You are on page 1of 8

This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.


IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY

Tuning of Proportional Retarded Controllers: Theory and Experiments


Raul Villafuerte, Sabine Mondi, and Ruben Garrido

Abstract This brief provides simple tuning rules for the


proportional retarded (PR) control of second order systems
requiring strong closed-loop damping. A frequency domain
analysis allows determining the -stabilizability regions of the
controller. The analysis provides explicit formulae for tuning the
three parameters of the PR controller, namely, the proportional
gain, the retarded gain, and the delay. The performance of the
PR closed-loop control is experimentally compared with that
of a proportional derivative (PD) controller. The experiments
show that the PR controller outperforms the PD controller
fed using velocity estimates obtained from a high-pass filter in
terms of noise amplification, control effort, and position error,
and has a similar performance compared with a PD controller
supplied with velocity estimates produced by an observer. Numerical implementation of the PR controller is computationally
less demanding than the corresponding implementation of PD
algorithms using velocity estimation based on filters or observers,
since it does not need solving ordinary differential equations
and only requires performing two products and a few memory
registers for implementing the time delay.
Index Terms D-partition method, exponential decay, proportional retarded control, second order system, time delay systems.

I. I NTRODUCTION
HE PROPORTIONAL DERIVATIVE (PD) controller is a
key component in many control laws applied to mechanical systems, and together with an integral action yields the
proportional integral derivative controller, which is the most
used algorithm in motion control of industrial electrical drives
[1][6]. Moreover, a PD controller plus gravity compensation
globally stabilizes a robot manipulator [7], [8]; the proportional action shapes the closed-loop potential energy and the
derivative action injects damping [9]. The PD controller is also
crucial in many neural network-based controllers [10]. In the
case of vibration attenuation problems, the PD controller is
able to mitigate the vibratory behavior in structures [11].
A practical problem with the PD controller when applied
to servo drive control is the fact that in many situations it
is not possible to measure the angular velocity. Including a
tachogenerator for velocity measurement is not convenient
because it adds bulk and cost to a control system; moreover,
its measurements have a high level of noise thus precluding
the use of high values of the derivative gains. A simple way of

Manuscript received April 14, 2011; revised March 8, 2012; accepted March
27, 2012. Manuscript received in final form April 13, 2012. This work was
supported in part by CONACyT, Mexico, Postdoctoral Grant 217000 and
Project Grant 61076. Recommended by Associate Editor F. A. Cuzzola.
R. Villafuerte is with the CITIS-ICBI Department, UAEH, Carretera
Pachuca-Tulancingo, Pachuca 42084, Mexico (e-mail: kual-es@hotmail.com).
S. Mondi and R. Garrido are with the Department of Automatic Control CINVESTAV-IPN, Mexico 14-740, Mexico (e-mail:
smondie@ctrl.cinvestav.mx; garrido@ctrl.cinvestav.mx).
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TCST.2012.2195664

overcoming velocity measurements is to use a high-pass filter


(HPF) as described in [12] for robot control. An advantage
of this approach is that its design does not need explicit
knowledge on the mechanical systems under control; however,
since in many practical cases an optical encoder supplies the
servo drive angular position, the signals produced by these
sensors processed through a HPF may produce biased noisy
estimates. Finite differences algorithms such that the Euler
method for obtaining velocity estimates also suffer from the
same problems as shown in [13]. An alternative to these
approaches is the use of state observers for obtaining velocity estimates [14][16]. Building an observer would require
explicit knowledge on the system parameters; for instance,
the gain and viscous friction parameters of a DC motor, and
requires solving a set of ordinary differential equations; despite
being more complex than a HPF, an observer may produce
velocity estimates with less noise. In the particular case of
high gain observers, a single parameter sets the observer
bandwidth [17].
Consider now the class of second order systems described
by the following models:
(t) + 2 (t) + 2 (t) = bu(t)

(1)

where > 0 is the non damped frequency, > 0 is the


damping factor, and b > 0 is the input gain. System (1),
although simple, is the first choice model for a wide range
of physical processes, such as the DC servomechanism experimental platform used in this research work. An alternative to
the standard PD paradigm
u(t) = k1 (t) k2 (t)

(2)

applied to these systems is the proportional retarded (PR)


controller
u(t) = k p (t) + kr (t h)
(3)
where k p is the proportional gain, kr and h are, respectively,
the retarded gain and the delay. The PR controller has been a
research subject in [18] and [19]; more recently, it has been
used in vibration mitigation control [11] or combined with an
integral action [20]. Compared with the PD controller, the PR
algorithm does not seek to estimate the time-derivative ; this
last feature avoids most of the drawbacks associated with the
use of filters and observers; moreover, its numerical implementation does not involve solving differential equations and only
requires a few memory registers to approximate the time delay.
A key aspect of any controller is its tuning; in the particular
case of the PR controller, it has been performed using numerical methods for minimizing the integral of time absolute
error performance index. In this regard and to the best of the
Authors knowledge, there is no previous results concerning

10636536/$31.00 2012 IEEE

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
2

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY

explicit analytical rules for tuning the three parameters of a PR


controller. Obtaining such rules is not an easy task if one takes
into account that introducing a time delay in the controller
produces an infinite number of closed-loop poles. In order to
appreciate this point, substituting the PR control law (3) into
(1) gives the closed-loop characteristic quasipolynomial
p(s, k p , kr , h) = s 2 + 2s + 2 + bk p bkr ehs .

