Proceedings of the 19th World Congress

The International Federation of Automatic Control

Cape Town, South Africa. August 24-29, 2014

Performance-guaranteed Robust PID Controller Design

for Systems with Unstable Zero
Š. Bucz, V. Veselý, A. Kozáková, Š. Kozák

Faculty of Electrical Engineering and Information Technology,

Slovak University of Technology in Bratislava, Ilkovičova 3, SK-812 19 Bratislava, Slovak Republic
(e-mail: stefan.bucz@stuba.sk, vojtech.vesely@stuba.sk, alena.kozakova@stuba.sk, stefan.kozak@stuba.sk )

Abstract: The paper deals with the development of a new robust PID controller design method that guarantees
designer-specified maximum overshoot and settling time for non-minimum phase processes with unstable zero.
The PID controller design provides guaranteed gain margin GM. The parameter of the tuning rules is a
suitably chosen point of the plant frequency response obtained by a sine-wave signal with excitation frequency
ωn. Then, the designed controller moves this point into the phase crossover with the required gain margin GM.
The couple (ωn;GM) is specified with respect to closed-loop performance requirements in terms of ηmax
(maximum overshoot) and ts (settling time) according to developed parabolic dependences. The new approach
has been verified on a vast batch of benchmark examples; subsequently, the developed algorithm has been
extended to robust PID controller design for plants with unstable zero and unstructured uncertainties.
Keywords: Gain margin, PID tuning, guaranteed performance, unstable zero

1. INTRODUCTION margin of the plant decreases, and the phase crossover moves
The proposed new method is applicable for control of linear closer to (-1,j0). Due to significant changes of the gain margin
single-input-single-output non-minimum phase systems even of the plant brought about by the non-minimum phase
with unknown mathematical model with unstructured behavior, it is beneficial to use gain margin GM as a
performance measure when designing the PID controller.
uncertainties. A survey on PID controller tuning can be found
in (Åström and Hägglund, 1995), (Åström and Hägglund, Nyquist Diagram

2000), (Blickley, 1990), (Grabbe et al., 1959-61), (Karaboga 1

α=0.1 α=1
and Kalinli, 1996), (Kristiansson and Lennartson, 2002), 0.8 α=0.2 M1 α=2
(Morilla and Dormido, 2000), (O’Dwyer, 2000), (Tinham, 0.6
α=0.5 α=5
1989), (Veselý, 2003), (Visioli, 2006), (Yu, 2006), in the 0.4
famous paper (Ziegler and Nichols, 1942) and references
0.2
therein. The control objective is to provide required nominal
Imaginary Axis

maximum overshoot ηmax and settling time ts of the controlled

0

-0.2
process variable y(t). The key idea behind guaranteeing
specified values ηmax and ts consists in extending validity of the -0.4

relations ηmax=f(GM) and ts=f(ωn) derived for 2nd order systems -0.6

(Reinisch, 1974) for arbitrary plant orders; two-parameter -0.8

quadratic dependences were obtained for both the maximum -1

overshoot ηmax=f(GM,ωn) and settling time ts=f(GM,ωn). The -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1

resulting plots called B-parabolas enable the designer choosing Real Axis

such a couple (GM,ωn) that guarantees fulfillment of specified Fig. 1. Nyquist plots of G(s)=(1-αs)/(Ts +1)n for n=3, T=1
performance requirements thus allowing consistent and and different values of α
systematic shaping of the closed-loop step response with
regard to the controlled plant (Bucz and Kozáková, 2012). Consider a multipurpose loop shown in Fig. 2 (the switch in
position SW=1). Let G(s) be transfer function of an uncertain
2. PID CONTROLLER DESIGN OBJECTIVES FOR non-minimum phase plant, and GR(s) the PID controller.
PROCESSES WITH UNSTABLE ZERO
Sine-wave gen.
It is a well known difficulty to control the class of non-minimum
phase systems G(s)=(1-αs)/(1+Ts)n with unstable zero z=+1/α, w(t) e(t) 2 u(t) y(t)
PID controller
1
G(s)
even for small values of α; moreover, control complexity SW
increases with increasing α (Vítečková et al., 2000). Fig. 1 - 3
Relay
shows Nyquist plots of the non-minimum phase plant G(s) for
n=3 and T=1, with an unstable zero (α=0.1,0.2,0.5,1,2,5 are
considered). Fig. 1 reveals, that with increasing α the gain Fig. 2. Multipurpose loop for the designed sine-wave method

