v

' ∂ P (u(t ) , v (t )) du ∂ P (u(t) , v (t )) dv

' ' u ' v

' ' u ' v

∣ ' t ∣2=e u' t 2 2 f u ' t v ' t  g v ' t 2

, we get a quadratic form :

A=∬ √ det M 1 dudv =∬ det J dudv under some conditions

Simple or total curvature

where e.g. for l : U = dP dN

 min  max 1 1 e n g l−2 f m

The C are the Christoffel symbols (often

coefficients of the second fundamental form. 95

Coefficients of the 1st

Total tangential component Normal component 98

s is the curvilinear abscissa

Geodesic torsion (or relative torsion)

N. M. Patrikalakis and T. Maekawa, Shape Interrogation

(Available as a PDF electronic copy at the library)

Explicit curves and surfaces

Definition of an explicit surface

Some words on differential geometry...

=−T  s ∗∇ N s∗T s

∇  ∇   =H  ∇ ⋅T =∣∇ ∣⋅N⋅T =0

u  x , y=min [1  x , y , 2  x , y]

S-ar putea să vă placă și