(4)

As in the case of delay-free systems, when the system


has no zeros, the shaping of the closed-loop delay system
response depends on the location of the dominant roots of
the characteristic quasipolynomial, and on their nature, real or
complex conjugates. This response shaping problem is indeed
closely related to the stabilization with prescribed exponential
decay, named -stabilizability.
Since the PR control law deliberately introduces a delay,
it is convenient to review work related to this issue. In the
control literature, it is well known that the presence of delays
may induce instability or bad performance. At the same time,
there exist simple dynamical systems, such as second-order
oscillators and more general classes of oscillatory systems for
which a delay in the output feedback may have a stabilizing
effect [21], [22]. The basic ideas for studying the stability of
time delay systems in a parametric plane, originate from the
work of the Russian scientist Neimark [23]. On the other hand,
the root-locus technique has been employed for determining
the critical open-loop gain of a closed-loop system for a
fixed time delay [21], [24]. The main limitations of frequency
domain techniques are the case by case analysis they require
and the restriction to fairly simple linear systems. Their
advantage is that necessary and sufficient stability conditions
can be obtained, and that precise information on the roots
is available. Substantial advances, enlarging the classes of
systems that can be analyzed, were recently presented for
general single input single output systems with delayed control
[25] and for two delay systems [26]. The results presented in
this brief follow from a detailed frequency domain analysis of
the -stabilizability of the closed-loop quasipolynomial (4).
The contribution of this brief is twofold. On the one hand,
it presents an analytical tuning technique for the PR control
of second order systems where the introduction of strong
damping is important. On the other hand, it shows experiments
in a laboratory prototype for evaluating the performance of the
tuning technique. It is worth noting that most of the previous
works on PR controller tuning rely on numerical methods and
do not give explicit formulae for setting the proportional gain,
the retarded gain, and the delay. An exception is [11]; there,
the authors give analytic tuning rules for a retarded controller,
i.e., a controller without the proportional gain, for setting
up the retarded gain, and the delay. Moreover, most of the
published tuning methods for PR controllers are tested using
only numerical simulations; therefore, issues like measurement
noise, unmodeled dynamics, controller robustness with respect
to system parameters, and numerical implementation are not
taken into account.
This brief has the following structure. The main definitions
and tools are introduced in Section II. In Section II-A, the
-stabilizability boundaries and regions are determined.

The proposed three real roots assignment strategy is


characterized in Section II-C, leading to a low gain/oscillation
reduction tuning rule for the PR -stabilizabiling controller
presented in Section III. The last part of this brief addresses
the practical implementation of the PR control law on
a DC-servomotor prototype discussed in Section IV-A.
In Section IV-B, PD control schemes, using an observer and a
HPF for estimating the angular velocity of a servo drive, are
studied. A thorough evaluation of the regulation and tracking
performances, noise attenuation, and design complexity of
these schemes is performed in Section IV-C. The contribution
ends with some concluding remarks.
II. P RELIMINARY A NALYSIS
As proved in [27], the -stability of linear delay systems can
be characterized in the frequency domain: all the roots of the
characteristic equation must have real parts smaller than .
Moreover, it is well known that the change of variable
s (s ) in the frequency domain reduces the analysis
of the -stability of (4) to the stability of the transformed
quasipolynomial
p (s, k p , kr , h) = s 2 2( )s + ( )2

+ 2 (1 2 ) + bk p bk r eh ehs .

(5)

Remark 1: The decay of the autonomous system (1)


(u 0) is . The analysis presented in this brief is restricted
to the case of closed-loop exponential decay > , which
corresponds to an improved exponential decay when the gains
k p and kr are positive.
In the following, the -stabilizability is simply characterized
using the D-partition method [23]: the candidate boundaries
are determined by finding the crossings of the imaginary axis
of (5).
Stability charts for second order systems with time lag, and
quasipolynomials of the form q(s) + p(s)esh where q(s) and
p(s) are polynomials such that deg( p(s)) < deg(q(s)), have
been a long lasting topic of study in the delay community.
For formal analysis based on different techniques, the reader
is referred to [21], [24], and [28][32], among others.
A. -Stabilizability Boundaries
As the quasipolynomial (5) roots behavior in the complex
plane is continuous with respect to continuous changes of the
coefficients and time delay [23], loss of -stabilizability occurs
when the quasipolynomial (5) has roots either at s = 0 or
a pair of pure imaginary roots at s = j . Next, the root
crossings of the imaginary axis, which are indeed candidate
stability/instability boundaries are characterized.
Proposition 1: Root loci of the quasipolynomial (5) at
s = 0 satisfy
( )2 + 2 (1 2 ) + bk p
.
(6)
beh
Proof: The result follows straightforwardly by substituting
s = 0 into (5).
kr =