978-3-902823-62-5/2014 © IFAC 4933

19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014

The corresponding closed-loop characteristic equation a particular point of the plant frequency response
c(s)=1+L(s)=1+G(s)GR(s)=0 expresses the closed-loop corresponding to the excitation frequency ωn
stability can easily be broken down into the magnitude and
phase conditions G( jωn ) = G( jωn ) e j arg G( ωn ) = [Yn U n ]e jϕ( ωn ) (10)

G( jω*p ) GR( jω*p ) =1 GM , argG( ω*p ) + argGR ( ω*p ) = −π , (1) can be plotted in the complex plane.

where GM is required gain margin, L(jω) is the open-loop Im

transfer function, and ωp* is the open-loop phase crossover M1
frequency. Denote ϕ=argG(ωp*),Θ=argGR(ωp*), and consider 1
the ideal PID controller in the form
GM
 1 
GR ( s ) = K 1 + + Td s  , (2) -1 ωn 0 1 Re
 T s  ϕ G(jω
ω)
i
LP
where K is the proportional gain, and Ti, Td are integral and
derivative time constants, respectively. After comparing the Θ G
two forms of the PID controller frequency transfer functions ωn
 1 
GR ( jω*p ) = K + jK Td ω*p − , (3) ω)
L(jω
 Tiω*p 
Fig. 3. Graphical representation of the PID tuning principle
GR ( jω*p ) = GR ( jω*p ) [cosΘ + j sinΘ ] , (4)
Excitation frequency ωn is taken from the interval
PID coefficients can be obtained from the complex equation
at ω=ωp* ωn ∈ 0.5ωc ,1.25ωc , (11)
 1  cosΘ sinΘ where the plant critical frequency ωc can be obtained by the
K + jK Td ω*p − * 
= +j (5)
 Tiω p  GM G( jω p )
*
GM G( jω*p ) well-known relay experiment (Åström and Hägglund, 1995),
i.e. for SW=3.
using the substitution |GR(jωp*)|=1/[GM|G(jωp*)|] resulting
Using the PID controller with the coefficients {K;Ti=βTd;Td},
from (1a). The complex equation (5) is then solved as a set of
two real equations the identified point G(jωn) with coordinates (10) can be
moved into the phase crossover LP≡L(jωp*) on the negative
cos Θ  1  sinΘ real half-axis, where the required gain margin GM is
K= , K Td ω*p − * 
= , (6) guaranteed (Fig. 3), if the following identity between the
GM G( jω ) *
 Tiω p  GM G( jω*p )
p
excitation and phase crossover frequencies ωn and ωp*,
where (6a) is a general rule for calculating the controller gain respectively, is fulfilled
K; substituting (6a) into (6b), a quadratic equation in Td is ω *p = ω n . (12)
obtained

( ) 2 1 T Considering (11), the following relations result

Td2 ω *p − Td ω*p tgΘ − = 0 , where β = i . (7)
β Td G( jω*p ) = G( jωn ) , arg G( ω*p ) = arg G( ωn ) = ϕ , (13)
Expression for calculating Td is the positive solution of (7) Θ = −180° − arg G( ωn ) (14)
tgΘ 1 tg 2Θ 1 and the phase crossover coordinates are
Td = + + . (8)
2ω*p ω *p 4 β LP=[|L(jωn)|,argL(ωn)]=[1/GM,-180°]. Substituting (13a) into
(6a) and (12) into (8), the PID controller coefficients
Hence, (6a), (7b) and (8) are the resulting PID tuning rules, guaranteeing the required gain margin GM are obtained using
where the angle Θ is obtained from the phase condition (1b) the sine-wave type tuning rules expressed in the following form
Θ = −180° − arg G( ω *p ) = −180° − ϕ . (9)
cosΘ tgΘ 1 tg 2Θ 1
K= , Td = + + , (15)
3. PLANT IDENTIFICATION BY A SINUSOIDAL GM G( jωn ) 2ω n ω n 4 β
EXCITATION INPUT β = 4 , Ti = βTd , Θ = −180 ° − ϕ . (16)
Consider again Fig. 2; if SW=2, a sinusoidal excitation signal
u(t)=Unsin(ωnt) with magnitude Un and frequency ωn is 4. CLOSED-LOOP PERFORMANCE UNDER THE
injected into the plant G(s). The plant output DESIGNED PID CONTROLLER
y(t)=Ynsin(ωnt+ϕ) is also sinusoidal with magnitude Yn, This section answers the following question: how to transform
where ϕ is the phase lag between y(t) and u(t). After reading the maximum overshoot ηmax and settling time ts as required by
the values Yn and ϕ from the recorded values of u(t) and y(t), the designer into the couple of frequency-domain parameters

