ELEN90055 Control Systems

The University of Melbourne

Problem Set 2 with Reference Solutions

Consider the inverted pendulum system shown in Figure 1. The pendulum is mounted on the
shaft of an armature-controlled DC servomotor. A simplified schematic model for the motor is shown
in Figure 3.

mg
l θ

Figure 1: The Pendulum System.

I
b
R L J
θ
v vb

Figure 2: The armature-controlled DC servomotor

In Problem Set 1 (Problem 2) it was established that the transfer function for the linearized model
. .
with incremental input δv = v − vq and incremental output δθ = θ − θq is given by

−Kt
G(s) := .
(Ls + R)((Jm + ml2 )s2 + bs − mgl(cos θq ))

For the particular equilibrium (Iq , θq , vq ) = (0, 0, 0) and the parameter values L = b = 0, R = Kb =
Kt = l = 1, Jm = 0.1, m = 0.9, and g = 10 (all with appropriate units), the transfer function becomes
−1
G(s) = .
(s2 − 9)

1. Is this incremental model G(s) stable or unstable?

1
PSfrag
Di (s)

+
R(s) C(s) G(s) ∆θ (s)
−

Figure 3: Feedback Control System.

2. Consider the feedback control

. loop shown in Figure 3, where the plant G(s) is as above and the
compensator C(s) = P (s) L(s) where P (s) = K(s + 3), L(s) = (τ s + 1), K is a constant gain
and τ is a time-constant (i.e. both of the latter are real numbers and τ > 0). Find (closed-loop)
transfer functions T (s) and Si (s) such that ∆θ (s) = T (s) · R(s) + Si (s) · Di (s).

3. For what values of the controller parameters K and τ are both T (s) and Si (s) stable transfer
functions?

4. Derive an expression for the steady-state error (i.e. limt→∞ (r(t)−δθ (t)) for a unit step reference
r(t) in terms of K and τ ?

5. How should K and τ be set to achieve steady-state step-reference error of less than 1%.

6. How should K and τ be set to achieve an overshoot of less than 30% in the step-reference
response at the plant output?

7. Consider an input disturbance signal di (t) = 0.1 sin(100πt) for all t > 0. What is the steady-
state influence of this on the incremental output signal δθ in terms of K and τ ?

Reference Solution:

1. The transfer function G(s) = (s2−1−9)

−1
= (s−3)(s+3) has two poles: s = −3 and s = 3. Since one of
these has positive real part the corresponding linear time-invariant system is unstable.

2. From the block-diagram it follows that:

∆θ (s) = G(s) · (Di (s) + C(s) · (R(s) − ∆θ (s))) .

Equivalently,

(1 + G(s)C(s))∆θ (s) = G(s) · (Di + C(s) · R(s)) ,

whereby

G(s)C(s) G(s)
∆θ (s) = · R(s) + · Di (s).
1 + G(s)C(s) 1 + G(s)C(s)
G(s)C(s) G(s)
With T (s) = 1+G(s)C(s) and Si (s) = 1+G(s)C(s) it follows that ∆θ (s) = T (s) · R(s) + Si (s) · Di (s).
For the specified parameters,
K(s+3) −1
(τ s+1) · (s−3)(s+3) −K −K
T (s) = = = 2
1 + K(s+3) −1
(τ s+1) · (s−3)(s+3)
(τ s + 1)(s − 3) − K τ s + (1 − 3τ )s − (K + 3)

2
 and
−1
(s−3)(s+3) −(τ s + 1)
Si (s) = = .
1+ K(s+3) −1 (s + 3)(τ s2 + (1 − 3τ )s − (K + 3))
(τ s+1) · (s−3)(s+3)

3. From above it can be seen that T (s) and Si (s) share two poles at the roots of the quadratic
.
χ(s) = τ s2 + (1 − 3τ )s − (K + 3). In addition, Si (s) has a pole at s = −3. In view of this,
stability of the two transfer functions is equivalent to the roots of χ(s) having negative real
part. Let λ1 and λ2 be the roots of χ(s) = τ (s − λ1 )(s − λ2 ); as such,

λ1 λ2 = −(K + 3)/τ and − (λ1 + λ2 ) = (1 − 3τ )/τ.