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
VILLAFUERTE et al.: TUNING OF PR CONTROLLERS

Proposition 2: Pure imaginary


 root loci of the
quasipolynomial (5) occur at j 1,2 for real and positive
1,2 of the form


1,2 = 2 1 2 ( )2 + bk p



(bkr eh )2 4 ( )2 2 (1 2 ) + bk p . (7)
Proof: Setting
p ( j , k p , kr , h) = 2 2 j ( )+( )2


+ 2 1 2 +bk p bkr eh e j h
=0

and taking modulus yield


 

2
4 22 + bk p 2 ( )2 + + bk p

where = 2 2 + 2 . Introducing the change of variable


= 2 leads to the quadratic polynomial in
 

2
2 2 +bk p 2 ( )2 + +bk p b2 kr2 e2h = 0

whose roots are given by (7) and the result follows.


Proposition 3: The parametric equations for the time delay
h and the retarded gain kr of the quasipolynomial (5) corresponding to root crossings of the imaginary axis at purely
imaginary pairs are
 2

+ ( )2 + 2 (1 2 ) + bk p
1
h() = cot1

2( )

+ n , n = 0, 1 . . . , = 0
(8)

2( )
.
(9)
kr (, h) = h
be sin(h)
Proof: Substituting e j h = cos(h) j sin(h) into (5)
2

-stable region of (1) and (3).

kr in (6) is positive. Straightforward substitution shows that


the hypersurface (9) intersects (6) at frequencies

b 2 kr2 e2h = 0

Fig. 1.

Re{ p ( j , k p , kr , h)} = +( ) + (1 )+bk p


bkr eh cos(h)
=0

Im{ p ( j , k p , kr , h)} = bkr eh sin(h) 2( ) = 0


and the result follows by simple algebraic manipulations.
These parametric equations are sketched on Fig. 1 in
the bi-dimensional space (kr , h) for different values of .
The parameter numerical values b = 31, = 17.6, =
0.0128, and the fixed gain k p = 22.57, correspond to the
application considered in this contribution. Fig. 1 suggests that
the -stabilizability regions with larger delays have poorer
performance. As a consequence, our analysis is focused on
the region corresponding to shorter delays, which is depicted
on Fig. 2.
B. Regions of -Stabilizability
Notice that for the two hypersurfaces described by the
analytic expressions (6) and (8)(9) to be well defined, the
positivity under the square root in (7) must be insured, which
is true if k p > 2 (1 2 )/b > 0. This in turn implies that

a = 0



c = 2 2 (1 2 ) ( )2 + bk p

(10)
(11)

corresponding to locus (a) and (c) depicted on Fig. 2.


Moreover, it appears that at the intermediate frequency

c
(12)
b = 2 (1 2 ) ( )2 + bk p =
2
corresponding to locus (b), the argument of the square root in
(7) is null hence


2 ( ) 2 (1 2 ) + bk p
kr =
.
beh
Subtraction of this expression from (6), for same and h,
followed by squares completion yields



 2
kr kr = ( ) + 2 (1 2 ) + bk p
/beh > 0.
Hence, we conclude that the region described below is well
defined.
Proposition 4: Given > 0, b > 0 and > 0, then for
> and k p > 2 (1 2 )/b > 0, the first stabilizability
region in the parameter space (kr , h) is described as follows.
Upper Boundary: For the selected k p and , sketch in the
(kr , h) plane
kr =

( )2 + 2 (1 2 ) + bk p
beh

(13)

where h 2( )/( )2 + 2 (1 2 ) + bk p , h(c ) .
Here, c is defined in (11) and

h() =


 2
2
2
2
1 cot1 +( ) + (1 )+bk p , (0, e )

2( )

 2
2
2
2
1 cot1 +( ) + (1 )+bk p + , (e , c )

2( )


with e = min{c , ( )2 + 2 (1 2 ) + bk p }.
Lower Boundary: For the selected k p and , sketch in the
(kr , h) plane
2( )
(14)
kr () = h
be sin(h())
for h() defined as for the upper boundary.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
4

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY

Proof: When there is a triple root at  , the conditions


p (0, k p , kr , h) = 0, /s p (s, k p , kr , h)s=0 = 0 and
2 / 2 s p (s, k p , kr , h)s=0 = 0 hold, namely
( )2 + 2 (1 2 ) + bk p = bkr eh
hbkr eh 2( ) = 0
h 2 bkr eh = 2.

Fig. 2.

Fig. 3.

It follows from (18) and (19) that


2( )
h=
( )2 + 2 (1 2 ) + bk p
and (18) and (20) imply that
2
.
h2 =
( )2 + 2 (1 2 ) + bk p

Main -stable region of (1) and (3).

