System Modeling

Highlight:

Mathematical Modeling of Mechanical Systems Mathematical Modeling of Electrical Systems Linearization of Nonlinear Systems

Real Systems: usually very complicated, nonlinear, and time varying. Mathematical Models: Often very simple, linear, and time-invariant.
We always start by using simple models:

convenience, suciency, can always go back if model is too simple.

Mathematical Modeling of Mechanical Systems

We need to know to model mechanical systems because in many real systems the control rules are eventually applied through mechanical actuators such as motors.

Translational Systems
Simples models are based on Newton's second law, i.e. for each free-body

F = m a F: : a m:
where: vector sum of all forces applied to each body in the system in Newtons (N), vector acceleration of each body in m/s , and mass of the body in kg.

(1)

To use Newton's Second Law

1. Dene coordinates In particular, dene the positive direction. The displacement, velocity, acceleration must be consistent in their positive direction.

2. Use a free-body-diagram to nd all forces. Free-body-diagram is diagram used to show the relative magnitude and directions of all forces acting upon an object. (a) External forces (input, disturbance, etc.), (b) Spring,

x,

where

k is spring constant,
b m, b x,
where

(c) Friction (e.g. piston in a cylinder), 3. Use

b is friction coecient.

F = m, a

to get all dierential equations that describe the dependencies between coor-

dinates and forces. 4. Get the transfer function of interest. Example: One-mass system

Neglect rotational inertia of wheels, friction, etc. to get free-body-diagram:

Apply Newton's Law

u(t) = M x(t)
The Laplace transform

(2)

U (s) = M s2 X(s) X(s) 1 = U (s) M s2

Note that

1 s2

corresponds to a double integrator, i.e.,

x(t)

is much smoother than

u(t).

Example: one-mass system with friction

We model friction as proportional to velocity.

u bx = M x
Take Laplace transform

(3)

U (s) bsX(s) = X(s) = U (s)

Example: Two-mass system

M s2 X(s) 1 2 + bs Ms

Free-body diagram

Figure 1: Free-body diagram

m1 .

How to determine sign?

If If

x2 > x1 , x2 > x1 ,

spring is extended

force is pulling to the right.

m 2 is leaving away from m 1 force is pulling to the right.

Figure 2: Free-body diagram

m2 .

u + k(x2 x1 ) + b(x2 x1 ) = m1 x1 k(x2 x1 ) b(x2 x1 ) = m2 x2

Note: don't need to nd the sign again; simply use reaction force opposite to the acting force. Take Laplace transforms

X1 (s) m1 s2 + bs + k X2 (s) [bs + k] = U (s) X1 (s) [bs + k] + X2 (s) m2 s2 + bs + k = 0

Solve for

X1

and

X2 .

If we are only interested in

X2 .

X1 (s) =

X2 (s) [bs + k] + U (s) [m1 s2 + bs + k]

X2 (s) [bs + k] + U (s) [bs + k] + X2 (s) m2 s2 + bs + k = 0 [m1 s2 + bs + k] bs + k X2 (s) = 4 + (m + m )bs3 + (m + m )ks2 U (s) m1 m2 s 1 2 1 2

Rotational Systems
Newton's Second Law now reads:

M = I
where:

(4)

M : sum of all torques on body (N.m)

torque

= force distance

from the axis of rotation

I : moment of inertia of body around its axis of rotation (kg.m2 )

I
of a point mass

= mass (distance

from the axis of rotation)

of a more complex mass

=
0

r2 dm

: angular acceleration around the axis rotation (rad/s2 )

Figure 3: Rotational systems.

For Figure (3.a)

M =F l
For Figure (3.b)

M = F l mgr sin I = mr2 =

Note: To use

M = I

correctly, need to choose the right axis of rotation,

1. If the axis is static/xed, use it. 2. Use to body's center of mass as the axis. Caution: do not use an axis that is moving unless it is the center of mass. Example: The rotational equivalent of the two-mass system from previous example.

applied torque

=U R

I1 1 = U R + k (2 1 ) + b 2 1 I2 2 = k (2 1 ) b 2 1

Dimension of Freedom
One dimension of freedom

Figure 4: (a) One dimension of freedom, (b) two dimensions of freedom.