4934
 19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014

(ωn,GM) needed for identification and PID controller tuning? Substituting ωn=σωc we obtain relation for the relative
Consider typical gain margins GM given by the set settling time
{G } = {3dB ,5dB ,7 dB ,9dB ,11dB ,13dB ,15 dB ,17 dB} ,
Mj (17)
t sωc =
π π
γ ⇒τs = γ , (22)
j=1…8; let us split (11) into 5 equal sections of the size σ σ
∆ωn=0,15ωc and generate the set of excitation frequencies where ts is related to the plant critical frequency ωc. Due to
{ωnk } = {0.5ωc ,0.65ωc ,0.8ωc ,0.95ωc ,1.1ωc ,1.25ωc } , (18) introducing ωc, the l.h.s. of (22a) is constant for the given
plant and independent of ωn. The dependence (22b) obtained
k=1…6; its elements divided by the plant critical frequency empirically for different excitation frequencies ωnk is depicted
ωc determine excitation levels σk=ωnk/ωc given by the set in Fig. 5b and Fig. 6b, respectively; it is evident that with
{σ k } = {0.5,0.65,0.8,0.95,1.1,1.25} , (19) increased phase margin GM at every excitation level σ the
relative settling time τs first decreases and after achieving its
k=1…6. Fig. 4 shows the closed-loop step response shaping minimum τs_min, it increases again. Consider the benchmark
for different GM and ωn using the PID controller design for plants G1(s) and G2(s) with following parameters: G1.1(s):
the plant (20b) with parameters T2=0.75, α2=1.3, and (T1,n1,α1)=(0.75,8,0.2); G1.2(s): (1,3,0.1); G1.3(s): (0.5,5,1);
required gain margins GM=5dB, 9dB, 11dB and 13dB at G2(s): T2=0.5, α2=1.3. Couples of examined plants [G2(s),
different excitation levels σ1=ωn1/ωc=0.5, σ3=ωn3/ωc=0.8 G1.3(s)] and [G1.2(s), G1.1(s)] differ principally by the ratio
and σ5=ωn5/ωc=1.1. α/T, which for the 1st couple is [α2/T2=2.6, α1.3/T1.3=2] and
for the 2nd couple [α1.2/T1.2=0.1, α1.1/T1.1=0.27]. Hence, the
Closed-loop step responses y(t), ωn/ωc=0.5
c Closed-loop step responses y(t), ωn/ωc=0.8
c
1.5 1.5 ratio of the parameter α and the (dominant) time constant T
a) b) of the plant is significant for the closed-loop performance
1 1
assessment under the PID controller designed for a plant with
GM=5dB GM=5dB
unstable zero. Based on the previous analysis of design
y(t)

=5dB
y(t)

0.5 0.5 M
GM=9dB GM =9dB
M=9dB results of a series of benchmark examples, unknown plants
GM=11dB
=11dB
0
GM=11dB
0
G M
GM=13dB
=13dB
with unstable zero can be classified according to the ratio α/T
GM=13dB GM
in following two groups:
-0.5
1. plants with the ratio α/T<0,3;
-0.5
0 10 20 30 40 0 10 20 30 40
Time (s) Time (s)