-stable regions of (1) and (3) for = 32 and k p [22.57, 110].

Clearly, Proposition 4 allows drawing the stability charts


for any system parameters , b, , and given decay and k p
satisfying the conditions of the proposition.
Finally, the fact that the -stabilizability regions in the space
(kr , h) grow as the gain k p does is depicted in Fig. 3.
Clearly, for a given -stabilizability specification, the same
exponential decay is achieved at all the points of the region
boundaries. The upper boundary corresponds to loci of
p(s, k p , kr , h) with at least one dominant root at , while
the lower boundary corresponds to pairs of complex conjugate
roots with real part . Fig. 2 suggests that the largest
achievable exponential decay, named , occurs when these
two boundaries collapse into a point, which is characterized
by a rightmost root with multiplicity three.
C. Triple Dominant Real Roots Assignment
The analysis of the previous section motivates the following
design assigning a triple root at when k p is fixed.
The corresponding retarded gain kr and delay h are also
determined.
Lemma 1: Let the proportional gain of the controller k p >
2 (1 2 )/b be given. Then, a triple rightmost root of the
closed-loop system (1) and (3) at is achieved for

(15)
= + 2 (1 2 ) + bk p .

Moreover, the values of delayed gain kr and delay h that


-stabilize (1) and (3) with the exponential decay are
1
(16)
h = 
2 (1 2 ) + bk p
kr =

2( )
.
bh e h

(17)

(18)

(19)
(20)

(21)

(22)

Substituting (21) into (22) yields ( )2 = 2 (1 2 )+bk p ,


and (15) follows. Then (16) follows from substituting (15) into
(21), and (17) follows from (19).
We now prove that the locus (kr , h ) is -stable. Substitution of (20)(22) implies that
1
2
2

p (s, k p , kr , h ) = s 2 2 s + 2 2 eh s .
h
h
h
As h > 0, this is equivalent to verify that the quasipolynomial
p(s)

= s 2 2s + 2 2es has no roots with strictly positive


real part. Graphical methods based on the argument principle
show indeed that p(s)

has no such roots.


Remark 2: The above strategy insures robust stability of the
closed-loop system. In view of the continuity of the location
of the roots with respect to parameters [23], the obtained
stability margin allows parameter variations, including
delay uncertainties due to sampling, before the closed-loop
becomes unstable.
It is worth mentioning that the resulting controller is not the
less fragile (the one which admits the larger control parameter
perturbations without reaching instability, see [33], [34], and
the references therein).
III. T UNING OF THE PR C ONTROLLER
The following paragraphs describe a tuning strategy
obtained from the three repeated real dominant roots assignment of the previous section, which insures a non oscillatory
closed-loop system response. This feature is indeed useful in
applications where the introduction of damping avoids oscillatory closed-loop behavior, for instance, in robot manipulator
control [8] or in ship autopilot control [35]. Notice that Fig. 3
suggests that the three dominant real roots assignment at
corresponds to the minimum proportional gain k p required for
achieving this exponential decay.
Lemma 2: Let a specified exponential decay > be
given according to the designer specifications. Then, (3) that
-stabilizes (1) with triple dominant real root at is

determined by the parameters (k p , kr , h)


2
2
( ) (1 2 )
(23)
k p =
b
h = 1/ [ ]
(24)


2( )2
.
(25)
kr =
be h

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
VILLAFUERTE et al.: TUNING OF PR CONTROLLERS

Proof: The result follows from straightforward algebraic


manipulations of (15)(17).
Remark 3: The above strategy is not suitable for all
situations. On the one end, it may not be necessary in
plants tolerating underdamped response, such as gas turbine
temperature control. On the other hand, forcing multiple real
roots in poorly damped systems, such as flexible structures
may result in closed-loop characterized by high control effort
and poor robustness. Finally, multiple roots assignments are
highly sensitive, as reported in the literature (see [32] for
the case of delay systems). This suggests, for given k p and
< , a tuning strategy of two complex conjugate roots
with real part based on the observation of Fig. 2, and
on (7) describing the imaginary axis crossing frequencies that
range from 0 at locus (a) to c defined in (11) at locus (c).
For example, by choosing the locus where 1 = 2 in (7), the
assigned frequency is b defined in (12).
Notice that this locus
always exists because 0 b = c / 2. The corresponding
tuning is given by


