Professional Documents
Culture Documents
Notes MB
Notes MB
Simos Evangelou
1.1 Aims
• To introduce theoretical approaches for the modelling of multibody mechanical systems.
Multibody mechanical systems are those systems that consist of a finite number of rigid
bodies joined together either in a tree structure or with kinematic loop closures. Many physical
objects around us can be described accurately by such representations and their study requires
the use of ordinary differential equations. In contrast, we will not deal with the mechanics of
continua (for example, hydrodynamics (fluids), elasticity, electromagnetism, etc.).
• Distinguish between the two main branches of Classical Mechanics: vectorial and analyt-
ical mechanics.
• Develop models, in the form of differential equations, that describe the motion of real
mechanical systems, using Newton’s laws and Lagrangian equations of motion.
• Derive equations of motion for systems with holonomic and nonholonomic constraints.
1.3 Syllabus
• Basic vector calculus
• Newtonian mechanics
2
– Virtual work
– D’Alembert’s principle
– Hamilton’s principle and Lagrangian equations of motion
• Past papers
3
Modelling and Control of Multibody Mechanical Systems
Teaching and Assessment
– Pre-recorded videos
– Notes
4
Process:
• Simultaneous reporting
• Peer assessment
2.5 Assessment
Assessment is by open-book exam (75%) and coursework (25%).
The coursework is broken down into:
5
Modelling and Control of Multibody Mechanical Systems
Introduction
• satellites
• robots
• mems
• submersible vehicles
6
y x Steer axis
rider Twist axis x’
upper
body y’
z
aero z’
main
3.2 History
Important milestones in the science of Classical Mechanics:
• Vectorial mechanics
This treatment starts directly from Newton’s laws of motion. The main idea is to recognize
all the forces that act on any given particle at every instant and uniquely define its motion
as a consequence of these forces. The action of a force is measured by the momentum
produced by that force.
• Analytical mechanics
This treatment bases the entire study of equilibrium and motion on two fundamental
scalar quantities, the “kinetic energy” and the “potential energy”. Leibniz was the first
to use the kinetic energy as a gauge for the dynamical action of a force and he is the
originator of analytical mechanics. Euler and Lagrange were the first discoverers of the
exact principle of least action for conservative systems, where “action” is the time integral
of the kinetic energy over the entire motion. Hamilton’s principle is similar, with the
“action” being the difference between the kinetic and the potential energies, but it works
7
Modelling and Control of Multibody Mechanical Systems
for non conservative systems as well. The Hamiltonian formulation of the principle of
least action asserts that the actual motion realised in nature is that particular motion for
which this action assumes its smallest value.
k
m
8
↓
The spring force is
F = −kx, k > 0.
Newton’s Second Law gives
mẍ + kx = 0.
If we define the natural frequency ωn as
r
k
ωn = ,
m
we get the equation of a linear harmonic oscillator
ẍ + ωn2 x = 0.
k
x0
equilibrium
position
x
m
↓
There is no dissipation and therefore the total energy is conserved:
T + V = constant.
m
c
↓
The spring force is given by
Fs = −kx, k > 0,
and the damper force is obtained from
Fd = −cẋ, c > 0.
Use of Newton’s Second Law (Ftotal = mẍ) gives
Fs + Fd = mẍ,
or
mẍ + cẋ + kx = 0.
p
If we define the natural frequency ωn = k/m and the damping ratio ζ = c/(2mωn ) the
equation of motion becomes
ẍ + 2ζωn ẋ + ωn2 x = 0.
It is straightforward to show that the general solution of this equation is
√ √
2 2
x(t) = A1 e(−ζ+ ζ −1)ωn t + A2 e(−ζ− ζ −1)ωn t ,
where A1 and A2 are integration constants and can be found when specific initial conditions are
given. ↑
10
3.4.4 Forced spring-damper-mass system
k
Fext
m
c
↓
The spring force is given by
Fs = −kx, k > 0,
the damper force is given by
Fd = −cẋ, c > 0,
and the external input that forces the system is Fext . Use of Newton’s Second Law (Ftotal = mẍ)
gives
Fs + Fd + Fext = mẍ,
or
mẍ + cẋ + kx = Fext .
p
If we define the natural frequency ωn = k/m and the damping ratio ζ = c/(2mωn ) the
equation of motion becomes
Fext
ẍ + 2ζωnẋ + ωn2 x = .
m
The general solution of this equation is the solution of the unforced system (previous example)
plus the particular solution with the forcing term included.
↑
Fext
k2
k1
m1 m2
c2
x1 x2
↓
The total force on mass m1 is given by
11
Modelling and Control of Multibody Mechanical Systems
and the total force on mass m2 is given by
ẋ2
the input
u = Fext ,
and write the equations of motion in state space form
ẋ = Ax + Bu,
12
where
0 0 1 0
0 0 0 1
A=
− mk1 − k2 k2
,
1 m1 m1
− mc21 c2
m1
k2 k2 c2
m2
−m 2 m2
− mc22
and
0
0
1 .
B=
m1
0
These equations are linear and in order to solve them we can make use of well developed
techniques.
↑
13
Modelling and Control of Multibody Mechanical Systems
Basic tools
Our study will begin with the vectorial treatment of mechanics. Hence, we will review some of
the basic ideas and rules of vector algebra and vector calculus.
z
A
r
k rz
j y
O
i rx
ry
x
−→
origin O, any point in space, such as A, can be defined in terms of the line segment OA starting
from O and ending at A, which represents some vector r. The vector r is called the position
vector of the point A relative to the origin O.
The position vector can be given in terms of a basis set of vectors that define a particular
coordinate system. Most commonly we use an orthonormal right-handed set of unit vectors
denoted by i, j and k as shown in Figure 4.1. Such a coordinate system is known as Cartesian
or rectangular.
Therefore,
r = rx i + ry j + rz k. (4.1)
The position vector, r, can also be expressed in the form
r = rr̂, (4.2)
where r̂ is a unit vector in the direction of the position vector and r is the magnitude of the
position vector given by q
r = |r| = rx2 + ry2 + rz2 . (4.3)
14
In this and following sections, unless otherwise stated, we will make use of the Society
of Automotive Engineers (SAE) convention for Cartesian coordinates. That is a set of right-
handed axes with the z axis pointing downwards. Such set of axes and the associated unit
vectors are shown in Figure 4.2.
O i x
j
k
where θ is the angle between the two vectors. As the name of this product suggests the result
is a scalar quantity.
If the two vectors, r1 , r2 , are given in Cartesian coordinates,
r1 = x1 i + y1 j + z1 k, (4.5)
r2 = x2 i + y2 j + z2 k,
r1 · r2 = x1 x2 + y1 y2 + z1 z2 . (4.6)
The dot product can also be used to find the component of a given vector in a certain
direction. In Figure 4.3 the component of vector r in the direction of the unit vector n is rn .
It is given by
rn = |r| cos θ = r · n. (4.7)
Obviously if two vectors are perpendicular to each other their dot product is zero.
where θ is the smallest angle between r1 and r2 , and n is the unit vector perpendicular to the
plane that contains the two vectors, r1 and r2 , such that {r1 , r2 , n} is a right-handed set.
The result of the cross product is a vector quantity.
15
Modelling and Control of Multibody Mechanical Systems
n
rn
A
θ
r
O
If r1 and r2 are given in Cartesian coordinates as in Equations 4.5, then the cross product
can be found by evaluating the determinant
i j k
r1 × r2 = x1 y1 z1 . (4.9)
x2 y2 z2
By definition, r1 × r2 = −r2 × r1 , and also the cross product of two parallel vectors is zero.
16
The derivative of a vector function of a scalar variable is another vector. In the case of
the position vector, r, the derivative with respect to time furnishes the velocity vector. The
velocity vector can be differentiated once more to provide the acceleration vector.
When r is given in rectangular coordinates in the form
r = xi + yj + zk, (4.13)
where,
x = x(t), (4.14)
y = y(t),
z = z(t),
d2 r
= r̈ = ẍi + ÿj + z̈k. (4.16)
dt2
eθ A
er r
θ
θ=0
O
i
Figure 4.4: Planar polar coordinates of the point A and the polar unit vectors.
Figure 4.4 shows the polar coordinates r, θ of a point A and the polar unit vectors
er , eθ that have point O as their origin. As A moves around, the polar unit vectors do not
remain constant. They have a constant magnitude, which is unity, but their directions change
according to the θ coordinate of point A. Therefore, er and eθ are vector functions of the scalar
variable θ.
In order to obtain the derivatives of the two unit vectors with respect to θ we write er and
eθ in terms of the Cartesian basis vectors i, k as follows
17
Modelling and Control of Multibody Mechanical Systems
Since er , eθ are now expressed in terms of the constant vectors i, k, the differentiation with
respect to θ is simple and gives
der
= eθ , (4.18)
dθ
deθ
= −er .
dθ
Suppose now that A is a moving particle. Its coordinates, r, θ will be functions of time t. The
position vector of A relative to O has a magnitude r and direction er and can therefore be
written
r = rer . (4.19)
To obtain the velocity vector of A in polar form, we differentiate Equation 4.19 with respect
to t and we get
dr d der
v= = (rer ) = ṙer + r . (4.20)
dt dt dt
Now er is a function of θ which is, in its turn, a function of t. Hence, by the chain rule and
Equation 4.18,
der der dθ
= = θ̇eθ . (4.21)
dt dθ dt
By substituting into Equation 4.20 we rewrite the polar velocity vector into
The polar acceleration vector is obtained similarly by differentiating the velocity vector in
Equation 4.22 with respect to t. We get,
dv
a= = r̈ = (r̈ − r θ̇2 )er + (r θ̈ + 2ṙθ̇)eθ . (4.23)
dt
z
k
A eθ
er
r
k z
j y
O
i
θ r
18
them the out-of-plane z coordinate which is then treated in a Cartesian like manner. Every
point in space is determined by the r and θ coordinates of the projection of its position vector
in the xy plane, and the z coordinate of the same vector. Thus the position vector is
Note that when√using cylindrical coordinates, r is not the modulus of r. The modulus of r is
given by |r| = r 2 + z 2 .
4.2.4 Spherical
z eφ eθ
er
A
r
k
j φ y
O
i
θ
In spherical coordinates, we make use of two angles θ, φ and a radial distance r to specify
the position of a particle.
The spherical unit vectors written in Cartesian coordinates are
We can differentiate these vectors using a similar procedure used in the case of polar coordinates
earlier. The derivatives of the unit vectors are thus given by
19
Modelling and Control of Multibody Mechanical Systems
These expressions can be used to find the velocity and acceleration vectors. The position vector
is
r = rer , (4.29)
and by differentiation we obtain the velocity as
4.2.5 Gradient
If V is a scalar function of position, the gradient of V in Cartesian coordinates x, y, z is given
by
∂V ∂V ∂V
∇V = i+ j+ k. (4.32)
∂x ∂y ∂z
R r′
y
O′
z
S′
20
Such a translational motion of the reference frame can be characterised as follows. The
origin O ′ of a frame S ′ moves along some given curve with respect to the origin O of an inertial
frame S in a given way with respect to time t. As can be seen in Figure 4.7, the radius vector
r ′ , of a moving particle at A, measured in the moving system and the radius vector r measured
in the absolute system, are in the following relation to each other:
r = R + r′. (4.33)
r = r′. (4.38)
Nevertheless, the velocities and accelerations measured in both systems differ from each other
because of the fact that rates of change observed in the two systems are different. If a certain
vector B is constant in S ′ it rotates with the system and thus, if observed in S, undergoes in
the time dt an infinitesimal change
dB = (Ω × B) dt, (4.39)
where Ω is the angular velocity vector which gives the rotation of S ′ with respect to S. Hence
dB
= Ω × B, (4.40)
dt
while at the same time
d′ B
= 0. (4.41)
dt
21
Modelling and Control of Multibody Mechanical Systems
We have introduced here the notation d′ /dt which refers to the operation of observing the rate
of change of a quantity in the moving system S ′ .
If, on the other hand, the vector B is not constant in S ′ , so that d′ B/dt 6= 0, we find, by
making use of the superposition principle of infinitesimal processes, that
dB d′ B
= + Ω × B. (4.42)
dt dt
Applying this equation to the radius vector r = r ′ we get
dr d′ r ′
= + Ω × r′. (4.43)
dt dt
This yields the following relation between the velocities v and v ′ :
v = v′ + Ω × r′, (4.44)
where
d′ r ′
v′ = . (4.45)
dt
In order to obtain the relation between the absolute and relative accelerations a and a′ ,
where
d′ v ′ d′2 r ′
a′ = = , (4.46)
dt dt2
we differentiate the Equation 4.44:
dv dv ′ dr ′
= + Ω̇ × r ′ + Ω × . (4.47)
dt dt dt
On the right-hand side, the relation 4.42 can be employed in order to express everything in
terms of quantities which belong to S ′ . The result is:
dv d′ v ′
= + 2Ω × v ′ + Ω × (Ω × r ′ ) + Ω̇ × r ′ , (4.48)
dt dt
or
a = a′ + 2Ω × v ′ + Ω × (Ω × r ′ ) + Ω̇ × r ′ . (4.49)
The first term on the right-hand side of this equation is the acceleration measured in the moving
frame, the second term is the Coriolis acceleration, the third one is the centripetal acceleration
and the fourth term is the Euler acceleration.
22
The relation between the accelerations is obtained by differentiating Equation 4.51 and we get
a = R̈ + a′ + 2Ω × v ′ + Ω × (Ω × r ′ ) + Ω̇ × r ′ . (4.52)
The expressions obtained here for the velocity and acceleration are thus the linear super-
position of the expressions obtained for the two separate cases of pure translation and pure
rotation.
Example
We derived earlier the expression for the acceleration vector of a moving particle in terms of
planar polar coordinates (Equation 4.23). We now show how to derive the same expression by
making use of the ideas of rotating reference frames explained above.
We start by making the assumption that the Cartesian unit vectors i, j, k in Figure 4.4
(j comes out of the page towards the reader) define the inertial reference frame. We take er ,
eθ , j as the unit vectors defining a moving reference frame which has the same origin as the
inertial reference frame and rotates about an axis in the j direction.
↓
The rotational velocity of the moving frame relative to the inertial frame is given by
Ω = θ̇j, (4.53)
Ω̇ = θ̈j. (4.54)
We assume that a particle is moving along the direction of the er vector and we would like to
find its acceleration. The position vector of this particle is
r = r ′ = rer , (4.55)
since the two frames have a common origin. The velocity v ′ as seen by an observer on the
moving frame is given by
d′ r ′
v′ = = ṙer . (4.56)
dt
Note that, the term which results from the differentiation of er with time should be omitted
because in the local reference frame the particle looks as if it is travelling in a straight line.
Similarly, the relative acceleration is given by,
′ d′ v ′
a = = r̈er , (4.57)
dt
again ignoring the term arising from the time derivative of er .
If we now substitute into the Equation 4.49 we get
which reduces to
a = (r̈ − r θ̇2 )er + (r θ̈ + 2ṙ θ̇)eθ . (4.59)
This is exactly the same as Equation 4.23.
↑
23
Modelling and Control of Multibody Mechanical Systems
Newtonian mechanics
24
5.1.2 Momentum
By examining Equation 5.3 above we can see that if the total external force, F , is zero then
ṗ = 0 and the linear momentum, p, remains constant. This is the conservation theorem for the
linear momentum of a particle.
We can also examine the properties of the angular momentum, H, of the particle about
point O. This is defined as
H = r × p, (5.7)
where r is the radius vector from O to the particle. Obviously if we change the order of the
product factors the sign will change. The moment of force or torque about O is
N = r × F. (5.8)
5.1.3 Work
We now consider the work done by the external force F on the particle in going from point 1
to point 2. By definition this work is
Z 2
W12 = F · ds. (5.12)
1
W12 = T2 − T1 . (5.15)
25
Modelling and Control of Multibody Mechanical Systems
If the force field is such that the work W12 is the same for any physically possible path between
points 1 and 2, then the force (and the system) is said to be conservative. In such a case the
force obeys the following property: I
F · ds = 0, (5.16)
which says that the work done around a closed path is zero. It is clear that a system cannot
be conservative if friction or other dissipation forces are present.
A necessary and sufficient condition that the W12 be independent of the physical path taken
by the particle is that F be the gradient of some scalar function of position:
where V is called the potential, or potential energy 1 . We note that in Equation 5.17 we can add
to V any quantity constant in space, without affecting the results. Hence, the zero level of V
is arbitrary. For a differential path length we have the relation
It is therefore obvious that for a conservative system the work done by the forces is
W12 = V1 − V2 . (5.18)
T1 + V1 = T2 + V2 (5.19)
which states the energy conservation theorem for a particle: If the forces acting on a particle
are conservative, then the total energy of the particle, T + V , is conserved.
The above theorem does not hold when the force applied to the particle is given by the
gradient of a scalar function that depends explicitly on both the position of the particle and
time.
F = Fx i + Fy j + Fz k. (5.20)
If we then equate the components on the two sides of this equation we get
Fx = mẍ,
Fy = mÿ, (5.22)
Fz = mz̈.
1
This is a consequence of a well known theorem of vector calculus, the Stokes’ theorem.
26
This is a system of ordinary differential equations (ODE’s). In order to find a solution to these
equations we need to specify the initial position and initial velocity vectors:
r0 = x0 i + y0 j + z0 k, (5.23)
ṙ0 = v0 = vx0 i + vy0 j + vz0 k.
In general, the forces acting on a body will depend on its position, r, its velocity, v (= ṙ),
and sometimes explicitly on the time t.
Analytical solutions
Apart from a few special cases we cannot solve analytically the equations of motion and the
only way to obtain solutions is via numerical integration. One such special case is when F is
either constant or depends on t only. We can then integrate the Equations 5.22 one by one to
compute the components of the velocity vector:
1 t
Z
vx (t) = vx0 + Fx (t)dt,
m 0
1 t
Z
vy (t) = vy0 + Fy (t)dt, (5.24)
m 0
1 t
Z
vz (t) = vz0 + Fz (t)dt,
m 0
and then integrate the velocity vector components to obtain the position vector components:
Z t
x(t) = x0 + vx (t)dt,
0
Z t
y(t) = y0 + vy (t)dt, (5.25)
0
Z t
z(t) = z0 + vz (t)dt,
0
Other special cases for which an analytical solution is possible are when the equations of motion
are linear in terms of the differential terms.