Closed-loop step responses y(t), ωnn/ωc=1.1 Time responses of the plant G(s)
2. plants with the ratio α/T>0,3.
c
1.5 1.6

c) 1.4
d) According to this classification, empirical dependences
1
1.2

1
ηmax=f(GM), τs=f(GM) for non-minimum phase systems with
0.8 an unstable zero were constructed for different open-loop
gain margins GM and excitation levels σ, and are depicted in
G
GMM=5dB
=5dB
y(t)

y(t)

0.5 0.6

GM=9dB
G =9dB
M 0.4

GM=11dB
G =11dB α/T=0.1 Fig. 5a (for α/T>0.3), and Fig. 6a (for α/T<0.3). The network
α/T=1
M 0.2
0 GM=13dB
=13dB
GM 0
of dependences shows that increasing gain margin GM brings
about decreasing of ηmax.
-0.2

-0.5 -0.4
0 10 20 30 40 0 5 10 15

Time (s) Time (s)

As the empirical dependences in Fig. 5 and Fig. 6 were
Fig. 4. a)-c) Closed-loop step responses of G2(s) with T2=0.75, approximated by quadratic regression curves they are called
α2=1.3 for various GM and ωn; d) Time responses of G2(s) for B-parabolas (Bucz and Kozáková, 2012). B-parabolas are a
α/T=1 and α/T=0.1 during the relay test useful design tool to carry out the transformation
ℜ:(ηmax,ts)→(ωn,GM) that enables to choose appropriate
Consider the following benchmark plants values of gain margin GM and excitation frequency ωn,
−α1s +1 −α2s +1 respectively, to guarantee the performance specified by the
G1( s ) = , G2 ( s ) = . (20)
( T1s +1)n 1
( s +1)(T2s +1)(T22s +1)(T23s +1) designer in terms of maximum overshoot ηmax and settling
time ts (Bucz and Kozáková, 2012). Note that pairs of
The proposed method has been applied for each element of
B-parabolas at the same level (Fig. 5a, Fig. 5b) or (Fig. 6a,
the Cartesian product ωnk×GMj of the sets (18) and (17) for
Fig. 6b) are to be used. When a real plant with an unstable
j=1...8 and k=1...6. Significant differences between dynamics
zero is to be controlled, the ratio α/T cannot be specified
of individual control loops under designed PID controllers
exactly due to unavailability of the plant model. To decide to
can be observed for the benchmark systems (20).
which category a given plant belongs (α/T>0.3 or α/T<0.3) it
is sufficient to analyze the rise portion of the output variable
The settling time ts can be expressed by the relation
during the relay test for finding ωc. If y(t) has an S-form with
γπ a tiny undershoot, the plant is included in the category
ts = , (21)
ωn α/T<0.3 and B-parabolas from Fig. 6 are to be used. If a
considerable undershoot of y(t) occurs having a “square root
where γ is the curve factor of the step response. To examine sign” form (Fig. 4d in the red dashed ellipse), the plant
settling times of closed-loops for various plant dynamics, it is belongs to the category α/T>0.3 and its performance will be
advantageous to define the relative settling time τs=tsωc. assessed using B-parabolas in Fig. 5.

4935
19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014

Závislost
Dependences η ωn -=f(G
ηmax =f(GM) pre rôzne systémy) sfor
nestabilnou
different ωα /T>0.3
nulou, τreg=f(GM) pre rôzne τωn=f(G
ZávislostDependences - systémy s nestabilnou
) for different ω α /T>0.3
nulou,
90
max M n
45
s M n

R+L=χ
χLRL Im
ωn=0.5ωc ωn=0.5ωc
80 40
LP ω)

Relative settling time τs=ωcts=f(GM,ωn)

ωn=0.65ωc
ωn=0.65ωc
+ + L0(jω
Maximum overshoot ηmax=f(GM,ωn)

RL L1 L2
70 ωn=0.8ωc ωn=0.8ωc
35

60
ωn=0.95ωc
30
ωn=0.95ωc
-1 0 ω) Re
G0(jω
ωn=1.1ωc ωn=1.1ωc
50
ωn=1.25ωc ωn=1.25ωc
R+L L0 L3 ϕ0
40
25
ML + Θ0 + +G2
20 ωn G1
30
GN
15
20
RG G0 G3+
10 10
|1+L0(jωn)|=[GM-1]/GM |L0(jωn)|=1/GM
0 5
ωn MG
3 4 5 6 7 8 9 10 11 12 13 4 6 8 10 12 14 16
Požadovaná amplitúdová bezpecnost G [dB] Požadovaná
Required gain margin GM [dB] Requiredamplitúdová bezpecnost
gain margin G G [dB]
[dB]
M