1
2b ( )
1
h(b ) =
cot
.
,
kr (b ) = h( )
b sin(h )
b
b
be
b
Remark 4: Additional tuning rules assigning two or three
dominant roots are given in [36]. It is worth mentioning that
the stability charts of Figs. 2 and 3 allow visualizing the effect
of parameter variations on the root dominance. For example,
Fig. 2 shows that one can reduce expectations regarding the
decay rate, and increase kr to reach a level curve < with
a single dominant root at , or reduce kr to reach a level
curve with complex conjugate roots with real part . Similar
observations apply to the choice of the time delay h and of
the proportional gain k p .
IV. E VALUATION OF THE PR C ONTROL S TRATEGY
Experiments are conducted on the PR control of a
DC-servomotor for the evaluation of the three dominant real
roots assignment tuning strategy of Lemma 2. Experiments
for other root assignment are available in [36]. A comparative
analysis with different popular implementations of PD control
laws avoiding the measurement of the angular velocity is
presented.
A. Experimental DC-Servomotor Setup
The servomechanism employed for the experiments consists
of a DC brushed motor controlled through a Copley controls
power amplifier, model 413, configured in current mode.
A BEI optical encoder directly coupled to the motor shaft
gives angular position measurements. The resolution of the
optical encoder is 2500 pulses per revolution. A Servotogo
Card endowed with inputs for optical encoders performs data
acquisition. The electronics associated to these inputs multiply
by four the encoder resolution. In this way, one motor turn
corresponds to 10 000-encoder pulses. A factor of 10 000
scales down the angular position measurements. The card also
has 12 bits digital-to-analog converters with an output voltage
range of 10 V. The Matworks M ATLAB/Simulink graphical programming together with Quanser Wincon real-time

Fig. 4.

Servomechanism.

environment allow implementing all the controller studied in


the next sections. The sampling period is 1 ms that corresponds
to 1000 Hz. The RungeKutta method is used for implementing the controllers. The servomechanism is shown in Fig. 4.
We consider the following second order model for the DC
servomechanism:
J q(t)
+ f q(t)
= (t) = ku(t)

(26)

where q is the angular position, (t) the input torque, u(t)


the control input voltage, J the motor and load inertia, f
the viscous friction, and k the amplifier gain. A brushed
servomotor, a power amplifier, and a position sensor compose
the servomechanism. The power amplifier is set to current
mode; therefore, the electromagnetic torque is proportional to
the input voltage applied to the amplifier. This approach also
works for both DC and AC brushless servomotors.
Observe that (26) can be written as
q(t)
= a q(t)
+ bu(t)

(27)

where a = f /J , b = k/J are positive parameters.


The estimated parameter values for this platform, obtained via
the identification algorithm proposed in [37], are a = 0.45 and
b = 31.
B. Controllers Design
The triple real roots assigning PR design is now evaluated.
The PR controller is compared to two well known control laws
that also avoid measuring the angular position time derivative.
The first one is the PD control plus a HPF. The second one is
the PD controller, where an estimate of the angular velocity is
obtained via a Luenberger state observer. In both cases, the use
of the so called tachometric feedback is considered. For a fair
evaluation, the same rightmost closed-loop roots are assigned
in all the schemes.
PR Control: Notice that (27) is not in the general form (1).
The auxiliary proportional control law u(t) = kpre q(t) +
(t), where kpre is a preliminary proportional control gain, is
applied to (27) and (t) is a control signal of (3), leads to a
system of (1) with

a
= bkpre , = 
.
2 bkpre

Notice that the above indications concern position regulation.


For position tracking, the variable q must be replaced by the

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
6

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY

TABLE I
M EAN S QUARE P OSITION E RROR

0.3
Ref
PR
PD+Obs
PD+HPF

0.2

q(t)

0.1

Controller
mse
Controller
mse

0
0.1

PR
0.2279
PD+Obs+Tac
0.3366

PD+HPF
0.2384
PD+HPF+Tac
0.3387

PD+Obs
0.2436

0.2
0.3

0.3

time

Ref
PR
PD+Obs+Tac
PD+HPF+Tac

0.2

q(t)

0.1

Fig. 5. Output variable q(t) of (26) with feedback controllers PR, PD+Obs,
and PD+HPF.

0
0.1

0.01

PR

0.2

0
0.3

e(t)

0.01
0
0.01

time

0
PD+Obs
0.01
0
0.01

Fig. 7.
Output variable q(t) of the closed-loop system (26) with the
tachometric feedback controllers PR, PD+Obs, and PD+HPF.

0
PD+HPF
0.01
0

Fig. 6.

time

Position error e(t) for the controllers PR, PD+Obs, and PD+HPF.

is designed





q(t)

d q(t)
.
.
=A
+ Bu(t) + K 0 (q(t) q(t))

(29)
dt q(t)

q(t)