Numerical integration
With numerical integration of the equations of motion we do not compute the exact solution
of the problem but an approximation to it.
In order to demonstrate the method of numerical integration we convert Equations 5.22 into
first order equations as
ẋ = vx ,
ẏ = vy ,
ż = vz ,
1
v̇x = Fx (x, y, z, vx , vy , vz , t), (5.26)
m
1
v̇y = Fy (x, y, z, vx , vy , vz , t),
m
1
v̇z = Fz (x, y, z, vx , vy , vz , t),
m
27
Modelling and Control of Multibody Mechanical Systems
These equations can be written more compactly as
The array X is usually referred to as the state vector or vector of state variables, even though it
is not a real vector, such as the physical vectors of velocity and acceleration. The components
of X do not even have the same units.
The topic of finding numerical solutions to systems of ODE’s is mature and well developed.
Many algorithms for obtaining solutions are available. To illustrate the main idea we make use
of the simple “forward Euler method”.
Consider a time increment ∆t which has been chosen to be small enough so that the solution
is accurate, but at the same time it is chosen not to be smaller than necessary so as to reduce
the number of steps needed to reach a solution. We then start from the initial condition X0
and move forward in time by calculating X at time ∆t. The procedure is then repeated by
using the previously calculated value of X as the initial condition and computing the new X
at time 2∆t. This procedure is repeated as many times as it is necessary. The update of X at
each discrete time t is given by the formula
5.1.5 Example 1
Consider a particle that moves in space under the influence of gravity only. If we make use of
Cartesian coordinates then Equations 5.22 are applicable and give
↓
ẍ = 0,
ÿ = 0, (5.30)
z̈ = g.
This is a set of linear differential equations which can easily be integrated to give
x = vx0 t + x0 ,
y = vy0 t + y0 , (5.31)
z = 21 gt2 + vz0 t + z0 ,
where vx0 , vy0 , vz0 are the initial velocity components and x0 , y0 , z0 are the initial position
coordinates of the particle. The motion problem is thus solved completely. Equations 5.31 are
the equations of motion of projectiles (in the presence of gravity).
↑
28
O i
k
c r
Fθ Fr
eθ
mg
er
5.1.6 Example 2
Consider the problem of a small mass m suspended from a coiled spring of negligible mass as
shown in Figure 5.1. Assume that m is free to move in a vertical plane under the action of
gravity, the spring and two controlling forces Fr and Fθ .
↓ Since
the motion is confined into the vertical plane the position vector of the mass particle is given
by
r = xi + zk. (5.32)
with the point O being the origin. The vector sum of the external forces on the particle is
r z x
F = mgk + (−c(r − r0 ) + Fr ) + Fθ ( i − k), (5.33)
r r r
where r0 is the unstretched length of the spring and c the usual spring constant. If we substitute
into this expression the position vector from Equation 5.32 we get
x z z x
F = ((−c(r − r0 ) + Fr ) + Fθ )i + (mg + (−c(r − r0 ) + Fr ) − Fθ )k, (5.34)
r r r r
The right-hand side of Equation 5.5 is
d2 r
m = mẍi + mz̈k. (5.35)
dt2
By combining Equation 5.34 and 5.35 we obtain the equations of motion in vector form
x z z x
((−c(r − r0 ) + Fr ) + Fθ )i + (mg + (−c(r − r0 ) + Fr ) − Fθ )k = mẍi + mz̈k, (5.36)
r r r r
and therefore
x z
mẍ = (−c(r − r0 ) + Fr ) + Fθ , (5.37)
r r
z x
mz̈ = mg + (−c(r − r0 ) + Fr ) − Fθ .
r r
29
Modelling and Control of Multibody Mechanical Systems
r is given by √
r= x2 + z 2 , (5.38)
and by substitution the equations of motion become
r0 x z
mẍ + c(1 − √ )x = √ Fr + √ Fθ , (5.39)
+z x2
2 2
x +z 2 x + z2
2
r0 z x
mz̈ + c(1 − √ )z − mg = √ Fr − √ Fθ , (5.40)
2
x +z 2 2
x +z 2 x + z2
2
which are second order coupled nonlinear differential equations of x and z. In this case it is
not as straight forward to integrate these equations as was the case in the previous example.
It seems that, instead of rectangular coordinates, polar coordinates would fit more naturally
to the problem. We now show how to solve the same problem using polar coordinates.
We rewrite the x, z coordinates in terms of polar coordinates r, θ,
x = r sin θ, (5.41)
z = r cos θ.
In order to substitute the new coordinates in Equations 5.37 we need to obtain ẍ, z̈. By
differentiating Equation 5.41 with respect to time once we get
We now substitute Equations 5.41 and 5.43 into Equation 5.37 and get
c Fr Fθ
(r̈ − r θ̇2 ) sin θ + (r θ̈ + 2ṙθ̇) cos θ + (r − r0 ) sin θ = sin θ + cos θ, (5.44)
m m m
c Fr Fθ
(r̈ − r θ̇2 ) cos θ − (r θ̈ + 2ṙθ̇) sin θ − g + (r − r0 ) cos θ = cos θ − sin θ. (5.45)
m m m
We can process these equations further to separate the r̈ and θ̈ terms. If Equation 5.44 is
multiplied by sin θ and Equation 5.45 is multiplied by cos θ and the two added together we
obtain
c Fr
r̈ − r θ̇2 − g cos θ + (r − r0 ) = . (5.46)
m m
Furthermore, if Equation 5.44 is multiplied by cos θ and Equation 5.45 is multiplied by − sin θ
and the two added together we obtain
Fθ
r θ̈ + 2ṙθ̇ + g sin θ = . (5.47)
m
Equations 5.46 and 5.47 are the equations of motion in polar coordinates. Integration of these
second order differential equations provides r and θ as functions of time. The forces Fr and Fθ
can be used to provide control. We will study this later.
A more straightforward way to obtain these equations is possible when we make use of the
polar unit vectors that rotate with the particle. To see this consider writing the force vector
30
and acceleration vector in terms of the polar unit vectors. The polar unit vectors belong to a
reference frame which has O as its origin. The acceleration vector is given in Equation 4.23.
We therefore get,
(mg cos θ − c(r − r0 ) + Fr )er + (−mg sin θ + Fθ )eθ = m((r̈ − r θ̇2 )er + (r θ̈ + 2ṙ θ̇)eθ ). (5.48)
The coefficients of the unit vectors, er , eθ furnish directly the polar equations of motion given
above. ↑
(e)
where Fi is the total external force acting on the particle, and Fji is the internal force on
the ith particle due to the jth particle. Fii is obviously zero. At this point we will make the
(e)
assumption that the Fij (as well as the Fi ) obey Newton’s third law of motion which states
that the forces two particles exert on each other are equal and opposite. This assumption,
which is not universally true, is sometimes referred to as the weak law of action and reaction.
For all particles together Equation 5.49 takes the form
d2 X X (e) X
mi r i = Fi + Fji . (5.50)
dt2 i i i,j
i6=j
The first sum on the right is the total external force F (e) , while the second term vanishes, due
to the law of action and reaction which states that each pair Fij + Fji is zero. The left-hand
side can be reduced by defining a vector R as the average of the radii vectors of the particles,
weighted in proportion to their mass:
P P
m i r i mi ri
R = Pi = i . (5.51)
i mi M
The vector R defines a point known as the centre of mass of the system (Figure 5.2). Using
this definition Equation 5.50 reduces to
d2 R X (e)
M = Fi ≡ F (e) , (5.52)
dt2 i
which states that the motion of the system is defined by the motion of the centre of mass as
if all the mass was concentrated at that point. The internal forces, when they obey Newton’s
third law, therefore have no effect on the motion of the centre of mass .
31
Modelling and Control of Multibody Mechanical Systems
mj
centre of mass
mi
rj R
ri
5.2.2 Momentum
The total linear momentum of the system is given by
X X dri dR
P = pi = mi =M , (5.53)
i i
dt dt
which, by making use of Equation 5.51, amounts to the total mass of the system times the
velocity of the centre of mass. If the external forces are zero, Equation 5.52 gives that the total
linear momentum is conserved, which is the conservation theorem for the linear momentum of
a system of particles.
To obtain the total angular momentum of the system we form the cross products ri × pi
and sum over all particles. This operation is performed in Equation 5.49 and with the aid of
the identity in Equation 5.10 we obtain
X X d X (e)
X
(ri × ṗi ) = (ri × pi ) = Ḣ = ri × Fi + ri × Fji . (5.54)
i i
dt i i,j
i6=j
The last term on the right hand side of Equation 5.54 amounts to a sum of the pairs of the
form
ri × Fji + rj × Fij = (ri − rj ) × Fji , (5.55)
where the equality of action and reaction has been used. The term ri − rj is the vector from
j to i (see Figure 5.3) and therefore the right-hand side of Equation 5.55 can be written as
rij × Fji .
If the internal forces between the particles lie along a line joining the particles in addition to
being equal and opposite (a condition known as the strong law of action and reaction) then all
these cross products vanish. With this assumption Equation 5.54 can be written as
dH
= N (e) , (5.56)
dt
32
rij
rj
ri
Figure 5.3: The vector rij between the ith and jth particle.
which states that the time derivative of the total angular momentum is equal to the moment
of the external force about the given point. From Equation 5.56 we obtain the conservation
theorem for the total angular momentum: the total angular momentum vector is constant in
time if the external torque is zero.
We have seen earlier in Equation 5.53 that the total linear momentum of the system is the
same as if the entire mass was concentrated at the centre of mass and moving with it. There is
a corresponding theorem for the angular momentum which is more complicated. With a given
reference point O the total angular momentum of the system is given by
X
H= r i × pi (5.57)
i
Let R be the position vector from O to the centre of mass and let ri′ be the vector from the
centre of mass to the ith particle. We have (see Figure 5.4)
ri = ri′ + R (5.58)
and
dri
= vi = viR + Ṙ, (5.59)
dt
where Ṙ is the velocity of the centre of mass relative to O, and
dri′
viR =
dt
is the velocity of the ith particle relative to the centre of mass of the system. Equation 5.59 is
equivalent to Equation 4.51 when the centre of mass is the origin of the moving reference frame.
Equation 5.58 can be used to expand the angular momentum expression in the following form
!
X X X d X
H= R × mi Ṙ + ri′ × mi viR + mi ri′ × Ṙ + R × mi ri′ . (5.60)
i i i
dt i
P
Both the last two terms in the right-hand side contain the factor mi ri′ which is a null vector
because it defines the radius vector of the centre of mass in the coordinate system whose origin
33
Modelling and Control of Multibody Mechanical Systems
ri′
centre of mass
ri
Figure 5.4: The vectors involved in the calculation of the angular momentum.
is the centre of mass. Therefore the last two terms vanish and rewriting the remaining terms,
the total angular momentum about O is
dR X ′
H =R×M + r i × pR
i . (5.61)
dt i
In words, Equation 5.61 says that the total angular momentum about point O is the angular
momentum of the system concentrated at the centre of mass, plus the angular momentum of
motion about the centre of mass.
We can now simplify Equation 5.56 to describe the motion about the centre of mass. Let
the angular momentum of motion about the centre of mass be denoted by H0 ,
X
H0 = ri′ × pR
i , (5.62)
i
(e)
and the moment of the external forces about the centre of mass be denoted by N0 ,
(e) (e)
X
N0 = ri′ × Fi . (5.63)
i
in which by substituting ri form Equation 5.58 the moment of the external force about a fixed
point becomes
(e) (e)
X
N (e) = (ri′ + R) × Fi = N0 + R × F (e) . (5.66)
i
34
By combining Equations 5.64 and 5.66 we obtain
dH0 (e)
= N0 , (5.67)
dt
which states that the time derivative of the angular momentum of motion about the centre of
mass is equal to the moment of the external force about the centre of mass.
5.2.3 Work
We now consider the work done by all the forces in moving the system from an initial configu-
ration 1 to a final configuration 2:
XZ 2 X Z 2 (e) XZ 2
W12 = Fi · dsi = Fi · dsi + Fji · dsi (5.68)
i 1 i 1 i,j 1
i6=j
As before, we can use the equation of motion to reduce the integral on the left-hand side to
XZ 2 XZ 2 XZ 2 1
Fi · dsi = mi v̇i · vi dt = d( mi vi2 ). (5.69)
i 1 i 1 i 1 2
Consequently, the work done can still be written as the difference of the final and initial kinetic
energies:
W12 = T2 − T1 ,
where the total kinetic energy, T, is
1X
T = mi vi2 . (5.70)
2 i
With the same reasoning already employed in calculating the angular momentum, the last term
vanishes, leaving
1 1X 2
T = M|Ṙ|2 + mi viR . (5.71)
2 2 i
Same as for the angular momentum, the kinetic energy also consists of two parts, the kinetic
energy as if all the mass was concentrated at the centre of mass plus the kinetic energy of the
motion about the centre of mass.
We now consider the right-hand side of Equation 5.68. In the special case that the external
forces are derivable in terms of the gradient of a scalar potential, the first term can be written
as
2
X Z 2 (e) XZ 2 X
Fi · dsi = − ∇i Vi · dsi = − Vi , (5.72)
i 1 i 1 i 1
where the subscript i on the del operator indicates that the derivatives are with respect to
the components of ri . If the internal forces are also conservative, the action reaction forces
35
Modelling and Control of Multibody Mechanical Systems
between the ith and jth particles, Fij and Fji , can be obtained from a potential function Vij .
Furthermore, to satisfy the strong law of action and reaction, Vij can be a function only of the
distance between the particles
Vij = Vij (|ri − rj |). (5.73)
Consequently, the two forces are equal and opposite
Fji = −∇i Vij = +∇j Vij = −Fij (5.74)
and lie along the line joining the two particles,
∇Vij (|ri − rj |) = (ri − rj )f (5.75)
where f is some scalar function.
With the assumption that Vij exists the second term on the right-hand side of Equation
5.68 can be written as a sum over pairs of particles, the terms for each pair being
Z 2
− (∇i Vij · dsi + ∇j Vij · dsj ).
1
The difference vector ri − rj is denoted as rij and ∇ij stands for the gradient with respect to
rij . Then
∇i Vij = ∇ij Vij = −∇j Vij ,
and
dsi − dsj = dri − drj = drij ,
so that the term for the ij pair has the form
Z
− ∇ij Vij · drij .
Summation over i and then j includes each term of a given pair twice and hence the need to
use the 1/2 factor.
Putting all the considerations together, it is clear that if the internal and external forces are
both derivable from potentials it is possible to define a total potential energy, V , of the system,
X 1X
V = Vi + Vij (5.77)
i
2 i,j
i6=j
such that the total energy T + V of the system is conserved, in analogy with Equation 5.19
which gives the conservation theorem for a single particle.
The second term on the right in Equation 5.77 will be called the internal potential energy
of the system. In general it varies as the system changes with time. Only for a particular
class of systems known as rigid bodies will the internal potential always be constant. Rigid
bodies are systems in which the distances rij between the particles remain fixed at all times
and therefore the vectors drij can only be perpendicular to the corresponding rij and therefore
to the Fij which obey the strong law of action reaction. Consequently in a rigid body the
internal forces do not do any work. The internal potential must remain constant and can be
completely disregarded in discussing the motion of the system.
36
eθ1
O i
er1
k
θ1
r1 eφ
el2
φ
m1 g l2
m2 g
5.2.4 Example 1
Consider the problem of a small mass m1 suspended from a rigid rod of negligible mass and a
second mass m2 suspended from a rigid rod of negligible mass onto the first mass as shown in
Figure 5.5. Assume the masses are free to move in a vertical plane under the action of gravity.
The distance between the first particle and point O and the distance between the two particles
are kept constant by the supporting rods.
↓
Two sets of polar unit vectors are employed to solve the problem. er1 and eθ1 rotate with
the position vector of the first particle, and el2 and eφ move with the first particle and rotate
with the connecting rod between the two masses. The system is described by two equations of
motion in terms of θ1 , φ. The position vectors of the two masses are
r1 = r1 er1 , (5.78)
r2 = r1 er1 + l2 el2 . (5.79)
By differentiation we get the velocities,
ṙ1 = r1 θ̇1 eθ1 , (5.80)
ṙ2 = r1 θ̇1 eθ1 + l2 φ̇eφ . (5.81)
The acceleration in the inertial reference frame is obtain by differentiating once more,
r̈1 = −r1 θ̇12 er1 + r1 θ̈1 eθ1 , (5.82)
r̈2 = −r1 θ̇12 er1 + r1 θ̈1 eθ1 − l2 φ̇2 el2 + l2 φ̈eφ . (5.83)
37
Modelling and Control of Multibody Mechanical Systems
Newton’s law can be applied to each particle individually. For the first particle we get,
m1 (−r1 θ̇12 er1 + r1 θ̈1 eθ1 ) = −F1 er1 + F2 el2 + m1 gk, (5.84)
where F1 and F2 are the magnitudes of the unknown forces applied from the connecting rods.
Newton’s law applied to the second particle gives,
m2 (−r1 θ̇12 er1 + r1 θ̈1 eθ1 − l2 φ̇2 el2 + l2 φ̈eφ ) = −F2 el2 + m2 gk. (5.85)
The unit vectors are related between them in the following way,
Equations 5.86, 5.87, 5.90 can be substituted in Equation 5.85 in order to get the two
equations in the two directions el2 , eφ ,
and
F2 = m2 (r1 θ̇12 cos(φ − θ1 ) − r1 θ̈1 sin(φ − θ1 ) + l2 φ̇2 + g cos φ). (5.92)
The first equation is one of the equations of motion whereas the second one provides one of the
unknown forces. Similarly, Equations 5.88, 5.89 are substituted into Equation 5.84 and from
that we obtain,
This is the second of the equations of motion and we have thus completed the problem. The
unknown force F1 can be found by multiplying Equation 5.93 by cos(φ − θ1 ) and Equation 5.94
by sin(φ − θ1 ) and adding them together, followed by a substitution of Equation 5.91. The
result is
F1 sin(φ − θ1 ) = −m1 l2 φ̈. (5.97)
38
eθ
O i
eR
k
θ1 R
θ
r1
r1′ eφ
φ
m1 g er′
A
r2′
m2 g
5.2.5 Example 2
Consider the problem of a small mass m1 suspended from a coiled spring of negligible mass and
a second mass m2 suspended from a rigid rod of negligible mass onto the first mass as shown in
Figure 5.6. Assume the masses are free to move in a vertical plane under the action of gravity
and the spring. The distance between the two particles is kept constant by the supporting rod
and therefore the centre of mass A remains at fixed distances r1′ and r2′ from the two particles.