F = m x

Two dimension of freedom

Fx Fy

= =

m x m y

Translational and Rotational Systems

Example: Inverted pendulum - segway HT, space shuttle.

Figure 5: (a) Inverted pendulum, (b) Free body diagram.

Objective: apply force

u to keep the inverted pendulum in the upright position. For this system, the object m has both translational and rotational motions.
We can consider it as the combination of translational motion at the center of mass and rotational motion around the center of mass

two sets of equations for object

m.

1. Set up coordinate, Figure (5.a) 2. Free body diagram for

M, Figure (5.b)
U H = Mx

3. Free body diagram for

m, Figure (5.b) m in the horizontal direction?

H=m d2 (x + l sin ) dt2 d2 (l cos ) dt2

Apply Newton's Law on the horizontal direction. Q: What is the displacement of A: Relative to

M : l sin , relative to earth: x + l sin .

On the vertical direction

V mg = m

Around the center of mass

V l sin H l cos = I
There are four equations. Quantities of interest:

x. Additional unknown: V, H

will be eliminated.

d (sin ) = cos dt d d2 (sin ) = (cos ) = sin + cos dt2 dt d (cos ) = sin dt d d2 (cos ) = ( sin ) = cos sin dt2 dt
Hence

U H = Mx

(5)

2 H = m ml sin () + ml cos x 2 V mg = ml cos () ml sin V l sin H l cos = I

Eliminate

(6)

(7)

(8)

H H

with (5) + (6)

2 U = (M + m) ml sin () + ml cos x
Eliminate and

(9)

by substituting (6) and (7) into (8),

mgl sin ml2 ml cos = I x mgl sin ml cos = I + ml2 x

Equations (9) and (10) describe the relationship between There are non-linear terms

(10)

u and x, .

sin () , cos , sin , cos x

Q: How to linearized those equations?

Mathematical Modeling of Electrical Systems

Models of some common elements

Element
Resistor Capacitor Inductor Voltage Source Current Source

Time Domain Laplace Transform Complex Impedance

v(t) = Ri(t) dv(t) i=C dt di(t) v(t) = L dt v(t) = Vs i(t) = Is V (s) = RI(s) I(s) = CsV (s) V (s) = LsI(s) 1 sC sL

Operational Amplier (op-amp)

Figure 6: Ideal operational amplier.

We usually assume that the op-amp is ideal:

e0 = K(e2 e1 ) = K(e1 e2 )
innite input impedance zero output impedance gain

zero current drain by input,

i+ = i = 0

and

e+ = e .

e0

does not change due to load.

K = +

Kirchho's Laws
Governs the relationship between the various currents and voltages.

Kirchho current law (KCL)

The algebraic sum of all currents leaving and coming from/to junction or node in the circuit is zero.

Kirchho voltage law (KVL)

The algebraic sum if all voltages taken around a closed path in a circuit is zero. Example: Use KCL to nd voltages at nodes.

Current equation at each node

Ei E2 E2 Eo C1 sE2 =0 R1 R2 C2 s (Ei Eo ) +
Eliminating

(11)

E2 Eo =0 R2 Ei (s)
and output

(12)

E2 (s),

to obtain the relationship between input

Eo (s),

we have:

Eo R1 R2 C1 C2 s2 + (R1 + R2 ) C2 s + 1 = Ei R1 R2 C1 C2 s2 + ((R1 + R2 ) C2 + R1 C1 ) s + 1
Derive (13) for exercise. Example: Inverting op-amp and ideal. Using KCL

(13)

i1 i2 i3 = 0 ei e e eo d C (e eo ) =0 R1 dt R2
Because

e = 0,

then

deo eo ei +C + =0 R1 dt R2
Take Laplace transform

Ei Eo + CsEo + =0 R1 R2 Eo = Ei
Putting everything together:

1 R1 1 Cs + R2

R2 1 R1 1 + R2 Cs

mechanical electrical block diagram

10

An electro-mechanical system

Armature control of a DC serve-motor

Figure 7: Electro-mechanical system.