where

negated tracking error. For kpre = 10, one gets = 17.6


and = 0.0128. Then, the PR control law is designed
according to the strategy presented in Section III. For = 32,
the proportional gain k p = 22.57, the retarded gain kr =
23.7941, and delay h = 0.03147 are readily computed from
(23)(25), respectively. Clearly, the proportional gain that is
actually applied to the servomotor is kpre + k p = 32.57.
If the above strategy is not completely satisfactory for the
user, the fine tuning of the control can be done with the
help of Figs. 2, and 3 with the full characterization of the
dominant roots presented in [36]. Note also that the closedloop undamped frequency,
including the preliminary propor
tional gain is = b(kpre + k p ). Its corresponding numerical
value is 31.75 rad/s, which roughly corresponds to 5 Hz. Thus,
the sampling frequency used in the experiments is well above
the closed-loop undamped frequency. On the other hand, the
time delay h = 0.03147 s is implemented using 31 sampling
periods, i.e., the implemented delay has a value of 0.031 s.
PD Control With a HPF (PD+HPF): The PD controller
(2) is designed by setting the closed-loop polynomial to
(s + )2 for = 32. This is achieved with the proportional
gain k1 = 33 and the derivative gain k2 = 2.
The servomotor angular velocity of the state is obtained
through the use of the HPF
300s
(28)
300 + s
applied to the position measurements.
PD Control With an Observer (PD+Obs): The PD controller is the same as the one designed for the PD+HPF
scheme (k1 = 33, k2 = 2). In view of the separation principle,
a Luenberger observer with measurement of the position q(t)
G(s) =

A=

0 1
0 a

B=


0
b

K0 =

k01
k02

with a = 0.45, b = 31. The choice k01 = 319.55, k02 =


25456.20 for the observer gain gives an observation error
dynamic five times faster than the dynamics assigned by the
control law. In the frequency domain, the transfer of the
estimated variables with respect to the position is


q(s)

(30)
= (s I A + K 0 C + B K )1 K 0 q(s)
s q(s)





where C = 1 0 and K T = k1 k2 .

C. Performance Evaluation

Next, the response of the system in closed-loop with the


above control schemes is discussed in the light of tracking
and regulation, noise attenuation, and design/computational
complexity.
1) Tracking and Regulation: The three controllers, PD,
PD+Obs, and PD+HPF, are tested with the tracking of a
signal comprised of a sinusoid followed by a step. The position
and the reference are shown in Fig. 5, and the position error
is displayed in Fig. 6. It is possible to conclude that the
performance of the PR, PD+Obs, and PD+HPF controllers
is comparable,
 T with a slightly smaller mean square error
mse = 1/T 0 |e(t)|dt for the PR, as shown in Table I.
The same experiments are also conducted by using tachometric feedback, denoted by +Tac, that consists of feeding
back the derivative of the state variable q(t) instead of the
derivative error signal e(t). Figs. 7, 8, and Table I show that
the observer-based and the HPF-based strategies introduce a
large error when tracking the sinusoid.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
VILLAFUERTE et al.: TUNING OF PR CONTROLLERS

90

0.05
PR

e(t)

0.05
0
0.05

0
PD+Obs+Tac

0.05
0
0.05

Magnitude (dB)

80

PD (theoretical)

70
PD+HPF

60

PD+Obs

50
40
30
20

0
PD+HPF+Tac

0.05
0

time

Fig. 8. Position error e(t) for the controllers PR, PD+Obs, and PD+HPF
with tachometric feedback.

PR

10 1
10

Fig. 9.

10

10

10

10

Frequency (rad/sec)

10

Bode gain diagram.

0.2
PR
0
0.2
0
0.2

u(t)

2) Control Signal Frequency Characteristics: The transfer functions of the controllers under consideration are the
following.
1) PD (Theoretical): The position time derivative is
assumed to be available. Expression (2) implies that

0.2
1

4
PD+HPF

0
0.2
0

Fig. 10.

time

Control signal u(t) of the schemes PR, PD+Obs, and PD+HPF.

0.2
PD+Obs+Tac
0

0.2
0

u(t)

3) PD+Obs: Substituting the estimates (30) given by the


observer (29) into (2) leads to

PD+Obs

0
0.2

u(s)
= k1 + k2 s.
q(s)

2) PD+HPF: Using the HPF (28) output instead of the


angular velocity variable in (2) yields


u(s)
300s
(k1 + 300k2 )s + 300k1
= k1 +k2
.
=
q(s)
300 + s
s + 300

0.2

u(s)
q(s)

(k01 k1 + k2 k02 )s + k01 k1 a + k02 k1


= 2
.
s + (bk2 + k01 + a)s + k01 a + k01 bk02 + k02 + bk01

4) PR: Equation (3) gives the transfer function


u(s)
= k p kr ehs .
q(s)

The Bode gain diagrams of these transfer functions are


sketched in Fig. 9. They show that the PR and the PD+Obs
control laws have a lower gain at high frequencies, hence
they attenuate high frequency measurement noise. Indeed,
one can see in Fig. 10 that the control signals for the PR
control and the PD+Obs are significantly smoother than the
HPF control. Moreover, the magnitude of the PR control
is slightly smaller than the PD+Obs. This is due to the
fact that, as one can see in Fig. 9, the proportional gain
of the PR controller is smaller than that of the PD+Obs,
while they both achieve the same response exponential decay.
Clearly, one could use a PR controller with a larger
and, according to Lemma 2, larger k p while staying into a
region where actuators do not saturate. Fig. 11 shows that
the noise in the control signals of controllers PD+HPF and
PD+Obs is significantly amplified when using tachometric
feedback.
The control signals depicted in Figs. 10 and 11 show clearly
that the use of the HPFs results in control law with large
amplitude peak, significantly greater magnitudes, and great
sensitivity to noise. It should be mentioned that the sound