↓
Two sets of polar unit vectors are employed to solve the problem. eR and eθ rotate with
the position vector of the centre of mass, R, and er′ and eφ move with centre of mass and
rotate with the connecting rod between the masses. The motion of the system is described by
three equations in terms of R, θ, φ. Therefore,
R = ReR , (5.98)
r1′ = −r1′ er′ , (5.99)
r2′ = r2′ er′ , (5.100)
r1 = R + r1′ = ReR − r1′ er′ , (5.101)
r2 = R + r2′ = ReR + r2′ er′ , (5.102)
where r1 , r2 are the position vectors of the two particles in the inertial reference frame.
By combining Equation 5.51 with Equations 5.101 and 5.102 we get
39
Modelling and Control of Multibody Mechanical Systems
or
−m1 r1′ + m2 r2′ = 0. (5.104)
We take the total length of the supporting link between the two masses to be l2 ,
and therefore
m2
r1′ = l2 , (5.106)
m1 + m2
m1
r2′ = l2 . (5.107)
m1 + m2
The equations of motion of the centre of mass can be found by making use of Equation 5.52,
d2 R r1
(m1 + m2 ) 2
= (m1 + m2 )gk − c(r1 − r10 ) , (5.108)
dt r1
where r10 is the unstretched length of the spring. If we express everything in terms of the polar
vectors eR , eθ we get,
(m1 + m2 ) (R̈ − Rθ̇2 )eR + (Rθ̈ + 2Ṙθ̇)eθ = (5.109)
R − r1′ cos(φ − θ)
(m1 + m2 )g cos θ − c(r1 − r10 ) eR
r1
r1′ sin(φ − θ)
+ −(m1 + m2 )g sin θ + c(r1 − r10 ) eθ ,
r1
where we made use of Equation 4.23. Two (out of the three) equations of motion are therefore
given by,
2 R − r1′ cos(φ − θ)
R̈ − Rθ̇ − g cos θ + c(r1 − r10 ) = 0, (5.110)
(m1 + m2 )r1
and
r1′ sin(φ − θ)
Rθ̈ + 2Ṙθ̇ + g sin θ − c(r1 − r10 ) = 0. (5.111)
(m1 + m2 )r1
The length r1 is given by q
r1 = r1′2 + R2 − 2r1′ R cos(φ − θ). (5.112)
In order to obtain the third equation of motion we calculate the angular momentum of the
system about the centre of mass A as given by Equation 5.62 and substitute into Equation
5.67. The angular momentum about point A is,
X
H0 = mi ri′ × viR , (5.113)
i
where ri′ and viR are the position and velocity vectors of the particles relative to the centre of
mass. The rate of change of this angular momentum is,
X dviR
Ḣ0 = mi ri′ × . (5.114)
i
dt
40
The term dviR /dt = d2 ri′ /dt2 is the acceleration of the ith particle relative to the centre of
mass. This acceleration is given by,
dviR d′ vi′
= + 2Ω × vi′ + Ω × (Ω × ri′ ) + Ω̇ × ri′ , (5.115)
dt dt
where vi′ is the velocity of the particles in the local rotating reference frame—defined by the
unit vectors er′ , eφ ,—and Ω, Ω̇ are the angular velocity and angular acceleration of the local
frame relative to the inertial reference frame. The first two terms of the above equation depend
on the velocities vi′ which are zero. These velocities are zero because the two particles are not
getting closer or farther from the centre of mass due to the rigid connection. The rate of change
of angular momentum is therefore,
X
Ḣ0 = mi ri′ × Ω × (Ω × ri′ ) + Ω̇ × ri′ . (5.116)
i
The result of vector triple product Ω × (Ω × ri′ ) is a vector in the direction of ri′ and thus the
term ri′ × (Ω × (Ω × ri′ )) is zero. The rate of change of angular momentum reduces to
X
Ḣ0 = mi ri′2 Ω̇ − (ri′ · Ω̇)ri′ , (5.117)
i
where use of the vector triple product equation (4.11) was made. The term ri′ · Ω̇ is zero since
ri′ and Ω̇ are perpendicular to each other (Ω̇ is colinear with Ω since in the present example
the angular velocity can only change in magnitude and not in direction because we study the
motion of particles that remain in a particular plane). The rate of change of angular momentum
is, X
Ḣ0 = mi ri′2 φ̈j, (5.118)
i
where we made use of the fact that Ω = φ̇j (j is the rectangular unit vector coming out of the
page).
The sum of the external moments about the centre of mass A is given by,
′ r1
N0 = r1 × −c(r1 − r10 ) + r1′ × (m1 gk) + r2′ × (m2 gk). (5.119)
r1
If we substitute into Equation 5.67 we get,
X
′ R + r1′
′2
mi ri φ̈j = r1 × −c(r1 − r10 ) + m1 gr1′ × k + m2 gr2′ × k. (5.120)
i
r 1
or,
X c(r1 − r10 )Rr1′ sin(φ − θ)
mi ri′2 φ̈j = − j + m1 gr1′ sin φj − m2 gr2′ sin φj. (5.121)
i
r1
The last two terms on the right vanish because of Equation 5.104 and the expression reduces
to,
c(r1 − r10 )Rr1′ sin(φ − θ)
(m1 r1′2 + m2 r2′2 )φ̈ + = 0. (5.122)
r1
With this equation we complete the derivation of the equations of motion for this system. Note
that the unknown force between the two particles which keeps them at a fixed distance from
each other does not come into our calculation since it is an internal force.
↑
41
Modelling and Control of Multibody Mechanical Systems
Systems with constraints
42
are generalised so that the old variables are expressed as arbitrary functions of the new variables.
The number of variables is not 3 but 3N, since the position of the mechanical system requires
3N coordinates for its characterisation. The general form of such a coordinate transformation
appears as follows:
x1 = f1 (q1 , . . . , q3N ),
..
. (6.4)
zN = f3N (q1 , . . . , q3N ).
We can prescribe these functions, f1 , . . . , f3N , in any way we wish and thus shift the original
problem of determining the xi , yi , zi as functions of t to the new problem of determining the
q1 , . . . , q3N as functions of t. With proper skill in choosing the right coordinates, we may solve
the new problem more easily than the original one. The flexibility of the reference system
makes it possible to choose coordinates which are particularly suitable for the given problem.
The advantage of generalised coordinates is even more obvious if mechanical systems with
given kinematical constraints are considered. Such constraints are given as certain functional
relations between the coordinates. We have already met one type of system involving con-
straints, namely rigid bodies, where the constraints on the motions of the particles keep the
distances rij unchanged. As a simpler example consider two atoms that form a molecule, the
distance between the two atoms being determined by strong forces which are in equilibrium
at that distance. Dynamically this system can be considered as composed of two mass parti-
cles with coordinates x1 , y1 , z1 and x2 , y2 , z2 which are kept at a constant distance a from one
another. This implies the condition
independent parameters
q1 , q2 , . . . , qn , (6.7)
43
Modelling and Control of Multibody Mechanical Systems
in such a way that the rectangular coordinates of all the particles are expressible as functions
of the variables in 6.7:
x1 = f1 (q1 , . . . , qn ),
..
. (6.8)
zN = f3N (q1 , . . . , qn ).
The number n is a characteristic constant of the given mechanical system which cannot be
altered. We express the fact that n parameters are necessary and sufficient for a unique char-
acterisation of the configuration of the system by saying that it has “n degrees of freedom.”
Moreover, we call the n parameters q1 , q2 , . . . , qn the “generalised coordinates” of the sys-
tem. The number N of particles which compose the mechanical system is immaterial for the
analytical treatment, as are also the coordinates of these particles. It is the generalised coordi-
nates q1 , q2 , . . . , qn and certain basic functions of them which are of importance. A rigid body
may be composed of an infinity of particles, yet for the mechanical treatment it is a system of
not more than 6 independent coordinates.
The generalised coordinates q1 , q2 , . . . , qn may or may not have a geometrical significance.
However, it is necessary that the functions 6.8 be finite, single valued, continuous and differen-
tiable and invertible.
It is not always advisable to eliminate all the kinematical conditions of a problem by in-
troducing appropriate generalised coordinates. We sometimes prefer to eliminate only some of
the kinematical conditions, and to leave the others as additional restricting conditions. Hence
the general problem of mechanical systems with kinematical constraints presents itself in the
following form. We have the equations 6.8 and in addition m′ equations of the form:
φi (q1 , q2 , . . . , qn ) = 0, (i = 1, . . . , m′ ). (6.9)
n′ = n − m′ .
Some examples of systems with various degrees of freedom are given in Table 6.1.
44
Figure 6.1: One degree-of-freedom piston
θ
l
θ1
l1
l2
θ2
45
Modelling and Control of Multibody Mechanical Systems
6.2 The configuration space
We now introduce the concept of the configuration space. If we associate with the three numbers
x, y, z a point in a three-dimensional space, there is no reason why we should not do the same
with the n numbers q1 , q2 , . . . , qn , considering these as the rectangular coordinates of a “point”
P in an n-dimensional space. Similarly, if we can associate with the equations
x = f (t),
y = g(t), (6.10)
z = h(t),
the idea of a “curve” and the motion of a point along that curve, there is no reason why we
should not do the same with the corresponding equations
q1 = q1 (t),
..
. (6.11)
qn = qn (t).
These equations represent the solution of a dynamical problem. In the associated geomet-
rical picture, we have a point P of an n-dimensional space which moves along a given curve of
that space.
This geometrical picture is a great aid to our thinking. No matter how numerous the
particles constituting a given mechanical system may be, or how complicated the relations
existing between them, the entire mechanical system is pictured as a single point of a many-
dimensional space, called the “configuration space.” For example, the position of a rigid body–
with all the infinity of mass particles which form it–is symbolised as a single point of a 6-
dimensional space. This 6-dimensional space has, to be sure, nothing to do with the physical
reality of the rigid body. It is merely correlated to the rigid body in the sense of a one-to-one
correspondence. The various positions of the rigid body are “mapped” as various points of
the 6-dimensional space, and conversely, the various points of a 6-dimensional space can be
physically interpreted as various positions of a rigid body.
The concept of the configuration space links intimately with the concepts and methods of
Riemannian geometry, in which the motion of an arbitrary mechanical system can be studied as
the motion of a free particle in a certain n-dimensional space of definite Riemannian structure.
The mechanical problem is thus translated into a problem of differential geometry. These
concepts are beyond the scope of this course and will not be studied further here.
This, however, is an oversimplification since it may be necessary to take into account the
constraints that limit the motion of the system. We have already introduced the idea of
kinematical constraints earlier and have shown that these amount to certain functional relations
46
between the generalised coordinates. Apart from the obvious example of the rigid body, other
examples of constrained systems are easily provided. The beads of an abacus are constrained
to one-dimensional motion by the supporting wires. Gas molecules within a container are
constrained by the walls of the vessel to move only inside the container. A particle placed on
the surface of a solid sphere is subject to the constraint that it can move only on the surface
or in the region exterior to the sphere.
Constraints introduce two types of difficulties in the solution of mechanical problems. First,
the coordinates ri are no longer all independent, since they are connected by the equations of
constraint; hence the equations of motion 5.49 are not all independent. Second, the forces of
constraint, (e.g., the force that the wire exerts on the bead, or the wall on the gas particle), are
not furnished a priori. They are among the unknowns of the problem and must be obtained
from the solution we seek. Indeed, imposing constraints on the system is simply another method
of stating that there are forces present in the problem that cannot be specified directly but are
known rather in terms of their effect on the motion of the system.
We can classify constraints in various ways. If a system is subject to kinematical conditions
given as a function of the generalised coordinates in the form
f (q1 , . . . , qn ) = 0, (6.13)
then the constraint is said to be holonomic. Perhaps the simplest example of holonomic
constraints is the rigid body, where the constraints are expressed by equations of the form
given in Equation 6.5. A particle constrained to move along any curve or on a given surface is
another obvious example of a holonomic constraint, with the equations defining the curve or
surface acting as the equations of constraint.
Constraints not expressible in this fashion are called nonholonomic. The walls of a gas
container constitute a nonholonomic constraint. The constraint involved in the example of a
particle placed on the surface of a sphere is also nonholonomic, for it can be expressed as an
inequality
r 2 − a2 ≥ 0,
which is not in the form of Equation 6.13 (where a is the radius of the sphere). Thus, in a
gravitational field a particle placed on the top of the sphere will slide down the surface part of
the way but will eventually fall off.
Occasionally, conditions of a more general nature occur which can only be stated in in-
finitesimal form. A characteristic example is the rolling of an object on a table. Consider a
disk rolling on the horizontal xy plane constrained to move so that the plane of the disk is
always vertical. The disk, moving freely in space, has 6 degrees of freedom. Since the disk
rests on the surface, the height of the centre is a given constant, which reduces the number
of degrees of freedom to five. Furthermore, the imposed constraint that the disk be vertical
reduces the degrees of freedom to four. We may characterise the position of the disk by the
two rectangular coordinates x and y of its centre, plus two angles φ and θ, the angle of rotation
about the axis of the disk and the angle between the axis of the disk and, say, the x axis (see
Figure 6.4). If the disk can slide along the surface, it can make use of all of its four degrees of
freedom. However, if it is confined to rolling, the point of contact has to be momentarily at
rest. Pure rolling cuts down the degrees of freedom to two. If the path of the point of contact
is determined by giving x and y as functions of the time t, the condition of rolling determines
the position of the disk at any time uniquely. This would suggest that perhaps the angles φ
and θ can be given as functions of x and y. This, however is not possible. To see this consider
the following analysis.
47
Modelling and Control of Multibody Mechanical Systems
x
φ
z
θ v
As a result of the constraint the velocity of the disk, v, has a magnitude proportional to φ̇,
v = aφ̇,
where a is the radius of the disk, and its direction is perpendicular to the axis of the disk:
ẋ = v cos θ,
ẏ = v sin θ.
Thus the differentials of θ and φ are expressible in terms of the differentials of x and y, but these
differential relations are not integrable. They cannot be changed into finite relations between
the coordinates as in Equation 6.13 and are therefore nonholonomic. Indeed, if the disk starts
from a certain position and rolls along two different paths so that the final position of the point
of contact is the same in each case, the two final positions of the disk are rotated relative to
each other. If the angles φ, θ could be given as functions of x and y, the final positions of the
disk would be the same in both cases.
Now consider the holonomic kinematical condition as given in Equation 6.13. From this it
follows by differentiation that
∂f ∂f
dq1 + . . . + dqn = 0. (6.15)
∂q1 ∂qn
However, if we start with a differential relation of the form
(with coefficients Ak which are given functions of the qk ) this relation is convertible into the
form 6.13 only if certain integrability conditions are satisfied. The only exception is the case
n = 2, because a differential relation between two variables is always integrable–a familiar
example is the two-dimensional motion of a circle rolling on an inclined plane.
Holonomic kinematical conditions can be attacked in two ways. If there are m equations
between n variables, we can eliminate m of these and reduce the problem to n − m independent
variables. Or, in the case of the analytical treatment, we can operate with a surplus number of
48
variables and retain the given relations as auxiliary conditions (method of Lagrange multipliers).
Problems involving holonomic constraints are always amenable to a formal solution.
However, there is no general way of attacking nonholonomic examples. There is a formal
procedure in special cases such as those when the conditions are stated as relations between
the differentials of the coordinates. A reduction in variables in such case is not possible and we
have to operate with more variables than the degrees of freedom of the system demand. Again,
this is achieved with the method of Lagrange multipliers that we will study at a later point.
If there is a finite relation between the variables q1 , q2 , q3 on account of the given condition, we
must have
∂B1 ∂B2
= . (6.18)
∂q2 ∂q1
This relation has to be reinterpreted in view of the fact that the coefficients B1 and B2 may
depend implicitly on q3 . We thus replace 6.18 by
∂B1 ∂B1 ∂q3 ∂B2 ∂B2 ∂q3
+ = + ; (6.19)
∂q2 ∂q3 ∂q2 ∂q1 ∂q3 ∂q1
the quantities ∂q3 /∂q1 and ∂q3 /∂q2 are given by 6.17, and we get
∂B1 ∂B1 ∂B2 ∂B2
+ B2 = + B1 . (6.20)
∂q2 ∂q3 ∂q1 ∂q3
This relation has to be satisfied for all values of q1 and q2 , which means that it must be an
identity. If that is the case, the relation 6.17 is integrable, otherwise not. However, it may
happen that q3 does not drop out from the resulting equation and can be expressed in terms
of q1 and q2 . With this solution for q3 we then go back to 6.17 and test whether the partial
derivatives of q3 with respect to q1 and q2 agree with B1 and B2 or not. If they do, we have
proved that the given differential relation is holonomic and we have replaced it by a finite
relation between the coordinates. In the case of more than two independent variables all the
integrability conditions
∂Bi ∂Bk
= , (6.21)
∂qk ∂qi
have to be tested in a similar manner. If a whole system of independent differential relations
is given, we first solve them for a suitable set of dependent differentials and then proceed in a
similar way with testing the integrability conditions.
Example
Investigate the integrability of the following differential relation:
49
Modelling and Control of Multibody Mechanical Systems
↓ (This
condition is holonomic and can be replaced by the finite relation
z = x2 − xy + y 2.) (6.23)
Xi , Y i , Z i . (6.24)
These forces come either from an external field or from the mutual interaction of the particles,
but they are not to include the forces which maintain the given kinematical constraints.
Let us change the coordinates of each one of the particles by arbitrary infinitesimal amounts
dxi , dyi , dzi . The total work1 of all the impressed forces is
N
X
dW = (Xi dxi + Yi dyi + Zi dzi ). (6.25)
i=1
If now the rectangular coordinates are expressed as functions of the generalised coordinates
q1 , . . . , qn according to Equations 6.8, which can be rewritten as
x1 = fx1 (q1 , . . . , qn ),
y1 = fy1 (q1 , . . . , qn ),
..