1. Input voltage 2.

i through the armature (rotor winding). The current i caused a torque T on the rotor and load. T is proportional to i
ei
sends a certain current

T = ki.
3. The rotor + load is modeled as a rotational system with damping. 4. The rotation of the rotor also generate a back voltage (emf ) velocity

m .

eb

proportional to the angular

This can be viewed as a feedback that impact the armature current .

eb = Kb m
Thus, we have the set of equations as

ei eb = i R + sL T = ki T b = J m m m e = K b b m
where

m =

T Jm s2 + bs eb = sKb m

Jm

is moment of inertia and

b is friction coecient.

We could model this as a block diagram

(s) ei (s)

1 1 k Ls + R Jm s2 + bs 1 1 1 + sKb k Ls + R Jm s2 + bs k 2 + s (bL + RJ ) + (Rb K k)] s [Jm Ls m b

11

Linearization of Nonlinear Systems

A linear function satises the following superposition property

f (x1 + x2 ) = f (x1 ) + f (x2 )

for all

x1 , x2 .

Linear systems are those where the principle of superposition applies

an y (n) + an1 y (n1) + + a1 y (1) + a0 y(t) = bm u(m) + bm1 u(m1) + + b1 u(1) + b0 u(t)
If

then so does

(y1 (t), u1 (t)) and (y2 (t), u2 (t)) both satisfy (y1 (t) + u1 (t), y2 (t) + u2 (t)).

the dierential equation,

We can rewrite a linear system as

f (y, y, , y (n) , u, u, , u(m) ) = 0

where

f ( ) = an y (n) +an1 y (n1) + +a1 y (1) +a0 y(t) bm u(m) + bm1 u(m1) + + b1 u(1) + b0 u(t)
Then

f (.)

is a linear function of its variables.

IMPORTANT: We can take Laplace transform only for linear systems.

Non-linear systems are those where the principle of superposition does NOT apply.

Example: Non-linear systems 1. Inverted pendulum, equation (9) and (10).

2 U = (M + m) ml sin () + ml cos x mgl sin ml cos = I + ml2 x

If we rewrite these equations as

f (, , , x, x, x, u) = 0
The function

is NOT a linear function of its variables.

2. Non-linear circuit.

Non-linear resistor

VR = i3 R. Vi = i3 R + L

di dt

We CANNOT take Laplace transform for non-linear systems. However, we could derive a linear system that is an approximation of the original system, then take Laplace transform.

12

Linearization
Basic idea: If the system operates only around a small neighborhood of an equilibrium points, by expanding the nonlinear functions into a Taylor series about the equilibrium point, we can retain the linear terms, and neglect the higher-order terms. In this way we obtain linear approximation of the original system. Such linear model would work, if the deviation of the signals from the equilibrium point is small. We can the study the linear system by taking Laplace transform.

Linearization of a function
Suppose

y = f (x),

f
y

is non-linear. Take Taylor series expansion, assuming all derivatives exist.

= =

f (x) f (x) + df dx (x x) +
x=x

1 d 2f 2! dx2 x

(x x)2 +
x=x

derivatives evaluated at when

(constant)

|x x|

is small,

df y f (x) + = dx
Letting

(x x)
x=x

y = y f (x),

and

x = x x,

we have

y=
Deviations from

df dx

x
x=x

f (x)

of the output

y are linear in deviations from x of the input x.

Linearization of a function with multiple variables

y = f (x1 , x2 , u1 , u2 )

13

f (x1 , x2 , u1 , u2 ) + f = x1 + f x2

x1 = x1 u1 = u1

x2 = x2 (x1 x1 ) u2 = u2 x1 = x1 u1 = u1 x2 = x2 (u1 u1 ) u2 = u2

x1 = x1 u1 = u1

f x2 = x2 (x2 x2 ) + u1 u2 = u2 x2 = x2 (u2 u2 ) u2 = u2

++

f u2

x1 = x1 u1 = u1

The coecient in front of each variable is simply the partial derivative evaluated at the equilibrium point.