Fig. 7. Dispersion circles MG and ML

Fig. 5. B-parabolas: a) ηmax=f(GM); b) τs=ωcts=f(GM) for
identification levels ωnk/ωc, k=1,2,3,4,5,6 valid for non- The proposed control law generated by the robust controller
minimum phase systems with the ratio α/T>0.3 GRrob(s) designed for the nominal point G0(jωn) actually carries
90
Závislost ηmax =f(GM) pre rôzne
Dependences η ωn -=f(G
max
systémy) sfor
M
nestabilnou
different ωα /T<0.3
nulou,
n
45
τreg=f(GM) pre rôzne τ
ZávislostDependences ωsn=f(G M) for different ωn
- systémy s nestabilnou nulou, α /T<0.3
out the transformation ℘:{RG→RL:RL=|GRrob|RG} of the set of
80
ωn=0.5ωc
40
ωn=0.5ωc
identified points Gi(jωn) encircled by MG with the radius RG
Relative settling time τs=ωcts=f(GM,ωn)

ωn=0.65ωc ωn=0.65ωc

into the set of points Li(jωn) delimited by ML, and also

Maximum overshoot ηmax=f(GM,ωn)

70 ωn=0.8ωc ωn=0.8ωc
35
ωn=0.95ωc
calculates the radius RL≡RL(ωn) of the dispersion circle ML
ωn=0.95ωc
60
ωn=1.1ωc 30 ωn=1.1ωc

corresponding to the points Li(jωn) of the Nyquist plot so as to

50 ωn=1.25ωc
ωn=1.25ωc
25
40

30
20
guarantee fulfillment of the robust stability condition. The
20
15
robust PID controller is designed using the sine-wave method
10 10
described in sections 2 and 3; the input data for the nominal
0
4 6 8 10 12 14 16 18 20 22
5
4 6 8 10
Požadovaná
12 14 16 18 20 22 24 model G0(jωn) are its coordinates: {|G0(jωn)|; ϕ0=argG0(ωn)}.
Required gain margin GM [dB] Requiredamplitúdová bezpecnost
gain margin G G [dB]
[dB]
M
Substituting them into (15) and (16) the following expressions
Fig. 6. B-parabolas: a) ηmax=f(GM); b) τs=ωcts=f(GM) for for calculating robust PID controller parameters are obtained
identification levels ωnk/ωc, k=1,2,3,4,5,6 valid for non-
minimum phase systems with the ratio α/T<0.3 cosΘ0 tgΘ0 1 tg2Θ0 1
Krob = , Td rob = + + , (26)
GM G0 ( jωn ) 2ωn ωn 4 β
5. ROBUST SINE-WAVE TYPE PID CONTROLLER Ti rob = βTi rob , Θ 0 = −180° − ϕ 0 , β=4. (27)
DESIGN
It can be seen that the gain margin GM appearing in (26a) is at
The main idea of the uncertain plant identification consists in
the same time a robust PID controller tuning parameter
repeating the sine-wave type excitation for individual
required for guaranteeing robust stability.
uncertainty changes using the excitation signal frequency ωn
yielding a set of identified points Gi of the uncertain plant Theorem 1 (Sufficient condition of robust stability under
frequency responses a PID controller)
Gi ( jωn ) = Gi ( jωn ) e j arg G ( ω ) = ai + jbi , i=1,2...N. i n
(23) Consider an uncertain continuous-time stable dynamic
system described by unstructured uncertainty. The closed-
Plant parameter changes are reflected in magnitude and phase loop system T(s) under the controller GR(s) is robustly stable
changes |Gi(jωn)| and argGi(ωn), where i=1...N; N=2p is the if the nominal closed-loop system (G0(s) under a PID
number of identification experiments and p is the number of controller GR(s)) is stable, and
varying technological quantities of the plant. The nominal
plant model G0(jωn) at ωn is obtained as mean values of real χ L RG ( ω n )
GM > 1 + , (28)
and imaginary parts of Gi(jωn), respectively G 0 ( jω n )