. (6.26)
zN = fzN (q1 , . . . , qn ),
the differentials dxi , dyi , dzi can be expressed in terms of the dqi , and the infinitesimal work
dW comes out as a linear differential form of the variables qi :
where
N
X ∂fxi ∂fyi ∂fzi
F1 = (Xi + Yi + Zi ),
i=1
∂q1 ∂q1 ∂q1
..
. (6.28)
N
X ∂fxi ∂fyi ∂fzi
Fn = (Xi + Yi + Zi ).
i=1
∂qn ∂qn ∂qn
1
The dash above the symbol is used to denote non-integrable differentials which cannot be considered as the
infinitesimal change of something.
50
In the analytical treatment the original force-components are of no importance, but only the
coefficients
F1 , F2 , . . . , Fn (6.29)
of the differential form 6.27. They take the place of the components 6.24 of the single forces
which act on each particle. The quantities 6.29 determine completely the dynamical action
of all the forces. These quantities form the components of a vector, but it is a vector of
the n-dimensional configuration space and not one of ordinary space. We call this vector the
“generalised force” with its components being the quantities Fi .
In general the forces acting on a mechanical system fall automatically into two categories.
It is possible that all we can say about the work dW is that it is a non-integrable differential
form. But it is also possible that dW turns out to be a true differential of a certain function,
in which case the dash above dW can be omitted. That function is called the “work function”
and is frequently denoted by the letter U. Hence we put
dW = dU, (6.30)
where
U = U(q1 , q2 , . . . , qn ). (6.31)
We thus have
X X ∂U
Fi dqi = dqi , (6.32)
∂qi
which gives
∂U
Fi =
. (6.33)
∂qi
The present practice is to use the negative work function and call this quantity V:
V = −U. (6.34)
The advantage of this change of sign is that in view of the law of conservation of energy V
can be interpreted as the “potential energy” of the system, as we have seen previously. The
Equation 6.33 can now be rewritten in the form
∂V
Fi = − , (6.35)
∂qi
where V is a function of the position coordinates:
V = V (q1 , q2 , . . . , qn ). (6.36)
Forces in this category are remarkable for two reasons. Firstly, they satisfy the law of the
conservation of energy. Secondly, although the generalised force has n components, they can
be calculated from a single scalar function, U.
The definition of a work function on the basis of the Equation 6.31 is too restricted.
We have forces in nature which are derivable from a time-dependent work function U =
U(q1 , q2 , . . . , qn , t). Then the Equation 6.30 still holds, with the understanding that in
forming the differential dU the time t is considered as a constant. The Equations 6.32 and 6.33
remain valid, but the conservation of energy is lost. The generalised force possesses a work
function without being conservative. An electrically charged particle revolving in a cyclotron2
2
A cyclotron is a type of particle accelerator.
51
Modelling and Control of Multibody Mechanical Systems
returns to the same point with increased kinetic energy, so that the energy is not conserved.
This is not because the work function does not exist, but because the work function is time-
dependent. On the other hand, a generalised force may have no work function and still satisfy
the conservation of energy, as for example the force which maintains rolling.
It is desirable to have a distinctive name for forces which are derivable from a scalar quantity,
irrespective of whether they are conservative or not. As in some of the literature we will adopt
the name “monogenic” (which means “single-generated”) as a distinguishing name for this
category of forces, while forces which are not derivable from a scalar function–such as friction–
will be called “polygenic.”
The work function associated with a monogenic force is in the most general case a function
of the coordinates and the velocities:
For example , the electromagnetic force of Lorentz, which acts on a charged particle in the
presence of an electric and a magnetic field, is derivable from a work function of this kind.
As the differential equation of Euler-Lagrange shows (we will study this later), the connection
between force and the work function is now given by the equation
∂U d ∂U
Fi = − , (6.38)
∂qi dt ∂ q̇i
which is a generalisation of the simpler relation 6.33.
f (q1 , . . . , t) = 0. (6.39)
This is for example, the situation if a mass point moves on a surface which itself is moving
according to a given law. The equation of that surface takes the form
f (x, y, z, t) = 0. (6.40)
Or consider a pendulum whose length is being constantly changed by pulling the thread in
a definite manner. This again amounts to an auxiliary condition which contains the time
explicitly.
The terms “rheonomic” and “scleronomic” are used to describe kinematic constraints
that do or do not involve the time t. If rheonomic conditions are present among the given
kinematical constraints, the elimination of these conditions by a proper choice of generalised
coordinates will have the consequence that the equations 6.8 will contain t explicitly:
x1 = f1 (q1 , . . . , qn , t),
..
. (6.41)
zN = f3N (q1 , . . . , qn , t).
52
A similar situation arises even without time-dependent kinematical conditions, when the coor-
dinates chosen belong to a reference system which is in motion.
Differentiation of the Equations 6.41 with respect to t yields:
∂f1 ∂f1 ∂f1
ẋ1 = q̇1 + . . . + q̇n + ,
∂q1 ∂qn ∂t
..
. (6.42)
∂f3N ∂f3N ∂f3N
żN = q̇1 + . . . + q̇n + ,
∂q1 ∂qn ∂t
If these expressions are used to calculate the kinetic energy, this will not come out as a purely
quadratic form of the generalised velocities, q̇i ; we obtain additional terms which are linear in
the velocities, and others which are independent of the velocities.
An example of a system with a scleronomic constraint is a bead sliding on a rigid curved
wire fixed in space; if the wire is moving in some prescribed fashion the constraint is rheonomic.
Note that if the wire moves, say, as a reaction to the bead’s motion, then the time dependence
of the constraint enters in the equation of constraint only through the coordinates of the curved
wire, which are now part of the system coordinates. The overall constraint is then scleronomic.
The time may also enter explicitly in the work function U. In this case the system becomes
rheonomic again. The essential difference between a scleronomic and a rheonomic system is
that in a scleronomic system the total energy remains constant during the motion3 . The total
energy is equal to the sum of the kinetic and the potential energy. Rheonomic systems do not
satisfy any conservation law.
3
Assuming that all the external forces are furnished by the work function, i.e there are no external forces of
polygenic nature.
53
Modelling and Control of Multibody Mechanical Systems
Kinematics of rigid body motion
A rigid body, as we have seen previously, may be defined in mechanics as a system of particles
such that the distances between the particles do not vary. This condition can, of course, be
satisfied only approximately by systems which actually exist in nature. The majority of solid
bodies, however, change so little in shape and size under ordinary conditions that these changes
may be entirely neglected in considering the laws of motion of the body as a whole.
x′
x
z ′
y′
z
y
Figure 7.1: x, y, z axes represent an external set of reference axes; x′ , y ′, z ′ are fixed in the
rigid body.
The configuration of a rigid body can be completely specified by locating a Cartesian set
of coordinates fixed in the rigid body (the primed axes shown in Figure 7.1) relative to the
54
x′
i′
i x
′
j j
k′ k
z′ z y′
y
Figure 7.2: Direction cosines of the body set of axes relative to an external set of axes.
coordinate axes of the external space. Clearly three of the coordinates are needed to specify
the coordinates of the origin of this “body fixed” set of axes. The remaining three coordinates
must then specify the orientation of the primed axes relative to a coordinate system parallel to
the external axes, but with the same origin as the primed axes.
α1 = cos(i′ , i) = i′ · i,
α2 = cos(i′ , j) = i′ · j, (7.1)
α3 = cos(i′ , k) = i′ · k.
or
i′ = α1 i + α2 j + α3 k. (7.2)
Similarly the direction cosines of the y ′ and z ′ axes with x, y, z may be designated by β1 , β2 ,
β3 , and γ1 , γ2 , γ3 and these will be the components of j ′ and k′ in the unprimed reference
frame:
j ′ = β1 i + β2 j + β3 k, (7.3)
k′ = γ1 i + γ2 j + γ3 k. (7.4)
These sets of nine direction cosines then completely specify the orientation of the x′ , y ′ , z ′ axes
relative to the x, y, z set. One can equally well invert the process, and use the direction cosines
to express the i, j, k unit vectors in terms of their components along the primed axes. Thus
we can write
i = (i · i′ )i′ + (i · j ′ )j ′ + (i · k′ )k′ ,
55
Modelling and Control of Multibody Mechanical Systems
or
i = α1 i′ + β1 j ′ + γ1 k′ , (7.5)
with analogous equations for j and k.
The direction cosines also furnish directly the relations between the coordinates of a given
point in one system and the coordinates in the other system. Thus, the coordinates of a point
in a given reference frame are the components of the position vector, r, along the axes of the
system. The x′ , y ′, z ′ coordinates are then given in terms of x, y, z by
x′ = (r · i′ ), y ′ = (r · j ′ ), z ′ = (r · k′ ),
or
x′ = α1 x + α2 y + α3 z,
y ′ = β1 x + β2 y + β3 z, (7.6)
z ′ = γ1 x + γ2 y + γ3 z.
The set of nine direction cosines thus completely spells out the transformation between the two
coordinate systems.
If the primed axes are taken as fixed in the body, then the nine direction cosines will be
functions of time as the body changes its orientation in the course of the motion. In this sense
the α’s, β’s and γ’s can be considered as coordinates describing the instantaneous orientation
of the body, relative to a coordinate system fixed in space but with origin in common with
the body system. But, clearly, they are not independent coordinates, because there are nine of
them and it has been shown that only three coordinates are needed to specify an orientation.
The connections between the direction cosines arise from the fact that the basis vectors in
both coordinate systems are orthogonal to each other and have unit magnitude,
i · j = j · k = k · i = 0,
and
i · i = j · j = k · k = 1, (7.7)
with similar relations for i′ , j ′ and k′ . We can obtain the conditions satisfied by the nine
coefficients by forming all possible dot products among the three equations for i, j, k in terms
of i′ , j ′ , k′ , making use of the Equations 7.7:
αl αm + βl βm + γl γm = 0, l, m = 1, 2, 3; l 6= m,
αl2 + βl2 + γl2 = 1, l = 1, 2, 3. (7.8)
56
is called the matrix of transformation, and will be denoted by a capital letter A. Equation 7.6
in matrix form becomes ′
x α1 α2 α3 x
y ′ = β1 β2 β3 y . (7.10)
z′ γ1 γ2 γ3 z
To make these formal considerations more meaningful consider the simple example of motion
in a plane, so that we are restricted to two dimensional coordinate systems. The transformation
matrix reduces to the form
α1 α2
. (7.11)
β1 β2
The four matrix elements are connected by three orthogonality conditions
α1 α2 + β1 β2 = 0, (7.12)
α12 + β12 = 1, (7.13)
α22 + β22 = 1, (7.14)
and therefore only one independent parameter is needed to specify the transformation. But this
conclusion is not surprising. A two-dimensional transformation from one Cartesian coordinate
system to another corresponds to a rotation of the axes in the plane, see Figure 7.3, and such
x′
φ
x
z′
z
Figure 7.3: Rotation of the coordinate axes; equivalent to two-dimensional orthogonal trans-
formation.
a rotation can be specified completely by only one quantity, the rotation angle φ. Expressed in
terms of this single parameter, the transformation equations become
x′ = x cos φ − z sin φ,
z ′ = x sin φ + z cos φ.
r ′ = Ar, (7.16)
57
Modelling and Control of Multibody Mechanical Systems
which is to be read: The matrix A operating on the components of a vector in the unprimed
system yields the components of the vector in the primed system. It is to be emphasised
that A acts on the coordinate systems only, the vector is unchanged, and we ask merely for
its components in two different coordinate frames. In two dimensions the transformation of
coordinates, it has been seen, is simply a rotation, and A is then identical with the rotation
operator in a plane.
Despite this, A can also be thought of as an operator acting on the vector r, changing
it to a different vector r ′ with both vectors expressed in the same coordinate system. Thus,
in two dimensions, instead of rotating the coordinate system counterclockwise one can rotate
the vector r clockwise by an angle φ to a new vector r ′ . The components of the new vector
will then be related to the components of the old by the same Equations 7.6 that describe the
transformation of coordinates. The interpretation as an operator acting on the coordinates is
the more pertinent one when using the orthogonal transformation to specify the orientation of
a rigid body. When the transformation is taken as acting only on the coordinate system we
speak of the passive role of the transformation. In the active sense the transformation is looked
on as changing the vector or other physical quantity.
x
φ r
r′
Figure 7.4: Rotation of a vector leaving the coordinate system unchanged; another interpreta-
tion of an orthogonal transformation.
BA 6= AB. (7.18)
The inverse transformation from body coordinates to space axes is given by the inverse of the
operator matrix which for an orthogonal transformation is the same as the transpose of the
matrix,
A−1 = AT . (7.19)
58
roll
φ x′
y ′ pitch
θ
ψ
z ′
yaw
to describe the motion of rigid bodies it will therefore be necessary to seek three independent
parameters specifying the orientation of a rigid body in such a manner that the corresponding
matrix of transformation has a determinant of +1. The determinant of an orthogonal trans-
formation matrix has one of two possible values, +1 or −1. A value of −1 would result in a
transformation that inverts (or reflects) the coordinate axes which is an operation that never
corresponds to a physical displacement of a rigid body, hence the requirement for a determi-
nant of +1. Only when such generalised coordinates have been found can one proceed with the
derivation of the equations of motion. The most common and useful such set of parameters are
the Euler angles.
One can carry out the transformation from a given Cartesian coordinate system to another
by means of three successive rotations performed in a specific sequence. The Euler angles are
then defined as the three successive angles of rotations. Within limits, the choice of rotation
angles is arbitrary, as is the sequence of rotation. A number of conventions is possible in
defining the Euler angles and we will use here the one that is commonly used in engineering
applications relating to the orientation of moving vehicles such as cars, aircraft and satellites.
The sequence employed is started by rotating the initial system of axes, fixed in the vehicle,
by an angle ψ clockwise about the vertical axis (looking towards the positive direction of the
axis) and gives the yaw or heading angle. In the second stage the intermediate axes are rotated
clockwise by an angle θ about a perpendicular axis fixed in the vehicle and along the lateral
direction to give the pitch or attitude angle. Finally the third angle is one of rotation by an
angle φ about the longitudinal direction of the vehicle and it is the roll or bank angle.
The Euler angles θ, φ, ψ thus completely specify the orientation of the x′ y ′z ′ system, given
by the roll, pitch and yaw axes in Figure 7.5, relative to the initial (unrotated) xyz system and
can therefore act as the three needed generalised coordinates.
The elements of the complete transformation A can be obtained by writing the matrix as
the triple product of the separate rotations, each of which has a relatively simple matrix form.
The matrix of the complete transformation
r ′ = Ar,
is the product of the successive matrices,
A = BCD.
Now the D transformation is a rotation about z, and hence has a matrix of the form
cos ψ sin ψ 0
D = − sin ψ cos ψ 0 . (7.20)
0 0 1
59
Modelling and Control of Multibody Mechanical Systems
The C transformation corresponds to a rotation about an intermediate y, with the matrix
cos θ 0 − sin θ
C= 0 1 0 . (7.21)
sin θ 0 cos θ
and finally B is a rotation about an intermediate x and therefore has the form
1 0 0
B = 0 cos φ sin φ . (7.22)
0 − sin φ cos φ
r = A−1 r ′ ,
v = V + Ω × r′, (7.25)
where v = dr/dt and V = dR/dt. The vector V is the velocity of the origin O on the body,
and is also the translational velocity of the body. The vector Ω is called the angular velocity of
the rotation of the body. Its direction is along the axis of rotation. Thus the velocity v of any
point in the body relative to the external system of coordinates can be expressed in terms of
the translational velocity of the body and its angular velocity of rotation.
60
P x′
′
r
O
r
R
x
z′ y′
z
y
Figure 7.6: A point P in a rigid body represented in an external set of reference axes and axes
fixed in the rigid body.
Let us now assume that the system of coordinates fixed in the body is such that the origin
is not at point O, but at some point O1 at a distance d from O. Let the velocity of O1 be V1 ,
and the angular velocity of the new system of coordinates be Ω1 . We again consider the same
point P in the body, and denote by r1′ its position vector with respect to O1 . Then r ′ = r1′ + d,
and substitution in Equation 7.25 gives v = V + Ω × d + Ω × r1′ . The definition of V1 and Ω1
shows that v = V1 + Ω1 × r1′ . Hence it follows that
V1 = V + Ω × d, (7.26)
Ω1 = Ω. (7.27)
The second of these equations is very important. We see that the angular velocity of rotation,
at any instant, of a system of coordinates fixed in the body is independent of the particular
system chosen. All such systems rotate with angular velocities Ω which are equal in magnitude
and parallel in direction. This enables us to call Ω the angular velocity of the body. The velocity
of the translational motion, however, does not have this “absolute” property.
It can be seen from Equation 7.26 that, if V and Ω are, at any given instant, perpendicular
for some choice of the origin O, then V ′ and Ω′ are perpendicular for any other origin O ′.
Equation 7.25 shows that in this case the velocities v of all points in the body are perpendicular
to Ω. It is then always possible to choose an origin O ′ whose velocity V ′ is zero, so that the
motion of the body at the instant considered is a pure rotation about an axis through O ′ . This
axis is called the instantaneous axis of rotation. O ′ may, of course, lie outside the body. In the
general case where V and Ω are not perpendicular, the origin may be chosen so as to make
V and Ω parallel, so that the motion consists (at the instant in question) of a rotation about
some axis together with a translation along that axis. In general both the magnitude and the
direction of Ω vary during the motion.
The body set of axes proves to be the most useful for discussing the equations of motion,
and we will now express the components of the angular velocity vector Ω along the moving
axes x′ , y ′ , z ′ in terms of the Euler angles and their derivatives. To do this, we must find the
components along those axes of the angular velocities φ̇, θ̇, ψ̇.