Procedures for linearization

Given the dierential equation 1. Find equilibrium point by setting all

x, x, u, u,

to zero

solve for the equilibrium point.

2. For each term that is non-linear, view it as a non-linear function of the variables Evaluate the partial derivative at the equilibrium point: linear approximation.

x, x, x, u, u, u, .

x, 0, 0, u, 0, 0, .

Write down the

3. Simplify the linear equations. Replace each variables by its deviation from the equilibrium.

Linearization of the dierential equation of one variable

nonlinear. 1. Find equilibrium point of the system by setting the system. (Idea: If Hopefully, if

Suppose

dx = f (x(t)), dt

where

x = 0 f (x) = 0.

The set of solutions that satisfy the above equation is called the set of equilibrium points of

x(0)

is close to

x(0) = x, then x(t) = x for all t. The system is at equilibrium. x, x(t) will stay close to x for at least some interval (0, T )).

2. Evaluate partial derivatives for each non-linear term

dx df f (x) + dt dx
3. Simplify the linear equation. Note that Let

(x x)
x=x

f (x) = 0

by denition of equilibrium points

x.

x = x x.

Then

d d df x = x x = dt dt dx

x
x=x

Linearization of a dierential equation with inputs and higher order derivatives

d2 x(t) = f (x(t), x(t), u(t)) dt2
1. Find the equilibrium points. For a reference (constant) input solutions of

uref ,

the equilibrium points are

f (xeq , 0, ueq ) = 0

14

2. Evaluate partial derivatives for non-linear terms

d2 x(t) dt2

= f (xeq , 0, ueq ) +

f (x(t) xeq ) x x = xeq x=0 u = ueq

f f (x(t) 0) + (x(t) xeq ) x x = xeq x x = xeq x=0 x=0 u = ueq u = ueq

Let

3. Simplify. Note that

f (xeq , 0, ueq ) = 0.

x = x xeq , u = u ueq

x =

f f f x + x + u x x = xeq x x = xeq x x = xeq x=0 x=0 x=0 u = ueq u = ueq u = ueq

Rearrange

Example:

Vi = i3 R + L

di . dt

di Vi R = i3 dt L L Vref = 1 Vi R i3 = 0 L L ieq =
3

1. Find equilibrium points. Suppose Set

di = 0, dt

we have

Vref = R

1 R

2. Evaluate partial derivatives. Let

f (i) =

R 3 i L f i (i ieq )
i=ieq

f (i) f (ieq ) + f i

=
i=ieq

3R 2 i L

=
i=ieq

3 3 R L

Hence, the linear approximation is

di dt

= = =

Vi + f (i) L Vi f + f (ieq ) + (i ieq ) L i i=ieq Vi 3 3 + f (ieq ) R(i ieq ) L L

3. Simplify. Let

Vi = Vi Vref , i = i ieq i = Vi Vref 3 3 + + f (ieq ) Ri L L L

=0

3 Vi 3 Ri L L

15

Take Laplace transform

si(s) i(s) s + 3 3 R L i(s) Vi (s)

Example: pendulum

= =

Vi (s) 3 3 Ri(s) L L Vi (s) L 1 L 3 s+ 3R L

I = ul cos mgl sin bl

1. Find equilibrium points Suppose

uref = 0.

Set

, = 0.

We have

mgl sin = 0 = 0.
2. Evaluate partial derivatives. Let

f (, u) = ul cos mgl sin f (, u) = f (, uref ) + f u = uref ( ) =0

f (u uref ) u u = uref =0

u = uref =0

ul sin mgl cos | u = u = mgl ref =0 l cos | u = u =l ref =0

f u u = uref =0

Hence, the linear approximation is

I = f (, uref ) mgl( ) + l(u uref ) bl

16

3. Simplify. Note that Let

and

f (, uref ) = 0. u = u uref I = mgl + lu bl

Take Laplace transform

s2 I(s) (s) s I + sbl + mgl (s) u(s)

2

= mgl(s) + lu(s) sbl(s) = lu(s) = l s2 I + sbl + mgl