These roots are either both real or λ1 = λ̄2 , where ¯· denotes complex conjugate. In the former
case, both roots are negative if, and only if, λ1 λ2 > 0 and −(λ1 + λ2 ) > 0. In the latter case,
λ1 λ2 = λ̄2 λ2 = |λ2 |2 > 0 and the real part is negative if, and only if, −(λ1 + λ2 ) = −(λ̄2 + λ2 ) =
−2ℜ(λ2 ) > 0. Therefore, T (s) and Si (s) are both stable if, and only if,

−(K + 3)/τ > 0 and (1 − 3τ )/τ > 0.

Equivalently,

K < −3 and τ < 1/3,

because τ > 0.

4. The steady-state unit step-reference error is finite if, and only if, T (s) is a stable transfer
function; that is, K < −3 and τ < 1/3. Otherwise, the error grows unboundedly for a step at
.
the reference input. Let e(t) = r(t) − δθ (t) and suppose that di (t) = 0 for all t > 0, so that
taking the Laplace transform,

E(s) = R(s) − ∆θ (s) = (1 − T (s)) · R(s).

Applying the final-value theorem with R(s) = 1/s (the Laplace transform of a unit step),
.
eSS = lim e(t) = lim sE(s) = lim s · (1 − T (s)) · (1/s) = 1 − lim T (s).
t→∞ s→0 s→0 s→0

In terms of the controller parameters K and τ ,

−K 3
eSS = 1 − T (0) = 1 − = .
−(K + 3) K +3

5. In order to get a steady-state error of less than 1% the transfer function T (s) must be stable
and |eSS | = |3/(K + 3)| < 0.01; that is K < −3 and |K + 3| > 3/0.01, or equivalently,
K < −3 − 3/0.01. Note that the controller parameter τ has no influence on the step-reference
error in steady-state, provided τ < 1/3 for stability. Of course, if τ > 1/3, then the step-
reference error would grow without bound over time.

6. From Workshop 1 (Exercise 3) the overshoot for the unit step response of a canonical second
2
ωn
order dynamical system with transfer function s2 +2ψω 2 , where the damping ratio 0 < ψ < 1,
n +ωn
p
is given by given by Mp = exp(−πψ/ 1 − ψ 2 ) (this is a fraction of 1 which is the final value
as t → ∞). Note that
ωn2
T (s) = K̂ 2
s + 2ψωn s + ωn2

3
 for ωn2 = −(K +3)/τ , ψ = ((1−3τ )/τ )(1/2ωn ) and K̂ = (−K/τ )(1/(−(K +3)/τ )) = K/(K +3).
So provided the controller parameters are stabilizing (i.e. K < −3 and τ < 1/3), the final
value of the step response T (s) · (1/s) is K̂ = K/(K + 3) and the peak value is simply K̂
times the peak value of the canonical step response (at the same time), so that the percentage
overshoot (relative
p to the final value) is the same. To achieve less than 30% overshoot, we need
exp(−πψ/ 1 − ψ 2 ) < 0.3 ⇔ ψ > 0.36.

7. Note that di (t) = 0.1(exp(j100πt)−exp(−j100πt))/2j. We know that the steady-state response

of a stable transfer function Si (s) to an input ejωt is given by |Si (jω)|ej∠Si (jω) ejωt for any
frequency −∞ < ω < ∞. So by linearity, provided Si (s) is stable, the steady-state response to
the specified di (t) is given by

δθ,SS (t) = 0.1(|Si (j100π)|ej∠Si (j100π) ej100πt − |Si (−j100π)|ej∠Si (−j100π) e−j100πt )/2j
= 0.1|Si (j100π)|(ej(100πt+φ) − e−j(100πt+φ) )/2j = 0.1|Si (j100π)| sin(100πt + φ),

where φ = ∠Si (j100π). Note that since the coefficients of the numerator and denominator
polynomials of Si (s) are real (i.e. Si (s) is real rational), it follows that φ = −∠Si (−j100π)
and |Si (j100π)| = |Si (−j100π)|, whereby the second equality above holds. To summarize, in
steady-state the incremental signal δθ (·) will be a sinusoid with frequency 100π radians/sec,
which lags the disturbance input by −φ radians (i.e. it leads by φ radians) where

φ = tan−1 (ℑ(Si (j100π))/ℜ(Si (j100π))).

This sinusoidal will have magnitude

(τ (j100π) + 1)
0.1 .
(3 + j100π)(−τ (100π)2 − (K + 3) + j(1 − 3τ )100π)