The orthogonal transformations B, C, D of Section 7.3 may be used to furnish the compo-
nents of these vectors along any desired set of axes. Since Ωψ is parallel to the space z axis, its
components along the body axes are given by applying the complete orthogonal transformation
61
Modelling and Control of Multibody Mechanical Systems
A = BCD (Equation 7.23):
0
(Ωψ )′ = A 0 , (7.28)
ψ̇
which gives
The components of Ωθ with respect to the body axes are furnished by applying only the final
orthogonal transformation B (Equation 7.22):
1 0 0 0
′
(Ωθ ) = 0 cos φ sin φ
θ̇ , (7.32)
0 − sin φ cos φ 0
which gives
No transformation is necessary for the components of Ωφ , which lies along the x′ axis.
Collecting the components along each body axis, we have
Similar techniques may be used to express the components of Ω along the space set of axes
in terms of the Euler angles, or along body fixed axes in terms of Euler angles of different
conventions.
62
Dynamics of rigid bodies
In the previous section we have introduced all the necessary kinematical tools needed in the
discussion of rigid body motion. These tools will now be applied to obtain the dynamical
equations of motion of the rigid body in their most convenient form.
1
T = Mv 2 + T ′ (ψ, θ, φ),
2
as the sum of the kinetic energy of the entire body as if concentrated at the centre of mass,
plus the kinetic energy of motion about the centre of mass.
Often the potential energy can be similarly divided, each term involving only one of the
coordinate sets, either the translational or rotational. Thus, the potential energy in a uniform
gravitational field will depend only upon the Cartesian vertical coordinate of the centre of grav-
ity. The centre of gravity, of course, coincides with the centre of mass in a uniform gravitational
field.
It is of obvious importance to obtain expressions for the angular momentum and the kinetic
energy of the motion about some point fixed in the body. To do so we will make use of Equation
4.44 linking derivatives relative to a coordinate system fixed at some point in the rigid body.
We will also make use of the fact introduced previously that the rotation angle of a rigid body
displacement as also the instantaneous angular velocity vector, is independent of the choice of
origin of the body system of axes.
When a rigid body moves with one point stationary, the total angular momentum about
that point is
X
H= mi (ri × vi ), (8.1)
i
63
Modelling and Control of Multibody Mechanical Systems
where ri and vi are the position vector and velocity, respectively, of the ith particle relative to
the given point. Since ri is a fixed vector relative to the body, the velocity vi with respect to
the space set of axes arises solely from the rotational motion of the rigid body about the fixed
point. From Equation 4.44, vi is then
vi = Ω × ri . (8.2)
Hence Equation 8.1 can be written as
X
H= mi (ri × (Ω × ri )), (8.3)
i
We associate with the rigid body a Cartesian set of axes that moves with the body and has
i′ , j ′ , k′ unit vectors. We can then express the position vectors ri and the angular velocity Ω
in terms of these axes
ri = x′i i′ + yi′ j ′ + zi′ k′ , (8.5)
Ω = Ωx i′ + Ωy j ′ + Ωz k′ . (8.6)
By substituting into Equation 8.4 and expanding we obtain the components of H along the
body fixed axes,
!
X 2 2
X X
H= mi yi′ + zi′ Ωx − mi x′i yi′ Ωy − mi x′i zi′ Ωz i′ + . . . (8.7)
i i i
!
2 2
X X X
...+ − mi x′i yi′ Ωx + mi x′i + zi′ Ωy − mi yi′ zi′ Ωz j′ + . . .
i i i
!
2 2
X X X
...+ − mi x′i zi′ Ωx − mi yi′ zi′ Ωy + mi x′i + yi′ Ωz k′ .
i i i
It is obvious that each of the components of the angular momentum is a linear function of
all the components of the angular velocity. The angular momentum vector is related to the
angular velocity by a linear transformation. To emphasise the similarity of Equation 8.7 with
the equations of linear transformation we may write H as
H = Hx i′ + Hy j ′ + Hz k′ , (8.8)
and the three components of H as
Hx = Ixx Ωx + Ixy Ωy + Ixz Ωz , (8.9)
Hy = Iyx Ωx + Iyy Ωy + Iyz Ωz ,
Hz = Izx Ωx + Izy Ωy + Izz Ωz .
The nine coefficients Ixx , Ixy , etc., are the nine elements of the transformation matrix. The
diagonal elements are known as moments of inertia, and have the form illustrated by
2 2
X
Ixx = mi (yi′ + zi′ ), (8.10)
i
64
while the off-diagonal elements are designated as products of inertia, a typical one being
X
Ixy = − mi x′i yi′ . (8.11)
i
where the symbol I stands for the operator whose matrix elements are the inertia coefficients
appearing in 8.9 and is known as the inertia matrix or inertia tensor. Thus,
Ixx Ixy Ixz Ωx
H = Iyx Iyy Iyz Ωy , (8.13)
Izx Izy Izz Ωz
or
mi yi′ 2 + zi′ 2
P P P
i − i mi x′i yi′ − i mi x′i zi′ Ωx
′2 ′2
P P P
H= − i mi x′i yi′ i mP
i xi + zi − i mi yi′ zi′ Ωy . (8.14)
′2 ′2 Ωz
P P
− i mi x′i zi′ − i mi yi′ zi′ i mi xi + yi
The matrix elements appear in the form suitable if the rigid body is composed of discrete
particles. For continuous bodies the summation is replaced by a volume integration, with the
particle mass becoming mass density. Thus, the diagonal element Ixx appears as
Z Z
′2 ′2 2 2
Ixx = (y + z )dm = ρ(r)(y ′ + z ′ )dV. (8.15)
V
Similarly, an expression for all matrix elements can be stated for continuous bodies. Since the
coordinate system used in resolving the components of H is a system fixed in the body, the
various distances x′i , yi′ , zi′ are constant in time. The same is true for the matrix elements
(moments and products of inertia) which take particular values for the body involved, and
depend on the origin and orientation of the particular body set of axes in which they are
expressed.
In Equation 8.12 H has been expressed by an operator equation representing a linear
transformation. Of the two interpretations that have been given to the operator of a linear
transformation (see Section 7.2), it is clear that here I must be thought of as acting upon
the vector Ω, and not upon the coordinate system. The vectors H and Ω are two physically
different vectors, having different dimensions, and not merely the same vector expressed in two
different coordinate systems. Unlike the operator of rotation, I will have dimensions–mass
times length squared–and it is not restricted by any orthogonality conditions. Equation 8.12
is to be read as saying that the operator I acting upon the vector Ω results in the physically
new vector H.
65
Modelling and Control of Multibody Mechanical Systems
O x
h
z R
8.2 Example
Consider the cylinder with uniform density in Figure 8.1. The origin O is at the centre of mass
of the cylinder. Find the moments of inertia about the three axes x, y, z.
↓
Straightforward application of the Equation similar to 8.15 gives
Z Z Z
2 2 2 2
Izz = (x + y )dm = ρ (x + y )dV = ρ r 2 dV, (8.17)
V V
where r is the radial distance from the z axis. If we consider an infinitesimal volume element
given in cylindrical polar coordinates then Izz becomes
Z h Z 2π Z R
2 ρπhR4
r 3 drdθdz = . (8.18)
− h2 0 0 2
But the mass of the cylinder is given by
m = ρV = ρπR2 h, (8.19)
and therefore Izz becomes
1
Izz = mR2 . (8.20)
2
The moment of inertia about the x axes can be found similarly via the equation
Z Z Z
Ixx = (y + z )dm = ρ (y + z )dV = ρ (r 2 cos2 θ + z 2 )dV.
2 2 2 2
(8.21)
V V
By using a similar cylindrical polar volume element as above the volume integral becomes
Z h Z 2π Z R 2
R2
2
2 2 2 2 h
(r cos θ + z )rdrdθdz = ρπR h + , (8.22)
− h2 0 0 12 4
and by making use of the mass expression of the cylinder
1
Ixx = m(h2 + 3R2 ). (8.23)
12
66
Due to the symmetry of the cylinder Iyy is the same as Ixx ,
Iyy = Ixx . (8.24)
↑
The quantity summed over i will be recognised as the angular momentum of the body about
the origin, and in consequence the kinetic energy can be written in the form
Ω·H
T = . (8.28)
2
In matrix form, also substituting from Equation 8.12
1
T = ΩT IΩ, (8.29)
2
where ΩT is the transpose of Ω, i.e. a vector written in row instead of column form. Thus,
I xx I xy I xz Ωx
1
T = Ωx Ωy Ωz Iyx Iyy Iyz Ωy , (8.30)
2
Izx Izy Izz Ωz
or by multiplying out
1
T = (Ixx Ω2x + Iyy Ω2y + Izz Ω2z + 2Ixy Ωx Ωy + 2Ixz Ωx Ωz + 2Iyz Ωy Ωz ), (8.31)
2
where the symmetry of the off-diagonal terms of the inertia matrix has been used.
Let n be a unit vector in the direction of Ω so that Ω = Ωn. Then an alternative form for
the kinetic energy is
Ω2 T 1
T = n In = IΩ2 , (8.32)
2 2
where I is a scalar, defined by
X
I = nT In = mi (ri2 − (ri · n)2 ), (8.33)
i
and known as the moment of inertia about the axis of rotation. It can be shown to be the
same as the moment of inertia about an axis defined in the usual elementary discussions which
is defined as the sum, over the particles of the body, of the product of the particle mass and
the square of the perpendicular distance from the axis.
67
Modelling and Control of Multibody Mechanical Systems
8.4 Potential energy
The potential energy of a rigid body is the sum over all particles:
X
V = −mi ri · g (8.34)
i
where ri is the position vector of the ith particle measured from a fixed origin and g is the
constant vector for the acceleration of gravity. By Equation 5.51 defining the centre of mass,
this is equivalent to
V = −MR · g. (8.35)
The above equation says that in a constant gravitational field the potential energy is the same
as if the body were concentrated at the centre of mass.
centre
of mass ′
R ri
O ri
a b
Figure 8.2: The vectors involved in the relation between inertia tensors about different origins.
The inertia tensor depends upon the origin of the body set of axes. However, the inertia
tensor about some origin can be related to the inertia tensor when the origin is the centre of
mass. Let the vector from the given origin O to the centre of mass be R, and let the radii
vectors from O and the centre of mass to the ith particle be ri and ri′ , respectively. The three
vectors so defined are connected by the relation (see Figure 8.2)
ri = R + ri′ . (8.36)
By substituting ri into Equation 8.3 we get
X
H= mi (R + ri′ ) × (Ω × (R + ri′ )), (8.37)
i
or by expanding
X
H= mi (R × (Ω × R) + R × (Ω × ri′ ) + ri′ × (Ω × R) + ri′ × (Ω × ri′ )) . (8.38)
i
68
Use of the vector triple product identity gives
X
H= mi R2 Ω − (R · Ω)R + (R · ri′ )Ω − (R · Ω)ri′ + (ri′ · R)Ω − (ri′ · Ω)R + . . . (8.39)
i
2
...+ ri′ Ω − (ri′
· Ω)ri′ .
P
By the definition of the centre of mass, the terms that include the summation i mi ri′ vanish.
Hence H is given by
2
X X
H= mi ri′ Ω − mi (ri′ · Ω)ri′ + MR2 Ω − M(R · Ω)R, (8.40)
i i
where M is the total mass of the body. The first two terms are the inertia tensor I ′ relative to
the centre of mass multiplied with the angular velocity (similarly to Equations 8.12 and 8.13)
and the last two are the inertia tensor, relative to O, of a single particle of mass M located at
the centre of mass multiplied again by the angular velocity. If R and Ω are expressed in terms
of a body fixed set of axes with unit vectors i′ , j ′ , k′ as
R = Rx i′ + Ry j ′ + Rz k′ , (8.41)
Ω = Ωx i′ + Ωy j ′ + Ωz k′ ,
and substituted into Equation 8.40 we obtain
M(Ry2 + Rz2 ) −MRx Ry
Ωx −MRx Rz Ωx
H = I ′ Ωy + −MRx Ry M(Rx2 + Rz2 ) −MRy Rz Ωy , (8.42)
Ωz −MRx Rz −MRy Rz M(Rx2 + Ry2 ) Ωz
or
M(Ry2 + Rz2 ) −MRx Ry
−MRx Rz Ωx
H = I ′ + −MRx Ry M(Rx2 + Rz2 ) −MRy Rz Ωy . (8.43)
−MRx Rz −MRy Rz M(Rx2 + Ry2 ) Ωz
Therefore the inertia tensor about origin O is given by
M(Ry2 + Rz2 ) −MRx Ry
−MRx Rz
IO = I ′ + −MRx Ry M(Rx2 + Rz2 ) −MRy Rz , (8.44)
−MRx Rz −MRy Rz M(Rx2 + Ry2 )
where I ′ is the inertia tensor relative to the centre of mass (with its matrix components ex-
pressed in terms of the body fixed axes of course), and the second term is the inertia tensor,
relative to O, of a single particle of mass M located at the centre of mass. Thus, the inertia
tensor possesses a type of resolution, relative to the centre of mass, very similar to that found
for the linear and angular momentum and the kinetic energy in Section 5.2.
As a special case, it can be shown that the moment of inertia about some given axis is
related simply to the moment about a parallel axis through the centre of mass. The moment
of inertia about the axis a in Figure 8.2 can be expressed in terms of the moment of inertia
about the parallel axis b through the centre of mass as
Ia = Ib + M(|R × n|)2 , (8.45)
where n is a unit vector along axis a. The magnitude |R × n| is the perpendicular distance of
the centre of mass from the axis passing through O. Consequently the moment of inertia about
a given axis is equal to the moment of inertia about a parallel axis through the centre of mass
plus the moment of inertia of the body, as if concentrated at the centre of mass, with respect
to the original axis. This is known as the parallel axis theorem.
69
Modelling and Control of Multibody Mechanical Systems
8.5.2 Rotation of axes
The inertia matrix depends also upon the orientation of the body fixed axes with respect to
the body. The inertia matrix about some origin and with respect to an x′′ y ′′ z ′′ set of body axes
can be related to the inertia matrix when the origin is the same and the body fixed axis system
is x′ y ′ z ′ and has a different orientation. The position vector from the given origin O of the ith
particle can be expressed in the two coordinate systems as
However, the two representations of the position vector are connected between them by the
orthogonal transformation
ri′ = Ari , (8.48)
where A is the transformation matrix given by
α1 α2 α3
A = β1 β2 β3 . (8.49)
γ1 γ2 γ3
The elements of the A matrix are the direction cosines relating the orientations of the two sets
of axes, similarly to 7.9. It is clear from Equation 8.48 and the orthogonality of matrix A that
ri = AT ri′ . (8.50)
The angular momentum vector can be expressed in terms of ri′ by substituting ri from Equation
8.50 into Equation 8.4. The substitution gives
2
X T
H= mi (ri′ Ω − (ri′ A · Ω)AT ri′ ), (8.51)
i
where the property AT A = I was used. I is the identity matrix. The sum inside the brackets
will be recognised as the inertia tensor, I ′′ , in the new x′′ y ′′z ′′ system. The inertia tensor in
the original coordinate system, I ′ , is therefore given by
I ′ = AT I ′′ A, (8.53)
which means that the inertia tensor with respect to the rotated axes is given by
I ′′ = AI ′AT , (8.54)
or
Ix′′ x′′ Ix′′ y′′ Ix′′ z ′′ α1 α2 α3 Ix ′ x ′ Ix ′ y ′ Ix ′ z ′ α1 β1 γ1
Iy′′ x′′ Iy′′ y′′ Iy′′ z ′′ = β1 β2 β3 Iy′ x′ Iy′ y′ Iy′ z ′ α2 β2 γ2 . (8.55)
Iz ′′ x′′ Iz ′′ y′′ Iz ′′ z ′′ γ1 γ2 γ3 Iz ′ x ′ Iz ′ y ′ Iz ′ z ′ α3 β3 γ3
70
8.6 The principal axes
The inertia coefficients depend both upon the location of the origin of the body set of axes and
upon the orientation of these axes with respect to the body. It would be very convenient, of
course, if for a given origin one could find a particular orientation of the body axes for which
the inertia tensor is diagonal so that
I1 0 0
I = 0 I2 0 . (8.56)
0 0 I3
With respect to such a set of axes each of the components of H would involve only the corre-
sponding component of Ω, thus
H1 = I1 Ω1 , H2 = I2 Ω2 , H3 = I3 Ω3 . (8.57)
A similar simplification would also occur in the form of the kinetic energy:
ΩT IΩ 1 1 1
T = = I1 Ω21 + I2 Ω22 + I3 Ω23 . (8.58)
2 2 2 2
One can show that it is always possible to find such axes, and the proof is based essentially on
the hermitean (self-adjoint) nature of the inertia tensor.
The problem of finding a set of axes in which I is diagonal is equivalent to the eigenvalue
decomposition problem for the I matrix, I1 , I2 , and I3 being the three eigenvalues. It also
follows that for the coordinate system in which I is diagonal the direction of the axes coincides
with the direction of the eigenvectors. Thus, it can be proved that all the eigenvalues of I are
real, and that the three real directions of the eigenvectors are mutually orthogonal.
The methods of matrix algebra thus enable us to show that for any point in a rigid body
one can find a set of Cartesian axes for which the inertia tensor will be diagonal. The axes are
called the principal axes and the corresponding diagonal elements I1 , I2 , I3 are known as the
principal moments of inertia. Given some initial body set of coordinates one can transform to
the principal axes by a particular orthogonal transformation, which is therefore known as the
principal axis transformation.
In practice, of course, the principal moments of inertia, being the eigenvalues of I, are found
as the roots of the characteristic equation. Let wj be the jth eigenvector of I, with eigenvalue
Ij . The corresponding eigenvalue equation is then
Iwj = Ij wj . (8.59)
(I − II)w = 0, (8.60)
where the symmetry of I has been displayed explicitly. Equation 8.61 is the characteristic
equation, in the form of a cubic in I, whose roots are the desired principal moments. For
71
Modelling and Control of Multibody Mechanical Systems
each of these roots Equations 8.59 can be solved to obtain the direction of the corresponding
principal axis. In most of the easily solvable problems in rigid dynamics the principal axes can
be determined by inspection. For example, one usually has to deal with rigid bodies that are
solids of revolution about some axis, with the origin of the body system on the symmetry axis.
All directions perpendicular to the axis of symmetry are then alike, which is the mark of a
double root to the characteristic equation. The principal axes are the symmetry axis and any
two perpendicular axes in the plane normal to the symmetry axis.