0
PD+HPF+Tac
0.2
0

time

Fig. 11. Control signal u(t) of the schemes PD+Obs and PD+HPF with
tachometric feedback.

produced by the servomotor during experiments reflects these


facts, i.e., the PD controller using these velocity approximations produces a lot of acoustic noise while the PR controller
works silently.
3) Computational and Implementation Issues: The design
of the PR control strategy reduces to substituting the parameter
of the servomotor and the proportional gain k p used in this
setup into the simple formulae (15)(17). Moreover, tuning
the PR controller requires the same prior knowledge about the
servomechanism parameters than an observer design. Notice
also the availability of sketches allowing the fine tuning of the
leading roots.
Regarding real-time implementation, it is worth noting that
an observer-based control law requires solving on-line a pair of
differential equations, whereas the PR controller only requires
a few kilobytes of memory allocation for implementing the
delay. These issues are paramount when these controllers
are implemented in low-cost microprocessors. Another issue
deserving comments corresponds to the approximation of the
time delay; the experiments indicate that the error introduced

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
8

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY

in approximating the delay has not apparent consequence on


closed-loop performance, as expected from the observations
made in Remark 2.
V. C ONCLUSION
This brief presented a PR controller tuning strategy for
second order system where a highly damped closed-loop
was needed. The controller parameters assigning a triple
dominant real root were readily computed through simple
formulae after selecting the desired exponential decay for
the response. The obtained controller is non fragile in the
sense that it admits controller parameter variations without
reaching instability. An alternative tuning strategy assigning
two complex conjugate roots was also outlined.
The experimental evaluation shows that, for the triple real
roots assignment, the PR controller outperforms a PD controller where the time derivative was produced by a HPF,
in terms of the position error as well as control effort.
The PR controller is able to give the same performance
that an observer-based control law. A comparative study of
the Bode magnitude diagrams for the controllers employed
in the experiments reveals that the PR controller together with
the observer-based control law, have the lowest gain at high
frequencies; however, the PR controller is less computationally
demanding. Finally, it should be mentioned that, unlike matrix
linear inequality based control design approaches, the tuning
of the PR controller in the frequency domain presented here
gives a useful grasp on the dominant root location, as well as
the possibility of fine-tuning for additional purposes.
R EFERENCES
[1] G. Ellis, Control System Design Guide: A Practical Guide, 3rd ed.
Amsterdam, The Netherlands: Elsevier, 2004.
[2] W. Leonhard, Control of Electrical Drives. New York: Springer-Verlag,
1996.
[3] R.-E. Precup and S. Preitl, PI and PID controllers tuning for integraltype servo systems to ensure robust stability and controller robustness,
Electr. Eng. (Archiv Elektrotech.), vol 88, no. 2, pp. 149156, 2006.
[4] R. Kelly and J. Moreno, Learning PID structures in an introductory
course of automatic control, IEEE Trans. Edu., vol. 44, no. 4, pp. 4373
376, Nov. 2001.
[5] K. J. Astrom and T. Hagglund, PID Controllers, 2nd ed.
Research Triangle Park, NC: Int. Soc. Meas. Control, 1995,
pp. 14.
[6] Q. G. Wang, Z. Zhang, K. J. Astrom, and L. S. Chek, Guaranteed
dominant pole placement with PID controllers, J. Process Control,
vol. 19, no. 2, pp. 349352, 2009.
[7] M. Takegaki and S. Arimoto, A new feedback method for dynamic
control of manipulators, J. Dyn. Syst., Meas. Control, vol. 103, no. 2,
pp. 119125, 1981.
[8] M. W. Spong, S. Hutchinson, and M. Vidyasagar, Robot Modeling and
Control. Hoboken, NJ: Wiley, 2006.
[9] R. Ortega, J. A. L. Perez, P. J. Nicklasson, H. J. Sira-Ramirez, and
H. Sira-Ramirez, Passivity-Based Control of Euler-Lagrange Systems:
Mechanical, Electrical, and Electromechanical Applications. New York:
Springer-Verlag, 1998.
[10] F. L. Lewis, S. Jagannathan, and A. Yesildirek, Neural Network Control
of Robot Manipulators and Nonlinear Systems, London, U.K.: Taylor &
Francis, 1999.
[11] H. Elmali, M. Renzulli, and N. Olgac, Experimental comparison of
delayed resonator and PD controlled vibration absorbers using electromagnetic actuators, J. Dyn. Syst., Meas., Control, vol. 122, no. 3, pp.
514520, 2000.