where GM is the required gain margin, ωn is the excitation

N N
1 1
G0 ( jωn ) = a0 + jb0 =
N
∑ a + j N ∑b , i=1,2...N,
i =1
i
i =1
i (24)
frequency, χL is the safety factor, RG(ωn) is the radius of the
dispersion circle of the Nyquist plots of the plant at ωn, and
where |G0(jωn)|=(a0 +b0 ) ,ϕ0(ωn)=argG0(ωn)=arctg(b0/a0). 2 2 0,5
G0(jωn) is a point on the Nyquist plot of the nominal plant at ωn.
The points Gi representing unstructured uncertainties of the
plant can be enclosed in the circle MG centered in Proof
G0(jωn) with the radius RG≡RG(ωn) obtained as a maximum
distance between the i-th identified point Gi(jωn) and the The proof can easily be performed according to Fig. 7. If the
nominal point G0(jωn) nominal open-loop L0(s)=G0(s)GR(s) is stable, then according

RG = max
i
{ (a − a ) + (b − b ) }, i=1,2...N.
i 0
2
i 0
2
(25)
to the Nyquist stability criterion the closed-loop with the
uncertain plant will be stable if the distance between L0 and
the point (-1,j0), i.e. |1+L0(jωn)| is greater than the radius
The dispersion circle MG centered in the nominal point G0 RL(ωn) of the circle ML centered in L0, i.e.
with the radius RG encircles all identified points Gi of the
uncertain plant (Fig. 7). RL ( jω n ) < 1 + L0 ( jω n ) , (29)

4936
19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014

where ωn is the sine-wave generator frequency. The distance 3. Using the sine-wave method, eight points of Nyquist plots of
|1+L0(jωn)| is a complementary distance |0,L0|=|L0| to the unit the uncertain plant were identified at ωn=0,65ωc=0,65.0,04880=
value. Thus =0,03172 s-1: G31(jωn)...G38(jωn) (depicted by blue „x“ in
Fig.11). The nominal point G30(jωn), which position was
L0 ( jωn ) + 1 + L0 ( jωn ) = 1 , 1 + L0 ( jωn ) = 1 − L0 ( jωn ) . (30)
calculated from the coordinates of identified points G3i(jωn),
From the principles of the proposed PID controller tuning i=1...8, is located on the Nyquist plot of the nominal model
method results that the robust controller shifts the nominal G30(jωn) (blue curve) thus proving correctness of the
point of the plant frequency response G0(ωn) to a point L0 on identification. Radius of the dispersion circle MG drawn from
the negative real half-axis of the complex plane. Thus, the the nominal point G30(jωn) is RG=0.164.
magnitude |L0(jωn)|=|G0(jωn)||GR(jωn)|=1/GM yielding the
ratio |GR(jωn)|=1/[GM|G0(jωn)|] between the radii RG 4. As GM=18dB and the r.h.s. of (27) G0_RS=3.52 dB, the
and RL=|GR|RG of the circles MG and ML, respectively. The robust stability condition (26) GM>G0_RS is satisfied. The
radius RL of the dispersion circle ML is calculated as designed robust PID controller moves the nominal point
G30(jωn) on the negative half-axis into L30(jωn)=
1
RL = RG . (31) =G30(jωn)GR_rob(jωn)=0.12e-j180°, through which passes the
G M G0 ( j ω n ) Nyquist plot of the nominal open-loop L30(jωn) (Fig. 10 in
Substituting (30b) and (31) into the general robust stability green), where the gain margin GM=18 dB is guaranteed. The
condition (29) and considering the safety factor χL, the nominal closed-loop step response (Fig. 11a, green curve)
following inequality holds proves achieving the required nominal performance
ηmax0_obtained=4.55%, τs0_obtained=ωcts0_obtained=0,0488.243=11,86.
GM − 1 χ L RG
> , (32)
GM G M G0 ( j ω n ) 5. The dispersion circle ML (in green) radius RL=0.0573
encompasses all points L3i(jωn)=G3i(jωn)GR_rob(jωn) for
which after some manipulations is identical to the proven i=1…8. The PID controller has moved the worst point
condition (28). Let χL=1.2. According to the robust stability G3N(jωn) of the plant (blue symbol „+“ in Fig. 8) into
condition the chosen value GM is substituted into (26a) and
L3N(jωn)=0.16e-j197°, according to it the estimated worst gain
afterwards the robust PID controller parameters are obtained
margin is GMN=14.9 dB.
from (26) and (27). A setup of the proposed method is
extensively illustrated on the following example.
6. The smallest gain margin with the worst point G3N(jωn) of
the plant (blue symbol „+“ in Fig. 8) is specified by the
6. VERIFICATION OF THE PROPOSED ROBUST PID
intersection of the red Nyquist plot with the negative real
CONTOLLER DESIGN METHOD
axis, where the open-loop gain margin is G+MN=13.1 dB; here
Consider the following uncertain plant G3(s) with an unstable ηmaxN=25% and the relative settling time τsN=16 are expected
zero (according to „pink“ curves in Fig. 6 at ωn=0.65ωc). Achieved
performance ηmaxN_obtained=13.5%, tsN_obtained=301 s (red step
K 3 ( −α 3 s + 1 )
G3 ( s ) = , (33) response in Fig. 9b) prove this fact.
( T3 s + 1 )3
Nyquist Diagram
0.1
K 30 ( −α 30 s + 1 ) 0,8( −7.5s + 1 )
G30 ( s ) = = (34)
( T30 s + 1 )3 ( 27.5s + 1 )3 ω n)
L3N(jω
0
with parameters K3, T3 and α3 varying within ±15% around
the nominal values; G30(s) is the nominal model. For the
above plant, a robust PID controller is to be designed to -0.1 M L RL
guarantee a maximum overshoot ηmax0=5% and a maximum
relative settling time τs0=12 for the nominal model (33), and ωn)
L30(jω
Imaginary Axis