In general the moment of inertia about a given axis has been defined as I = nT In (Equation
8.33). Let the direction cosines of the axis be α1 , α2 , and α3 so that
n = α1 i′ + α2 j ′ + α3 k′ ;
I = Ixx α12 + Iyy α22 + Izz α32 + 2Ixy α1 α2 + 2Iyz α2 α3 + 2Izx α3 α1 , (8.62)
using explicitly the symmetry of I. This expression can be verified by making use of Equation
8.55.
A quantity closely related to the moment of inertia is the radius of gyration, R0 , defined
by the equation
I = MR02 , (8.63)
or r
I
R0 = , (8.64)
M
and can be interpreted as the root-mean-square of the mass element distances from the axis of
rotation.
A body whose three principal moments of inertia are all different is called an asymmetrical
top. If two are equal (I1 = I2 6= I3 ) we have a symmetrical top. In this case the direction of one
of the principal axes in the x1 x2 -plane may be chosen arbitrarily. If all three principal moments
of inertia are equal, the body is called a spherical top, and the three axes of inertia may be
chosen arbitrarily as any three mutually perpendicular axes.
It is worth reemphasising that the inertia tensor I and all the quantities associated with
it–principal axes, principal moments, etc.–are only relative to some particular point fixed in
the body. If the point is shifted elsewhere in the body all the quantities will in general be
changed. Thus Equation 8.44 gives the effect of moving the reference point from the centre of
mass to some other point. The principal axis transformation which diagonalises I ′ at the centre
of mass will not necessarily diagonalise the difference term (the last matrix term in Equation
8.44), and hence is not in general the principal axis transformation for the shifted tensor IO .
Only if the shift vector R is along one of the principal axes relative to the centre of mass will
the difference tensor be diagonal in that system. The new inertia tensor IO will in that special
case have the same principal axes as at the centre of mass. However the principal moments of
inertia are changed, except for that corresponding to the shift, where the diagonal element of
the difference tensor is clearly zero.
72
be taken to include the effects of these constraints in the equations of motion. For example,
if there are “rolling constraints” these must be introduced into the equations of motion by
the method of Lagrange undetermined multipliers. As has been described in Section 8.1, one
usually seeks a particular reference point in the body such that the problem can be split into
two separate parts, one purely translational and the other purely rotational about the reference
point. Of course, if one point of the rigid body is fixed in an inertial system then that is the
obvious reference point. All that has to be considered then is the rotational problem about the
fixed point. For bodies without a fixed point the most useful reference point is almost always
the centre of mass. We have already seen that the total kinetic energy and angular momentum
then split neatly into one term relating to the translational motion of the centre of mass and
another involving rotation about the centre of mass. Thus, Equation 5.71 can now be written
1 1
T = Mv 2 + IΩ2 . (8.65)
2 2
For many problems (certainly all those that will be considered here) a similar sort of division can
be made for the potential energy. One can then solve individually for the translational motion
of the centre of mass and for the rotational motion about the centre of mass. For example, the
Newtonian equations of motion can be used directly; Equation 5.52 for the motion of the centre
of mass and Equation 5.56 for the motion about that point. It is convenient to work in terms
of the principal axes system of the point of reference, so that the kinetic energy of rotation
takes the simple form given in Equation 8.58. So far the only suitable generalised coordinates
we have for the rotational motion of the rigid body are the Euler angles. Of course, the motion
is often effectively confined to two dimensions, as in the motion of a rigid lamina in a plane.
The axis of rotation is then fixed in the direction perpendicular to the plane; only one angle
of rotation is necessary and one may dispense with the cumbersome machinery of the Euler
angles.
For the rotational motion about a fixed point or the centre of mass, the direct Newtonian
approach leads to a set of equations known as Euler’s equations of motion. We consider either
an inertial frame whose origin is at the fixed point of the rigid body, or a system of space axes
with origin at the centre of mass. In these two situations Equation 5.56 holds, which here
appears simply as
dH
= N, (8.66)
dt
where the time derivative is with respect to axes that do not share the rotation of the body.
However, Equation 4.42 can be used to obtain the derivatives with respect to axes fixed in the
body:
dH d′ H
= + Ω × H, (8.67)
dt dt
or
d′ H
+ Ω × H = N. (8.68)
dt
Equation 8.68 is thus the appropriate form of the Newtonian equation of motion relative to
body axes. If now the body axes are taken as the principal axes relative to the reference point
then the angular momentum is given by
H = I1 Ω1 i′ + I2 Ω2 j ′ + I3 Ω3 k′ , (8.69)
and therefore its time derivative relative to the body fixed axes is
d′ H
= I1 Ω̇1 i′ + I2 Ω̇2 j ′ + I3 Ω̇3 k′ , (8.70)
dt
73
Modelling and Control of Multibody Mechanical Systems
since i′ , j ′ , k′ are not changing with respect to the body and the principal moments of inertia
are of course time independent. The second term in Equation 8.68 is given by
In expanded form the three equations making up Equation 8.68 are therefore
I1 Ω̇1 − Ω2 Ω3 (I2 − I3 ) = N1 ,
I2 Ω̇2 − Ω3 Ω1 (I3 − I1 ) = N2 , (8.72)
I3 Ω̇3 − Ω1 Ω2 (I1 − I2 ) = N3 .
Equations 8.72 are the so-called Euler equations of motion for a rigid body with one point fixed.
I1 Ω̇1 = Ω2 Ω3 (I2 − I3 ),
I2 Ω̇2 = Ω3 Ω1 (I3 − I1 ), (8.73)
I3 Ω̇3 = Ω1 Ω2 (I1 − I2 ).
The same equations, of course, will also describe the motion of a rigid body when one point is
fixed and there are no net applied torques. We know two immediate integrals of the motion;
both the kinetic energy and the total angular momentum vector must be constant in time.
For a spherical top for which the three principal moments of inertia, I are equal and H =
IΩ, the condition H=constant gives Ω=constant; that is, the most general free rotation of a
spherical top is a uniform rotation about an axis fixed in space.
The law of conservation of angular momentum also suffices to determine the more complex
free rotation of a symmetrical top. Using the fact that the principal axes of inertia x1 , x2
(perpendicular to the axis of symmetry, x3 , of the top) may be chosen arbitrarily, we take the
x2 -axis perpendicular to the plane containing the constant vector H and the instantaneous
position of the x3 -axis (line of nodes). Then H2 = 0, and formulae 8.57 show that Ω2 = 0. This
means that the directions of H, Ω and the axis of the top are at every instant in one plane1
(Figure 8.3). Hence, in turn, it follows that the velocity v = Ω × r of every point on the axis
of the top is at every instant perpendicular to that plane. That is, the axis of the top rotates
uniformly about the direction of H, describing a circular cone. This is called regular precession
of the top. At the same time the top rotates uniformly about its own axis.
1
To prove the argument we make use of an x2 -axis that is changing at every instant. When normally x2 is
fixed to the body, and rotates with it, the two terms on the left-hand side of Equation 8.68 will both be non-zero
but their total sum will be zero, since the net torque applied is zero.
74
H
Ω
Ωpr x3
x1 θ
Figure 8.3: Precession of the angular velocity about the axis of symmetry in the force-free
motion of a symmetrical rigid body.
The angular velocities of these two rotations can easily be expressed in terms of the given
angular momentum H and the angle θ between the axis of the top and the direction of H.
The angular velocity of the top about its own axis is just the component Ω3 of the vector Ω
along the axis:
Ω3 = H3 /I3 = (H/I3 ) cos θ. (8.74)
To determine the rate of precession Ωpr , the vector Ω must be resolved into components along
x3 and along H. The first of these gives no displacement of the axis of the top, and the
second component is therefore the required angular velocity of precession. Figure 8.3 shows
that Ωpr sin θ = Ω1 , and, since Ω1 = H1 /I1 = (H/I1 ) sin θ, we have
Ωpr = H/I1 . (8.75)
As a simple example of the use of the Euler angles, we shall use them to determine again the
free motion of the symmetrical top. We take the z-axis of the space fixed system of coordinates
in the negative direction of the constant angular momentum H of the top. The x3 -axis of the
moving system is along the axis of the top; let the x2 -axis coincide with the line of nodes at
the instant considered. Then the components of the vector H are, by Equations 7.36, 7.37,
7.38, H1 = I1 Ω1 = I1 (φ̇ − ψ̇ sin θ), H2 = I1 Ω2 = I1 θ̇, H3 = I3 Ω3 = I3 (−ψ̇ cos θ). Note that,
Ω2 in Equation 7.37 and Ω3 in Equation 7.38 reduce to θ̇ and −ψ̇ cos θ respectively, since x2
has been chosen such that φ = 0. Also, since x2 -axis is perpendicular to the z-axis, we have
H1 = H sin θ, H2 = 0, H3 = H cos θ. Comparison gives
θ̇ = 0, (8.76)
I1 (φ̇ − ψ̇ sin θ) = H sin θ, (8.77)
−I3 ψ̇ cos θ = H cos θ. (8.78)
75
Modelling and Control of Multibody Mechanical Systems
The first of these equations gives θ =constant, i.e. the angle between the axis of the top and the
direction of H is constant. With some effort one can see that Ωpr = φ̇/ sin θ − ψ̇ and therefore
the second equation gives the angular velocity of precession φ̇/ sin θ − ψ̇ = H/I1 in agreement
with 8.75. Finally, the third equation gives the angular velocity with which the top rotates
about its own axis: −ψ̇ cos θ = Ω3 = (H/I3 ) cos θ.
As an example of the use of the Euler equations, let us apply these equations to the free
rotation of the symmetrical top yet again. Putting I1 = I2 , we find from the third equation in
8.73 Ω̇3 = 0, i.e. Ω3 =constant. We then write the first two equations as Ω̇1 = −ωΩ2 , Ω̇2 = ωΩ1 ,
where
ω = Ω3 (I3 − I1 )/I1 (8.79)
is a constant. These are two first order coupled linear differential equations which can easily
be solved for Ω1 and Ω2 . Thus
This result shows that the component of the angular velocity perpendicular to p the axis of the
top rotates with an angular velocity ω, remaining of constant magnitude A = Ω21 + Ω22 . Since
the component Ω3 along the axis of the top is also constant, we conclude that the vector Ω
rotates uniformly with angular velocity ω about the axis of the top, remaining unchanged in
magnitude. On account of the relations H1 = I1 Ω1 , H2 = I2 Ω2 , H3 = I3 Ω3 between the
components of Ω and H, the angular momentum vector H evidently executes a similar motion
with respect to the axis of the top.
This description is naturally only a different view of the motion already discussed earlier in
this Section, where it was referred to the space fixed system of coordinates. In particular, the
angular velocity of the vector H (the z-axis in Figure 8.3) about the x3 -axis is the negative
of Ω3 − Ωpr cos θ. The last expression is the magnitude of the component of Ω along the x3 -
axis, when Ω is decomposed along that axis and the vertical z-axis; of course, the remaining
component is Ωpr along the z-axis. Using Equations 8.77, 8.78 we have ψ̇ = −H/I3 , φ̇ =
H sin θ(1/I1 − 1/I3 ). Thus
φ̇ φ̇ cos θ 1 1
Ω3 − Ωpr cos θ = −ψ̇ cos θ − ( − ψ̇) cos θ = − = −H cos θ( − ), (8.82)
sin θ sin θ I1 I3
or
H cos θ I3 − I1 Ω3 (I3 − I1 )
−(Ω3 − Ωpr cos θ) = = , (8.83)
I3 I1 I1
which is in agreement with 8.79.
Ω = θ̇j
76
O
i
θ l
OCM
mg
H = IO θ̇j,
where IO is the moment of inertia of the rigid body about the axis of rotation. Use of Equation
8.68 provides the equation of motion:
IO θ̈ = −mgl sin θ,
or
g
θ̈ + sin θ = 0.
IO /(ml)
Comparing this equation with the equation for a simple pendulum, we see that the motion of
a compound pendulum is identical to the motion of a simple pendulum of equivalent length
IO
leq = .
ml
↑
77
Modelling and Control of Multibody Mechanical Systems
Analytical mechanics
F = F (u1 , u2 , . . . , un ). (9.1)
78
The corresponding change of the function F becomes by the rules of elementary calculus
∂F ∂F ∂F
δF = δu1 + δu2 + . . . + δun . (9.3)
∂u1 ∂u2 ∂un
This expression is called the “first variation” of the function F .
In order to operate with finite rather than infinitesimal quantities, we put:
But now the “virtual” nature of our displacement implies that we may proceed in any direction
we like, so that the αk are arbitrary and hence
∂F
= 0, (k = 1, 2, . . . , n). (9.7)
∂uk
Conversely, if the Equations 9.7 are satisfied, the quantity 9.5 vanishes and thus F has a
stationary value. Hence the Equations 9.7 are both necessary and sufficient. 2
F = F (u1, u2 , . . . , un ), (9.8)
79
Modelling and Control of Multibody Mechanical Systems
n − 1 unrestricted variables u1 , . . . , un−1 and could be handled as a free variation problem.
This method is entirely justified and sometimes advisable. But frequently the elimination is a
rather cumbersome procedure. Moreover, the condition 9.9 may be symmetric in the variables
u1 , . . . , un and there would be no reason why one of the variables should be artificially designated
as dependent, the others as independent variables.
Lagrange devised a method for handling auxiliary conditions, the “method of the undeter-
mined multiplier”, which preserves the symmetry of the variables without eliminations, and
still reduces the problem to one of free variation. The method works quite generally for any
number of auxiliary conditions and is applicable even to non-holonomic conditions which are
given as non-integrable relations between the differentials of the variables, and not as relations
between the variables themselves.
In order to understand the nature of the Lagrangian multiplier method, we start with a
single auxiliary condition, given in the form 9.9. Taking the variation of this equation we
obtain the following relation between the δuk :
∂f ∂f
δf = δu1 + . . . + δun = 0; (9.10)
∂u1 ∂un
while the fact that the variation of F has to vanish at a stationary value, gives
∂F ∂F
δF = δu1 + . . . + δun = 0. (9.11)
∂u1 ∂un
We know from the previous section that the condition 9.11 would lead to the vanishing of each
∂F
∂uk
if the δuk were all independent of each other. This, however, is not the case, because of the
∂f
conditions 9.10. We agree to eliminate δun in terms of the other variations–assuming that ∂u n
is not zero at the point P –and then consider the other δuk as free variations. But before we do
so, we shall modify the expression 9.11. It is obviously permissible to multiply the left-hand
side of 9.10 by some undetermined factor λ, which is a function of u1 , . . . , un , and add it to
δF . This does not change the value of δF at all since we have added zero. Hence it is still true
that:
∂F ∂F ∂f ∂f
δu1 + . . . + δun + λ δu1 + . . . + δun = 0 (9.12)
∂u1 ∂un ∂u1 ∂un
This move is not trivial because, although we have added zero, we have actually added a sum;
the individual terms of the sum are not zero, only the total sum is zero.
We write 9.12 in the form
n
X ∂F ∂f
+λ δuk = 0. (9.13)
k=1
∂u k ∂u k
We wish to eliminate δun . But now we can choose λ so that the factor multiplying δun shall
vanish:
∂F ∂f
+λ = 0. (9.14)
∂un ∂un
This dispenses with the task of eliminating δun . After that our sum is reduced to only n − 1
terms:
n−1
X ∂F ∂f
+λ δuk = 0, (9.15)
k=1
∂uk ∂uk
80
and since only those δuk remain which can be chosen arbitrarily, the conditions of a free variation
problem are applicable. These require that the coefficient of each δuk shall vanish:
∂F ∂f
+λ = 0, (k = 1, 2, . . . , n − 1). (9.16)
∂uk ∂uk
The conditions 9.16, combined with the condition 9.14 on λ, lead to the conclusion that each
coefficient of the sum 9.13 vanishes, just as if all the variations δuk were free variations. The
result of Lagrange’s “method of undetermined multiplier” can be formulated thus: instead of
considering the vanishing of δF , consider the vanishing of
δF + λδf, (9.17)
and drop the auxiliary condition, handling the uk as free, independent variables.
We have
δF + λδf = δ(F + λf ), (9.18)
since the term f δλ vanishes on account of the auxiliary condition f = 0. Hence we can express
the result of our deductions in an even more compact form. Instead of putting the first variation
of F equal to zero, modify the function F to
F = F + λf, (9.19)
and put its first variation equal to zero, for arbitrary variations of the uk .
We generalise this λ-method for the case of an arbitrary number of auxiliary conditions.
Let us assume once more that the stationary value of F is sought, but under m independent
restricting conditions:
f1 (u1 , u2 , . . . , un ) = 0,
..
. (m < n) (9.20)
fm (u1 , u2 , . . . , un ) = 0.
These auxiliary conditions establish the following relations between the variations δuk :
∂f1 ∂f1
δf1 = δu1 + . . . + δun = 0,
∂u1 ∂un
..
. (9.21)
∂fm ∂fm
δfm = δu1 + . . . + δun = 0.
∂u1 ∂un
Because of these conditions m of the δuk can be designated as dependent variables and expressed
in terms of the others. We shall consider the last m of the uk as dependent, the first n − m as
independent, variables.
Now the given variational problem requires the vanishing of
n
X ∂F
δF = δuk (9.22)
k=1
∂u k
for all possible variations δuk which satisfy the given auxiliary conditions. We should express
the last m δuk in terms of the independent δuk . However, before doing so let us modify the
81
Modelling and Control of Multibody Mechanical Systems
expression 9.22 by adding the left-hand sides of the Equations 9.21 after multiplying each one
by some undetermined λ-factor. We thus get
n
X ∂F ∂f1 ∂fm
+ λ1 + . . . + λm δuk = 0. (9.23)
k=1
∂uk ∂uk ∂uk
Now the elimination of the last m δuk can be accomplished by the proper choice of the λ-factors,
so that
∂F ∂f1 ∂fm
+ λ1 + . . . + λm = 0, (k = n − m + 1, . . . , n). (9.24)
∂uk ∂uk ∂uk
This leaves
n−m
X ∂F
∂f1 ∂fm
+ λ1 + . . . + λm δuk = 0. (9.25)
∂uk ∂uk ∂uk
k=1
But all the δuk which remain in 9.25 are free variations. Hence the coefficient of each δuk must
vanish separately. In the final analysis we have the equations
∂F ∂f1 ∂fm
+ λ1 + . . . + λm = 0, (k = 1, . . . , n), (9.26)
∂uk ∂uk ∂uk
which can be considered as obtained from the variational principle
δF + λ1 δf1 + . . . + λm δfm = 0, (9.27)
considering all the uk as independent variables. Thus in the final result the distinction between
dependent and independent variables disappears.