[12] H. Berghuis and H. Nijmeijer, Global regulation of robots using only


position measurements, Syst. Control Lett., vol. 21, no. 4, pp. 289293,
1993.
[13] R. C. Kavanagh, Performance analysis and compensation of M/T-type
digital tachometers, IEEE Trans. Instrum. Meas., vol. 50, no. 4, pp.
965970, Aug. 2001.
[14] Y. Sheng-Ming and K. Shuenn-Jenn, Performance evaluation of a
velocity observer for accurate velocity estimation of servo motor
drives, IEEE Trans. Ind. Appl., vol. 36, no. 1, pp. 98104, Jan.Feb.
2000.
[15] G. Ellis, Observers in Control Systems: A Practical Guide. San Diego,
CA: Academic, 2002.
[16] R. D. Lorenz and K. W. Van Patten, High-resolution velocity estimation
for all-digital, AC servo drives, in Proc. IEEE Ind. Appl. Soc. Annu.
Meeting, vol. 1. Oct. 1988, pp. 363368.
[17] K. W. Lee and H. K. Khalil, Adaptive output feedback control of robot
manipulators using high-gain observer, Int. J. Control, vol. 67, no. 6,
pp. 869886, 1997.
[18] H. Suh and Z. Bien, Use of time-delay actions in the controller design,
IEEE Trans. Autom. Control, vol. 25, no. 3, pp. 600603, Jun. 1980.
[19] G. M. Swisher and S. Tenqchen, Design of proportional-minus-delay
action feedback controllers for second-and third-order systems, in Proc.
Amer. Control Conf., Atlanta, GA, 1988, pp. 254260.
[20] Q. C. Zhong and H. X. Li, A delay-type PID controller, in Proc. 15th
Triennial World Congr. Int. Federat. Autom. Control, Barcelona, Spain,
2002, pp. 16.
[21] C. Abdallah, P. Dorato, J. Benitez-Read, and R. Byrne, Delayed positive
feedback can stabilize oscillatory system, in Proc. Amer. Control Conf.,
San Francisco, CA, 1993, pp. 31063107.
[22] V. L. Kharitonov, S. I. Niculescu, J. Moreno, and W. Michiels, Static
output feedback stabilization: Necessary conditions for multiple delay
controllers, IEEE Trans. Autom. Control, vol. 50, no. 1, pp. 8286, Jan.
2005.
[23] J. Neimark, D-subdivisions and spaces of quasi-polynomials, Prikl.
Mat. Meh., vol. 13, no. 4, pp. 349380, 1949.
[24] K. L. Cooke and P. Van Den Driessche, On zeroes of some transcendental equations, Funkcialaj Ekvacioj, vol. 29, no. 1, pp. 7790,
1986.
[25] C. I. Morarescu, S.-I. Niculescu, and K. Gu, Stability crossing curves of
SISO systems controlled by delayed output feedback, Dyn. Continuous,
Discrete Impuls. Syst., Ser. B, vol. 14, no. 5, pp. 659678, 2007.
[26] K. Gu, S.-I. Niculescu, and J. Chen, On stability crossing curves for
general systems with two delays, J. Math. Anal. Appl., vol. 311, no. 1,
pp. 231253, 2005.
[27] R. Bellman and K. Cooke, Differential-Difference Equations. New York:
Academic, 1963.
[28] C.-I. Morarescu and S.-I. Niculescu, Stability crossing curves of SISO
systems controlled by delayed output feedback, Dyn. Continuous,
Discrete Impuls. Syst., Ser. B, Appl. Algorithms, vol. 14, no. 5, pp. 659
678, 2007.
[29] K. L. Cooke and Z. Grossman, Discrete delay, dristributed delay and
stability switches, J. Math. Anal. Appl., vol 86, no. 2, pp. 592627,
1982.
[30] C. S. Hsu and S. J. Bhatt, Stability charts for second-order dynamical
systems with time lag, J. Appl. Mech., vol. 33, no. 1, pp. 119124,
1966.
[31] G. Stepan, Retarded Dynamical Systems. London, U.K.: Longman, 1989.
[32] W. Michiels and S. I. Niculescu, Stability and Stabilization of TimeDelay Systems: Advances in Design and Control 12. Philadelphia, PA:
SIAM, 2007.
[33] C.-F. Mndez, S.-I. Niculescu, I.-C. Morarescu, and K. Gu, On the
fragility of PI controllers for time-delay SISO systems, in Proc. 16th
Medit. Conf. Control Autom., Ajaccio, France, 2008, pp. 529534.
[34] D. Melchor and S.-I. Niculescu, Computing non-fragile PI controllers
for delay models of TCP/AQM networks, Int. J. Control, vol. 82, no.
12. pp. 22492259, 2009.
[35] M. A. Johnson and M. H. Moradi, PID Control: New Identification and
Design Methods. London, U.K.: Springer-Verlag, 2005.
[36] R. Villafuerte and S. Mondi, A strategy for the tuning of a second
order system in closed loop, in Proc. 9th IFAC Workshop Time-Delay
Syst., Prague, Czech Republic, 2010, pp. 16.
[37] R. Garrido and R. Miranda, DC servomechanism parameter identification: A closed loop input error approach, ISA Trans., vol. 51, no. 1, pp.
4249, 2012.

You might also like