18/20
stability of the family of plants G3(s) (32) (robust stability). -0.2
RG 1/10
1. The measured critical frequency of the nominal model
ω)
L3N(jω 1/1014.9/20
is ωc=0.0488 s-1. From requirements on the nominal
closed-loop performance results ts=τs0/ωc=12/0.0488=
-0.3
ω)
L30(jω
=245.9 s. ωn)
G30(jω MG
-0.4
2. To achieve the expected nominal performance (ηmax0,τs0)=
=(5%,12), the gain margin and excitation frequency are ω)
G3N(jω ω)
G30(jω
chosen (GM,ωn)=(18dB,0.65ωc) using the „pink“ B-parabolas -0.5
-0.5 -0.4 -0.3 -0.2 -0.1 0 0.1
in Fig. 6 as according to (34) α30/T30=7.5/27.5=0.27<0.3.
Real Axis
Uncertainties of the plant are included in three parameters:
K3, T3 and α3, the number of identification experiments is Fig. 8. Nyquist plots of G30(jω), G3N(jω), L30(jω), L3N(jω):
therefore N=23=8. zoomed, for required performance ηmax0=5% and τs0=12

4937
19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014

1.5
Closed-loop time responses for the nominal model G30(s) (nominal maximum overshoot and settling time) into
frequency-domain performance specification in terms of
Controlled variable y(t)

1 nominal gain margin and phase crossover frequency.

ηmax0_obt.=4.5%, ts0_obt.=243 s
0.5
ACKNOWLEDGMENT
0 GM=18 dB, ωn=0.65ω
ωc This research work has been supported by the Scientific
-0.5
Grant Agency of the Ministry of Education of the Slovak
0 200 400 600 800 1000 1200 Republic under grants 1/1241/12, 1/0937/14 and the Slovak
Time (s) Research and Development Agency under grant
Closed-loop time responses for the worst plant model G3N(s) APVV-0772-12.
1.5
Controlled variable y(t)

REFERENCES
1

ηmaxN_obt.=13.5%, tsN_obt.=301 s Åström, K.J. and Hägglund, T. (1995). PID Controllers:

0.5
Theory, Design and Tuning, 2nd Edition. Instrument
Soceity of America. ISBN: 1556175167.
0 G+MN=13.1 dB, ωn=0.65ω
ωc Åström, K.J. and Hägglund, T. (2000). Benchmark Systems
for PID Control. IFAC PID’00. Terrasa, pp. 181-182.
-0.5
0 200 400 600 800 1000 1200 Blickley, G.J. (1990). Modern control started with Ziegler-
Time (s) Nichols tuning, Control Engineering, October, pp. 11-17.
Bucz, Š. and Kozáková, A. (2012). PID Controller Design
Fig. 9. Closed-loop step responses with the uncertain plant for Specified Performance. Introduction to PID
G3(s) and required values ηmax0=5% and τs0=12 Controllers: Theory, Tuning and application to frontier
areas. Department of Chemical Engineering, CLRI,
Modified version of sine-wave method Adyar, India, ISBN 978-953-307-927-1.
Grabbe, E.M., Ramo, S. and Wooldrige, D.E. (1959-61).
To avoid using the sine-wave generator, identification of the Handbook of Automation Computation and Control,
plant frequency transfer function point can be carried out as Vol.1,2,3, New York.
follows. Include some filter with the transfer function F(s) in Reinisch, K. (1974). Kybernetische Grundlegen und
closed-loop which does not violate the closed-loop. Using the Beschreibung Kontinuierlicher Systems. VEB Verlag
classical Ziegler-Nichols experiment (Ziegler and Nichols, Technik, Berlin.
1942) we obtain both the controller ultimate proportional gain Karaboga, D. and Kalinli, A. (1996). Tuning PID Controller
Kc and ultimate frequency ωc. With ultimate parameters the Parameters Using Tabu Search Algorithm, Proceedings
following equation holds of the IEEE International Conference on Systems, Man
and Cybernetics, pp. 134-136.
K c F ( jω c )G ( jω c ) = −1 + j 0 (35) Kristiansson, B. and Lennartson, B. (2002). Convenient
Almost Optimal and Robust Tuning of PI and PID
or, the complex-plane coordinates of G(jωc) are Controllers. IFAC 15th World Congress, Barcelona.
Morilla, F. and Dormido, S. (2000). Methodologies for the
1 Tuning of PID Controllers in the frequency domain.
G( jωc ) = − = m + jn . (36)
K c F ( jωc ) IFAC PID’00, Terasa, Spain, pp. 155-160.
O’Dwyer, A. (2000). A Summary of PI and PID Controller
Considering the crossover frequency to be set as ωc=ωp, a Tuning rules for Processes with time Delay. In IFAC
PID controller guaranteeing the prescribed gain margin using PID’00, Part 1, pp. 175-180, Part 2, pp. 242-247.
(15) and (16) can be designed. Tinham, B. (1989). Tuning PID controllers, Control and
Instrumentation, September, pp. 79-83.
If the plant is unstable, the additional transfer function F(s)
Veselý, V. (2003): Easy tuning of PID Controller. Journal of
with P controller can stabilize the closed-loop system; hence
El. Engineering, V 54, N5-6, 136-139
the proposed modification of the sine-wave method or the
Visioli, A. (2006). Practical PID Control. Advances in
modified Ziegler-Nichols method can be used for PID
Industrial Control. Springer London Limited, ISBN-10:
controller tuning even for unstable systems.
1846285852.
Vítečková, M., Víteček, A. and Smutný, L. (2000). Controller
7. CONCLUSIONS Tuning for Controlled Plants with Time Delay, Preprints
of Proceedings of PID'00: IFAC Workshop on Digital
The proposed robust PID controller design method is Control, Terrassa, Spain, pp. 283-288.
applicable for closed-loop output variable response shaping, Yu, Ch.-Ch. (2006). Autotuning of PID Controllers. A Relay
using various combinations of excitation signal values ωn and Feedback Approach, 2nd Edition, Springer-Verlag
required gain margins GM. Important contribution of the London Limited, ISBN 1-84628-036-2.
paper is construction of empirical plots converting time- Ziegler, J.G., Nichols, N.B. (1942): Optimum settings for
domain requirements specified by a process technologist Automatic Controllers. Trans. on ASME 64, pp. 759-768.

4938