Equation 9.27 can be stated even more compactly by writing it in the form
δ(F + λ1 f1 + . . . + λm fm ) = 0, (9.28)
and interpreting this equation as follows: instead of asking for the stationary value of F , we
ask for the stationary value of the modified function
F = F + λ1 f1 + . . . + λm fm , (9.29)
dropping the auxiliary conditions and handling this as a free variation problem. This yields n
equations. In addition to these equations we have to satisfy the m auxiliary conditions 9.20.
This gives n + m equations for the n + m unknowns
u 1 , u 2 , . . . , u n ; λ1 , λ2 , . . . , λm . (9.30)
The multiplier method of Lagrange changes a problem of n − m degrees of freedom to a
problem of n + m degrees of freedom. If we add to the n variables uk the m quantities λi as
additional variables and ask for the stationary value of the function F , this variation problem
gives the same n equations as we had before if we vary with respect to the uk , while the
variations of the λi give the m additional conditions
f1 = 0, . . . , fm = 0. (9.31)
These are exactly the given auxiliary conditions, but now obtained a posteriori, on account of
the variation problem.
This method of Lagrange permits the use of surplus coordinates–a great convenience in
many considerations of mechanics. It preserves the full symmetry of all coordinates by making
it unnecessary to distinguish between dependent and independent variables.
82
9.1.3 Non-holonomic auxiliary conditions
As was pointed out in Section 6.3, the restrictions on the mechanical variables of a problem
may be given in a differential (infinitesimal) instead of a finite form. We then have a variation
problem with non-holonomic auxiliary conditions. The Equations 9.20 do not exist in this case,
but we have relations analogous to the differentiated forms 9.21 of the auxiliary conditions. The
only difference is that the left-hand sides of these equations are no longer exact differentials but
merely infinitesimal quantities. We can write the non-holonomic conditions in the following
form:
Here the Aik are given functions of the ui which cannot be considered as the partial derivatives
of a function fi .
Non-holonomic conditions cannot be handled by the elimination method, because the equa-
tions for eliminating some variables as dependent variables do not exist. The Lagrangian
λ-method, however, is again available. By exactly the same procedure as before, we can obtain
an equation analogous to 9.27, namely:
δF + λ1 δf 1 + . . . + λm δf m = 0; (9.33)
and again all the δuk are handled as free variations. The only difference lies in the fact that we
cannot proceed to the equation 9.28 and have to be content with the differential formulation
of the procedure. The reduction of a conditioned variation problem to a free variation problem
is once more accomplished.
83
Modelling and Control of Multibody Mechanical Systems
y (b, β)
(a, α)
x
with the boundary conditions that the qk are given (and thus their variation is zero) at the two
end-points t1 and t2 :
[δqk (t)]t=t1 = 0, [δqk (t)]t=t2 = 0. (9.40)
The q1 , . . . , qn are unknown functions of t, to be determined by the condition that the actual
motion shall make the integral J stationary:
δJ = 0, (9.41)
4
See for example Lanczos, page 54, Goldstein, page 37.
84
for arbitrary independent variations of the qk , subject only to boundary conditions 9.40.
Now we can obviously select one definite qk and vary it all by itself, leaving the other qi
unchanged. Hence we can apply the differential equation 9.38 to our present problem, after
adapting it to the present notation. The y corresponds to qk , the ẏ to q̇k , the independent
variable x is now the time t. The function F is denoted by L and the limits of integration are
t1 and t2 . Hence we get
∂L d ∂L
− = 0, (t1 ≤ t ≤ t2 ). (9.42)
∂qk dt ∂ q̇k
These equations have to hold for each separate qk , k taking values from 1 to n.
The variations we have employed so far are but special variations, and the question arises
whether a simultaneous variation of all the qk would not bring in additional conditions. This
is actually not the case, on account of the superposition principle of infinitesimal processes.
Let us denote by δk J the variation of J produced by varying qk alone. Then the simultaneous
variation of all the qk produces the following resultant variation:
δJ = δ1 J + δ2 J + . . . + δn J. (9.43)
Now the differential equation 9.42 guarantees the vanishing of δk J. If that differential equation
holds for all indices k the sum 9.43 vanishes, and thus δJ is zero for arbitrary variations of the
qk .
The problem of finding the stationary value of J for arbitrary variations of the qk between
definite limits is thus solved. The conditions for the stationary value of J come out in the form
of the following system of simultaneous differential equations:
d ∂L ∂L
− = 0, (k = 1, 2, . . . , n). (9.44)
dt ∂ q̇k ∂qk
They are called “the differential equations of Euler and Lagrange”, or also, if applied to problems
of mechanics, “the Lagrangian equations of motion.”
With the exception of the singular case in which the function L depends on some or all
the q̇k in a linear way, the partial derivatives ∂∂L
q̇k
will contain all the q̇k , so that differentiation
with respect to t brings in all the second derivatives q̈k . We can solve the equations for the q̈k
algebraically and thus rewrite the differential equations 9.44 in the following explicit form:
The integration of such a system of differential equations of the second order involves 2n
constants of integration, so that the complete solution of the equations 9.45 may be written as
follows:
qk = qk (A1 , . . . , An ; B1 , . . . , Bn ; t). (9.46)
The constants of integration Ak and Bk can be adjusted to the given boundary conditions. Our
variational problem requires variation between definite limits, which means that the coordinates
qk are given at t = t1 and t = t2 . These are 2n boundary conditions which can be satisfied by
the proper choice of the constants Ak and Bk . The nature of mechanical problems is such that
more frequently initial conditions take the place of boundary conditions. The freedom of 2n
constants of integration allows all the initial position coordinates and velocities to be prescribed
arbitrarily.
85
Modelling and Control of Multibody Mechanical Systems
9.1.6 Variation with auxiliary conditions
We consider once more the problem of the previous Section, with the modification that the
variables q1 , . . . , qn shall not be independent, but restricted by given auxiliary conditions. These
conditions take the form of certain functional relations between the qk :
f1 (q1 , . . . , qn , t) = 0,
..
. (9.47)
fm (q1 , . . . , qn , t) = 0.
It is possible to eliminate m of the qk in terms of the other variables and thus reduce the problem
to n − m degrees of freedom. After the reduction the differential equations of Euler-Lagrange
come into play. However, this elimination may be rather cumbersome; moreover, the conditions
between the variables may be of a form which makes the distinction between dependent and
independent variables artificial. Here again the method of the Lagrangian multiplier, studied
before in Section 9.1.2, gives an adequate solution of the problem.
Variation of the equations 9.47 gives:
∂f1 ∂f1
δf1 = δq1 + . . . + δqn = 0,
∂q1 ∂qn
..
. (9.48)
∂fm ∂fm
δfm = δq1 + . . . + δqn = 0.
∂q1 ∂qn
These equations hold at any time t. According to the principal of the Lagrangian multiplier
we multiply each one of these equations by an undetermined factor λi . Since the auxiliary
conditions are prescribed for every value of the independent variable t, the λ-factors have also
to be applied for every value of t, which makes them functions of t. Moreover, the summation
over all the auxiliary conditions–after multiplication by λi –amounts to an integration with
respect to the time t. Thus the method of the Lagrangian multiplier appears here in the
following form: instead of putting the variation of the given integral J equal to zero, modify it
in the following fashion:
Z t2 Z t2
′
δJ = δ Ldt + (λ1 δf1 + . . . + λm δfm )dt = 0. (9.49)
t1 t1
This is obviously permissible since what we have actually added to the variation of J is zero.
Now in the first term we perform the standard integration by parts and reduce it to the form
Z t2 Z t2
δ Ldt = [E1 (t)δq1 + . . . + En (t)δqn ]dt, (9.50)
t1 t1
where
∂L d ∂L
Ek (t) = − . (9.51)
∂qk dt ∂ q̇k
We now unite the two integrands of 9.49 and collect the terms which belong to a certain δqk .
We should eliminate the last (n − m) δqk with the help of the equations 9.48, but we can avoid
this step by choosing the λi in such a manner that the coefficients of these δqk shall vanish. The
remaining δqi , on the other hand, can be chosen arbitrarily and hence our previous conclusion
86
that the coefficients of these δqi must vanish individually throughout the range becomes valid
again. But then we see that in the final analysis the coefficient of each individual δqk vanishes,
irrespective of whether a certain qk was designated as a dependent or an independent variable.
The resulting equations are:
∂L d ∂L ∂f1 ∂fm
− + λ1 + . . . + λm = 0. (9.52)
∂qk dt ∂ q̇k ∂qk ∂qk
This is equivalent to the following variational problem. Instead of considering the variation of
the definite integral Z t2
J= Ldt, (9.53)
t1
with the auxiliary conditions 9.47, consider the variation of the modified integral
Z t2
′
J = L′ dt, (9.54)
t1
where
L′ = L + λf1 + . . . + λm fm , (9.55)
and drop the auxiliary conditions.
Here again, as in the general procedure, the λi have to be considered as constants relative to
the process of variation. But we can also include them in the variational problem as additional
unknown functions of t. The variation of the λi gives then the auxiliary conditions 9.47 a
posteriori.
The solution of the differential equations 9.52 has to be completed by satisfying the m
auxiliary conditions 9.47. This determines the λi as functions of t. In the matter of initial
conditions we can choose arbitrarily only n − m position coordinates and n − m velocities,
because the remaining qi and q̇i are determined by the auxiliary conditions. This is in accord
with the fact that the given system represents a mechanical system of n − m degrees of freedom
and the use of surplus coordinates is merely a matter of mathematical convenience.
87
Modelling and Control of Multibody Mechanical Systems
is not performed instantaneously but during the infinitesimal time δt. The auxiliary conditions
now take the form
Ai1 δq1 + . . . + Ain δqn + Bi δt = 0, (9.57)
with coefficients Aik and Bi which are in general functions of the qi and the time t. The
quantities Bi do not enter into the equations of motion since the virtual displacements δqk are
performed without varying the time; but they do enter into the relations which exist between
the velocities:
Ai1 q̇1 + . . . + Ain q̇n + Bi = 0. (9.58)
These virtual displacements must be in harmony with the given kinematical constraints, and
we shall assume that they are reversible, i.e. the given constraints do not prevent us from
changing an arbitrary δRi into −δRi .
Now the principle of virtual work asserts that the given mechanical system will be in equi-
librium if, and only if, the total virtual work of all the impressed forces vanishes:
Let us translate this equation into analytical language. For this purpose we express the rect-
angular coordinates xi , yi , zi as functions of the generalised coordinates q1 , q2 , . . . , qn exactly
as we have done in Section 6.4. The differential form 9.60 is then transformed into the new
differential form
δw = F1 δq1 + F2 δq2 + . . . + Fn δqn , (9.61)
where F1 , F2 , . . . , Fn are called the components of the generalised force. They form a vector
of the n-dimensional configuration space.
The principle of virtual work requires that
88
We can give a geometrical interpretation of this equation. The left-hand side of the equation
is nothing but “scalar product” of force and virtual displacement. The vanishing of this scalar
product means that the force Fi is perpendicular to any possible virtual displacement.
Let us assume at first that the given mechanical system is free of any constraints. In that
case the C-point5 of the configuration space can be displaced in an arbitrary direction. Then
the principle 9.62 requires that the force Fi shall vanish, because there is no vector which can
be perpendicular to all directions in space.
Let us assume that the C-point has to stay within a certain (n−m)-dimensional subspace of
the configuration space, on account of m given kinematical constraints. Then the condition 9.62
no longer requires the vanishing of the force Fi , but only its perpendicularity to that subspace.
This amounts to n − m equations, in conformity with the n − m degrees of freedom of the
mechanical system.
We now come to the physical interpretation of the principle of virtual work. According to
Newtonian mechanics, the state of equilibrium requires that the resultant force acting on any
particle of the system shall vanish. This resultant force is the sum of the impressed force and
the forces which maintain the given constraints. These latter forces are usually called “forces of
reaction”. Since the principle of equilibrium requires that “impressed force plus resultant force
of reaction equals zero,” we see that the virtual work of the impressed forces can be replaced
by the negative virtual work of the forces of reaction. Hence the principle of virtual work can
be formulated in the following form, which we shall call Postulate A:
“The virtual work of the forces of reaction is always zero for any virtual displacement
which is in harmony with the given kinematic constraints.”
This postulate is not restricted to the realm of statics. It applies equally to dynamics, when
the principle of virtual work is suitably generalised by means of d’Alembert’s principle.
The principle of virtual work assumes a special significance in the particularly important
case where the impressed force Fi is monogenic, i.e. derivable from a single scalar function,
the work function. In this case the virtual work is equal to the variation of the work function
U(q1 , . . . , qn ). Since the work function can be replaced by the negative of the potential energy,
we can say that the state of equilibrium of a mechanical system is distinguished by the stationary
value of the potential energy, i.e. by the condition
δV = 0. (9.63)
If the equilibrium is stable, the potential energy must assume its minimum value–the minimum
understood in the local sense–while in general, equilibrium does not require the minimum, but
only the stationary value, of V .
89
Modelling and Control of Multibody Mechanical Systems
D’Alembert’s principle, which we shall discuss in the present Section, did not contribute directly
to the problem of integration. Yet it is an important landmark in the history of theoretical
mechanics, since it gives an interpretation of the forces of inertia which is fundamental for the
later development of variational methods.
F − ma = 0. (9.65)
If = −ma. (9.66)
This vector If can be considered as a force, created by the motion. We call it the “force of
inertia”. With this concept the equation of Newton can be formulated as follows:
F + If = 0. (9.67)
Apparently nothing is gained, since the intermediate step 9.66 gives merely a new name to the
negative product of mass times acceleration. It is exactly this apparent triviality which makes
d’Alembert’s principle such an ingenious invention and at the same time so open to distortion
and misunderstanding.
The importance of the equation 9.67 lies in the fact that it is more than a reformulation of
Newton’s equation. It is the expression of a principle. We know that the vanishing of a force
in Newtonian mechanics means equilibrium. Hence the equation 9.67 says that the addition of
the force of inertia to the other acting forces produces equilibrium. But this means that if we
have any criterion for the equilibrium of a mechanical system, we can immediately extend that
criterion to a system which is in motion. All we have to do is to add the new “force of inertia”
to the previous forces. By this device dynamics is reduced to statics.
This does not mean that we can actually solve a dynamical problem by statical methods.
The resulting equations are differential equations which have to be solved. We have merely
deduced these differential equations by statical considerations. The addition of the force of
inertia If to the acting force F changes the problem of motion to a problem of equilibrium.
D’Alembert generalised his equilibrium consideration from a single particle to any arbitrary
mechanical system. His principle states that any system of forces is in equilibrium if we add
to the impressed forces the forces of inertia. This means that the total virtual work of the
impressed forces, augmented by the inertial forces, vanishes for reversible displacements. It
seems appropriate to have a special name for the force which results if we add the force of
inertia If k to the given impressed force Fk which acts on a particle. We shall call this force
the “effective force” and denote it by Fke :
90
D’Alembert’s principle can now be formulated as follows: The total virtual work of the effective
forces is zero for all reversible variations which satisfy the given kinematical conditions:
N
X N
X
Fke · δRk ≡ (Fk − mk ak ) · δRk = 0. (9.69)
k=1 k=1
Notice that the impressed forces Fk may act at just a few points, while the effective forces Fke
are present whenever a mass is in accelerated motion.
A given system of impressed forces will generally not be in equilibrium. This requires the
fulfilling of special conditions. The total virtual work of the impressed forces will usually be
different from zero. In that case the motion of the system makes up for the deficiency. The
body moves in such a way that the additional inertial forces, produced by the motion, bring
the total up to zero. In this way d’Alembert’s principle gives the equations of motion of an
arbitrary mechanical system.
In contrast to the impressed forces, which are usually derivable from a single work function
by differentiation (monogenic forces), the forces of inertia are of a polygenic character. While
the virtual work of the impressed forces can be written as the complete variation of the work
function:
δw = δU, (9.70)
is merely a differential form, not reducible to the variation of a scalar function. (We shall see
later how this shortcoming can be remedied by an integration with respect to the time.)
The Newtonian equation 9.64 holds only if the mass is constant. If m changes during the
motion, the fundamental equation of motion has to be written in the form:
d
(mv) = F , (9.72)
dt
i.e., “rate of change of momentum equals moving force.” Accordingly, the force of inertia If
has to be defined as the negative rate of change of momentum:
d
If = − (mv). (9.73)
dt
For the customary case of constant m the general definition can be replaced by 9.66.
We can ask the question, what is the physical significance of d’Alembert’s principle? From
the definition of the “effective force” by 9.68 it follows that this force is zero in the case of
a free particle, while it is equal to the negative force of reaction if the particle is subject to
constraints. Hence the application of the principle of virtual work to the effective force F e is
equivalent to the assumption that “the virtual work of the forces of reaction is zero for any
virtual displacement which is in harmony with the given constraints.” We thus come back
to the same “Postulate A” that we have encountered before in Section 9.2. D’Alembert’s
principle generalises this postulate from the field of statics to the field of dynamics, without
any alteration.
91
Modelling and Control of Multibody Mechanical Systems
9.3.2 The place of d’Alembert’s principle in mechanics
D’Alembert’s principle gives a complete solution of problems of mechanics. All the differ-
ent principles of mechanics are merely mathematically different formulations of d’Alembert’s
principle. The most advanced variational principle of mechanics, Hamilton’s principle, can be
obtained from d’Alembert’s principle by a mathematical transformation. It is equivalent to
d’Alembert’s principle where both are applicable. In fact, Hamilton’s principle is restricted to
holonomic systems, while d’Alembert’s principle can equally well be applied to holonomic and
nonholonomic systems.
This principle is more elementary than the later variational principles because it requires no
integration with respect to the time. But the disadvantage of the principle is that the virtual
work of the inertial forces is a polygenic quantity and thus not reducible to a single scalar
function. This makes the principle unsuited to the use of curvilinear coordinates. However, in
many elementary problems of dynamics which can be adequately treated with the help of rect-
angular coordinates, or by the vector methods without the use of any coordinates, d’Alembert’s
principle is of great use.
For certain problems d’Alembert’s principle is even more flexible than the more advanced
principle of least action. The differential equations of motion are of second order, determining
the accelerations of the moving system. These accelerations are the second derivatives q̈i of
the position coordinates qi or the first derivatives of the velocities q̇i . Now it may happen–and
such a situation arises particularly in the dynamics of rigid bodies–that it is more convenient to
characterise the motion with the help of certain velocities which are not the derivatives of actual
position coordinates. Such quantities are called “kinematical variables”. A good example is
the spin of a top about its axis of symmetry. This spin is the angular velocity of rotation,
ω = dφ/dt, but the dφ is here merely an infinitesimal angle of rotation and not the differential
of an angle φ. This angle exists only if the axis is fixed but not if the axis is changing. And yet
it is convenient to use the spin as one of the quantities which characterise the motion of the top.
In the principle of least action we cannot use kinematical variables, whereas in d’Alembert’s
principle we can.
We now arrive at the typical “variational principles,” namely, those principles which operate
with the minimum–or more generally the stationary value–of a definite integral. The polygenic
character of the force of inertia can be overcome if we integrate with respect to the time. By
this procedure the problem of dynamics is reduced to the investigation of a scalar integral. The
condition for the stationary value of this integral gives all the equations of motion.
In spite of the various names connected with these principles–Euler, Lagrange, Jacobi,
Hamilton– they are all closely related to each other, and the name “principle of least action,”
if taken in the broader sense, applies to all of them.
In our exposition of these principles we shall not follow the historical development, but we
will concentrate on “Hamilton’s principle,” which is the most direct and most natural transfor-
mation of d’Alembert’s into a minimum principle. From this we can obtain by specialisation
the older form of the principle used by Euler and Lagrange, and likewise Jacobi’s principle.
92
9.4.1 Hamilton’s principle
D’Alembert’s principle operates with a non-integrable differential. A certain infinitesimal quan-
tity δw e –the total virtual work done by impressed and inertial forces–is equated to zero. The
two parts of the work done are very different in their nature. While the virtual work of the
impressed forces is a monogenic differential, deducible from a single function, the work func-
tion, the virtual work of the inertial forces cannot be deduced from a single function but has
to be formed for each particle separately. This puts the inertial forces at a great disadvantage
compared with the impressed forces. It is of the greatest theoretical and practical importance
that this situation can be remedied by a transformation which brings d’Alembert’s principle
into a monogenic form. Although implicitly used by Euler and Lagrange, it was Hamilton who
first transformed d’Alembert’s principle, showing that an integration with respect to the time
brings the work done by the inertial force into a monogenic form.
Let us multiply δw e by dt and integrate between the limits t = t1 and t = t2 :
Z t2 Z t2 X
e
d
δw dt ≡ Fi − (mi vi ) · δRi dt. (9.74)
t1 t1 dt
We separate the right-hand side into two parts. The first part can be written
Z t2 X Z t2 Z t2
Fi · δRi dt = − δV dt = −δ V dt. (9.75)
t1 t1 t1
(We assume that the work function is independent of the velocities and we put V = −U). In
the second part an integration by parts can be performed:
Z t2 Z t2 Z t2
d d d
− (mi vi ) · δRi dt = − (mi vi · δRi )dt + mi vi · (δRi )dt. (9.76)
t1 dt t1 dt t1 dt
The first part of this sum is integrable and we get a boundary term:
while the second part, making use of the interchangeable nature of variation and differentiation,
may be written
Z t2 Z t2
1 t2
Z t2
d 1
Z
mi vi · (δRi )dt = mi vi · δvi dt = mi δ(vi · vi )dt = δ mi vi 2 dt. (9.78)
t1 dt t1 2 t1 2 t1
We introduce the kinetic energy T of the mechanical system according to the definition 5.70
and we furthermore set
L = T − V. (9.80)
This function L, defined as the excess of kinetic energy over potential energy, is the most
fundamental quantity in the mathematical analysis of mechanical problems. It is frequently
referred to as the “Lagrangian function.”
93
Modelling and Control of Multibody Mechanical Systems
With these definitions we can write 9.79 in the form:
Z t2 Z t2 hX it2
e
δw dt = δ Ldt − mi vi · δRi . (9.81)
t1 t1 t1
So far the variations δRi are arbitrary virtual changes of the radius vectors Ri . We now require
that the δRi shall vanish at the two limits t1 and t2 :
This means that the position of the mechanical system is supposed to be given for t = t1 and
t = t2 , and no variations are allowed at these limits. We say that we “vary between definite
limits”, because the limiting positions of the system are prescribed. In that case the boundary
term on the right-hand side of 9.81 vanishes and the integral with respect to the time of the
virtual work done by the effective forces becomes the variation of a definite integral:
Z t2 Z t2
e
δw dt = δ Ldt = δA, (9.84)
t1 t1
with Z t2
A= Ldt. (9.85)
t1
Since d’Alembert’s principle requires the vanishing of δw e at any time, the left-hand side of
9.84 vanishes also. Hence d’Alembert’s principle can be reformulated as follows:
δA = 0. (9.86)
This is “Hamilton’s principle” which states that the motion of an arbitrary mechanical system
occurs in such a way that the definite integral A becomes stationary for arbitrary possible
variations of the configuration of the system, provided the initial and final configurations of the
system are prescribed.6
Our reasoning which led to Hamilton’s principle can be pursued in the opposite order. We
can start out with the postulate that δA vanishes for arbitrary variations of position, transform
δA to the left-hand side of 9.84 and deduce the vanishing of δw e , which is d’Alembert’s principle.
This shows that Hamilton’s principle and d’Alembert’s principle are mathematically equivalent
and their scopes are the same as long as the impressed forces which act on the mechanical
system are monogenic. For polygenic forces the transformation of d’Alembert’s principle into a
minimum–or strictly speaking stationary value–principle is not possible. Since holonomic kine-
matical conditions are mechanically equivalent to monogenic forces, nonholonomic conditions
to polygenic forces, we can say that Hamilton’s principle holds for arbitrary mechanical systems
which are characterised by monogenic forces and holonomic auxiliary conditions. The conser-
vative nature of these forces and auxiliary conditions–i.e. their independence of the time–is not
required.
While d’Alembert’s principle makes an independent statement at each instant of time during
the motion, Hamilton’s principle includes all these in one single statement. The motion is taken
as a whole.
6
Hamilton gave an improved mathematical formulation of a principle which was well established by the
fundamental investigations of Euler and Lagrange; the integration process employed by him was likewise known
to Lagrange. The name “Hamilton’s principle,” coined by Jacobi, was not adopted by the scientists of the 19th
century. It came into use, however, through the textbooks of more recent date.
94
9.4.2 The Lagrangian equations of motion and their invariance rel-
ative to point transformations
In the derivation of the principle of least action rectangular coordinates are used. However,
the mechanical system may be subject to kinematical conditions; if these conditions are holo-
nomic, the 3N rectangular position coordinates of the system will be expressible in terms of n
generalised coordinates q1 , . . . , qn , as in 6.8 and 6.41. Then both the potential and kinetic en-
ergies are functions of the generalised coordinates and the generalised velocities q̇1 , . . . , q̇n . The
motion of the system can now be pictured as that of the single C-point in the n-dimensional
configuration space of the qi .
The necessary and sufficient conditions for the action integral 9.85 to be stationary are (see
Section 9.1):
d ∂L ∂L
− = 0, (i = 1, . . . , n). (9.87)
dt ∂ q̇i ∂qi
Since the significance of the n generalised coordinates q1 , q2 , . . . , qn is not specified beyond
the requirement that it shall allow a complete characterisation of the system, we may choose
another set of quantities
q1 , q 2 , . . . , q n (9.88)
as generalised coordinates. There must, then, exist a functional relationship between the two
sets of coordinates expressible in the form:
q 1 = f1 (q1 , . . . , qn ),
..
. (9.89)
q n = fn (q1 , . . . , qn ).
The functions f1 , f2 , . . . , fn must satisfy the ordinary regularity conditions. They must be
finite, single valued, continuous and differentiable functions of the qk , etc.
It is a remarkable fact that the problem of minimising a definite integral is quite indepen-
dent of any special reference system. Let us assume that the original set of coordinates qi is
changed to a new set of coordinates by a point-transformation 9.89. This point-transformation
can be pictured as a mapping of the n-dimensional q-space on itself. The vanishing of the vari-
ation of the integral A requires the vanishing of this variation when expressed in terms of the
new coordinates. Hence the differential equations of Euler-Lagrange remain valid in the new
reference system. The Lagrangian function L and the action integral A are invariants of the
transformation. We merely substitute for qi the functions which express them in terms of the
new variables q i . The form of L in the new variables is quite different from what it was before,
but its value remains the same. The Lagrangian equations of motion, if taken separately for
a certain index i, are not transformed into the corresponding equations in the new reference
system. But the complete set of equations is transformed into the corresponding set in the new
coordinate system, because the complete set of Lagrangian equations expresses the vanishing of
the variation of the action integral A, and this statement is independent of any special system
of coordinates.
Since t is merely one more additional variable, these considerations remain valid even if
the relations between the old and the new qi depend on the time t. This is the case if the
mechanical phenomena are described in a reference system which is in motion. The Lagrangian
equations of motion remain valid even in arbitrarily moving reference systems.
The invariance of the Lagrangian equations of motion is one of their most important features.
It makes it possible to adjust the type of coordinates employed to the nature of the problem.
95
Modelling and Control of Multibody Mechanical Systems
There is no general method known by which the Lagrangian equations can be solved. The best
that we can do is to try to find a system of coordinates in which the equations are at least
partially integrable.
Another great advantage of the principle of least action, compared with d’Alembert’s princi-
ple, is the use of the single scalar function L. It is no longer necessary to find the acceleration of
each particle and the virtual work done by all the inertial forces. The scalar function L = T −V
determines the entire dynamics of the given system.
The two basic functions T and V correspond to the two sides of the Newtonian equation:
“Mass times acceleration equals force”. This equation can be interpreted as the balance between
the force of inertia and the moving force. A similar balance can be established in the analytical
treatment by separating the two basic scalars of analytical mechanics: the kinetic energy T and
the work function U. The Lagrangian equations can be written
d ∂T ∂T d ∂U ∂U
− =− + . (9.90)
dt ∂ q̇i ∂qi dt ∂ q̇i ∂qi
The n quantities
d ∂T ∂T
− (9.91)
dt ∂ q̇i ∂qi
represent the n components of the force of inertia in the configuration space. The n quantities
d ∂U ∂U
− + (9.92)
dt ∂ q̇i ∂qi
similarly represent the n components of the moving force.
Ordinarily, the first term of 9.92 is omitted on the ground that U is a function of the qi
alone. Moreover, U is replaced by −V . However, there is no inherent reason for excluding the
possibility that the work function may depend on the velocities q̇i .
9.4.3 Examples
Compound pendulum
Consider the compound pendulum as given in Example 8.9. Derive its equation of motion using
the Lagrangian approach.
↓
The kinetic energy of the pendulum, by making use of equation 8.31, is:
1
T = IO θ̇2 .
2
The potential energy, via equation 8.35, is:
96
or
d
IO θ̇ + mgl sin θ = 0,
dt
or
IO θ̈ + mgl sin θ = 0,
which is the same equation found in Example 8.9.
↑
Double pendulum
Consider the double pendulum in Example 5.2.4. Derive the equations of motion using the
Lagrangian approach.
↓
First, we calculate the kinetic energy. By equation 5.70 the kinetic energy is
1X 1
T = mi vi2 = (m1 v12 + m2 v22 ),
2 i 2
or
1
T = (m1 v1 · v1 + m2 v2 · v2 ).
2
But from equation 5.80
v1 = ṙ1 = r1 θ̇1 eθ1 ,
and from equation 5.81
v2 = ṙ2 = r1 θ̇1 eθ1 + l2 φ̇eφ = r1 θ̇1 sin(φ − θ1 )el2 + (r1 θ̇1 cos(φ − θ1 ) + l2 φ̇)eφ .
Therefore
1 2 2 1
2 2 2 2
T = m1 r1 θ̇1 + m2 r1 θ̇1 sin (φ − θ1 ) + (r1 θ̇1 cos(φ − θ1 ) + l2 φ̇) ,
2 2
or
1 1
T = m1 r12 θ̇12 + m2 r12 θ̇12 + 2r1 l2 θ̇1 φ̇ cos(φ − θ1 ) + l22 φ̇2 ,
2 2
or
1 1
T = (m1 + m2 )r12 θ̇12 + m2 l22 φ̇2 + 2r1 l2 θ̇1 φ̇ cos(φ − θ1 ) .
2 2
The potential energy of the system is
or
V = −(m1 + m2 )gr1 cos θ1 − m2 gl2 cos φ.
The Lagrangian function is given by
1 1
L = T −V = (m1 +m2 )r12 θ̇12 + m2 l22 φ̇2 + 2r1 l2 θ̇1 φ̇ cos(φ − θ1 ) +(m1 +m2 )gr1 cos θ1 +m2 gl2 cos φ.
2 2
The first Lagrangian equation is
d ∂L ∂L
− = 0.
dt ∂ θ˙1 ∂θ1
97
Modelling and Control of Multibody Mechanical Systems
We start by evaluating the following
∂L
= (m1 + m2 )r12 θ̇1 + m2 r1 l2 φ̇ cos(φ − θ1 ).
˙
∂ θ1
From this we get
d ∂L
= (m1 + m2 )r12 θ̈1 + m2 r1 l2 φ̈ cos(φ − θ1 ) − m2 r1 l2 φ̇(φ̇ − θ̇1 ) sin(φ − θ1 ).
dt ∂ θ1˙
98
we can proceed in two different ways. We may eliminate m coordinates qi with the help of the
auxiliary conditions, thus reducing our system to a free system without kinematical constraints.
Or we may dispense with eliminations and utilise the Lagrangian λ-method. In this method the
integrand L of the given variational problem is modified to L by adding the left-hand sides of
the equations 9.93, after multiplying each equation by some undetermined factor λi (t)7 . Then
the problem is again handled as a free problem, discarding the given auxiliary conditions.
Since the λi are undetermined factors, we can use −λi as factors equally well and write L
in the form
L = L − (λ1 f1 + . . . + λm fm ). (9.94)
Moreover, in ordinary problems of classical mechanics L appears in the form T − V . We can
combine the modification of L with the potential energy V by saying that V has been changed
to V defined as follows:
V = V + λ1 f1 + . . . + λm fm . (9.95)
This elegant mathematical method has a physical counterpart. The fact that we make the
variation problem free after modifying the Lagrangian function L means that we drop the given
kinematical conditions and consider the mechanical system as without constraints. But then
the modification of V to V means that we add to the potential energy of the impressed forces the
potential energy of the forces which maintain the given kinematical constraints. These forces
are given by:
∂
Ki = − (λ1 f1 + . . . + λm fm )
∂qi
∂f1 ∂fm
= − λ1 + . . . + λm . (9.96)
∂qi ∂qi
We see that the Lagrangian λ-method provides the forces of reaction which maintain kine-
matical constraints.
These Ki can be interpreted physically as the components of the force which acts on the
mechanical system in order to maintain the given nonholonomic conditions. This force is now of
7
The λ-method remains valid even if the auxiliary conditions contain not only the qi but the q̇i .
99
Modelling and Control of Multibody Mechanical Systems
a polygenic nature. Once again we see that nonholonomic auxiliary conditions are mechanically
equivalent to polygenic forces.
In view of this equivalence, we conclude that, when properly modified, the Lagrangian
equations of motion are applicable even to the case of polygenic forces. This is indeed the case.
We can characterise a polygenic system of forces by means of the virtual work of these forces.
Let that work be
δw = ρ1 δq1 + ρ2 δq2 + . . . + ρn δqn . (9.99)
The only difference compared with a monogenic force is that this work can no longer be ex-
pressed as the variation of a scalar function. Let us assume that all the impressed forces of
monogenic type are absorbed in the usual way into the Lagrangian function L, while the poly-
genic forces are given by their virtual work 9.99. Then we obtain the equations of motion in
the form
d ∂L ∂L
− = ρi . (9.100)
dt ∂ q̇i ∂qi
Here again, as in 9.97, the polygenic force produces a “right hand side” of the Lagrangian
equations.
9.4.6 Examples
Springy pendulum
Derive the Lagrangian equations of motion of the pendulum in Example 5.1.6.
↓
The position vector of the mass is given by
r = rer
Rolling hoop
Consider a hoop rolling, without slipping, down an inclined plane. Use the Lagrange multiplier
method to represent the rolling constraint and derive the Lagrangian equations of motion. In
this example the constraint of “rolling” is actually holonomic, but this fact will be immaterial
to the discussion. On the other hand, the holonomic constraint that the hoop be on the inclined
plane will be contained implicitly in the choice of generalised coordinates.
x θ
↓
The two generalised coordinates are x, θ, as in Figure 9.2, and the equation of rolling
constraint is
rdθ = dx.
The moment of inertia of the hoop about its centre is given by
Z Z 2π
2
I = ρ r dV = ρ r 3 dθ = 2πρr 3 ,
0
where ρ is the mass per unit length of the hoop. The total mass m is given by 2πρr and
therefore the moment of inertia is
I = mr 2 .
101
Modelling and Control of Multibody Mechanical Systems
The kinetic energy can be resolved into kinetic energy of the motion of the centre of mass plus
the kinetic energy of motion about the centre of mass:
1 1
T = mẋ2 + mr 2 θ̇2 .
2 2
The potential energy is
V = mg(l − x) sin φ,
where l is the length of the inclined plane and the Lagrangian is
mẋ2 mr 2 θ̇2
L=T −V = + − mg(l − x) sin φ.
2 2
Since there is one equation of constraint, only one Lagrange multiplier λ is needed. The
coefficients appearing in the constraint equation are
A1 = r,
A2 = −1.
The two Lagrange equations therefore are
mẍ − mg sin φ − λ = 0,
mr 2 θ̈ + λr = 0,
r θ̇ = ẋ,
r θ̈ = ẍ.
102