BULLETIN OF THE POLISH ACADEMY OF SCIENCES CONTROL AND INFORMATICS

TECHNICAL SCIENCES, Vol. 70(1), 2022, Article number: e139957

DOI: 10.24425/bpasts.2021.139957

Tuning PID and PI-PI servo controllers

by multiple pole placement
Andrzej BOŻEK ∗ and Leszek TRYBUS

Department of Computer and Control Engineering, Rzeszów University of Technology, W. Pola 2, 35-959 Rzeszów, Poland

Abstract. Tuning rules for PID and PI-PI servo controllers are developed using a pole placement approach with a multiple pole, i.e. a triple
one in the case of PID and a quadruple for PI-PI. The controllers involve complex roots in the numerators of the transfer functions. This is not
possible in the classical P-PI structure which admits real roots only. The settling time of the servos determined by the multiple time constant
is the only design parameter. Nomograms to read out discrete controller settings in terms of the time constant and control cycle are given. As
compared to the classical structures, the upper limit on the control cycle is now twice longer in the case of PID, and four times in the case of
PI-PI. This implies that the settling times can be shortened by the same ratios. Responses of a PLC-controlled servo confirm the validity of the
design.

Key words: servomechanism; double integrator; pole placement; PID controller; discrete-time control.

1. INTRODUCTION Fast tool servo devices, particularly direct drives, may be

Servomechanisms, called servos for short, as motion control equipped with additional acceleration feedback loop involv-
systems are final execution elements of robots, machine tools, ing an acceleration sensor and acceleration controller providing
conveyors, etc. [1]. Extensive analyses, design solutions, and a set-point for the current controller. The idea was initially pro-
applications of the servos are described in [2, 3]. Robot servos posed in [9]. The acceleration signal represents instantaneous
are typically equipped with permanent magnet (PM) brushless net reaction of the moved mass to current-generated control
DC (BLDC) motors or AC synchronous motors (PMSM) driven force and disturbance forces due to varying load, vibrations, etc.
by suitable power devices. A classical control structure involves The additional loop, usually proportional, reduces the effect of
a torque controller (two-axis controllers for PMSM), a PI ve- disturbance forces thus enhancing the performance of the servo.
locity loop, and an outer P position loop. This P-PI setup is For P-PI or PID position controllers a servo-drive equipped with
equivalent to a single PID with real roots (zeros) in the transfer the acceleration loop still looks like a double integrator.
function numerator. By neglecting viscous friction, a servomo- Application of the root-locus method for the design of P-PI
tor equipped with a torque controller is typically described by servo controller was described in this journal in [10]. The rela-
a double integrator. tive simplicity of the 3rd order system enabled analytic solution
The double integrator with a P-PI or PID controller is a 3rd for a controller with double real zero. The PI-PI structure, ap-
order system expected to provide smooth responses with a pre- propriate for suppressing varying disturbances, was considered
scribed settling time. Identification of electrical parameters and as well. Simple rules for setting controller parameters in terms
step-by-step tuning of current (torque), velocity, and position of required settling time, design nomograms for discrete imple-
controllers for a PMSM servo-drive is described in [4]. Con- mentations, and limits on control cycle time were provided.
ventional frequency methods are applied for design in [2, 3], From a practical point of view it may be useful to indicate
given natural frequency and damping ratio. The considerations that a controller with settings obtained by continuous design re-
on the application of root-locus method [5] can be found in [6]. quires sufficiently fast sampling what, if time requirements are
The detailed design of a constrained state feedback controller high, implies the application of an embedded servo-controller
for a PMSM servo described in [7] may be an example of ad- or an external PLC dedicated solely to the servo. If however,
vanced approaches. Assessments of PD, LQG, minimum time, the PLC executes also some other tasks, the computational load
sliding mode and a few other algorithms for control of the dou- required by fast sampling may turn out unacceptable. In such
ble integrator at off-normal conditions are presented in [8]. In a case discrete design is needed, together with knowledge of
practice, however, industrial servos are usually tuned by trial- the maximum control cycle which provides the lowest compu-
and-error. tational load.
Some time ago the authors became interested in Internal
∗ e-mail:
Model Control (IMC) method popular in the process control
abozek@[Link]
community, particularly for plants with delay, e.g. [11,12]. IMC
Manuscript submitted 2021-05-10, revised 2021-09-22, initially design provides smooth responses defined by a time constant λ
accepted for publication 2021-10-19, published in February 2022. (usually multiple) being the only design parameter. Due to such

Bull. Pol. Acad. Sci. Tech. Sci., vol. 70, no. 1, p. e139957, 2022 1

© 2022 The Author(s). This is an open access article under the CC BY license ([Link]
 A. Bożek and L. Trybus

features, application of the IMC to servo design seemed inter- with

esting. TI 2
kR = kP and α= . (2b)
As it turned out the IMC solution is equivalent to pole place- 4 TI
ment (modal) with a triple pole 1/λ . The resulting PID con-
troller has complex zeros, so has to be implemented as "stand- So it is a PID controller with a double real zero −α and equiv-
alone" (Fig. 1), and not as the P-PI structure with two nested alent gain kR . Taking into account the double integrator, the
loops. By application of a quadruple pole placement, the PI-PI open-loop transfer function
structure can also be considered. In the case of discrete imple-
(s + α)2
mentations, the new design relaxes considerably the require- Gopen (s) = K , K = kR ko (3)
ments on control cycle time, which may be beneficial for high- s3
performance servos [7] or when a PLC executes other tasks in is of 3rd order. Conversion of the P-PI structure into PID gen-
parallel. erates the filter (see Section 4)
This paper, often referring to [10], is organized as follows.
The next section reviews briefly the previous root-locus de- α
F(s) = . (4)
sign and presents a relatively uncomplicated new solution based s+α
on triple pole placement. A discrete PID controller for which
the design becomes more involved is considered in Section 3. Root-locus plot of the closed-loop poles shown in Fig. 2a
A nomogram for the selection of controller settings and upper consists of two circular branches meeting at the double pole
limit on control cycle are given. Continuous and discrete PI-PI s2 and the 3rd real branch approaching the zero −α. To get
structures are dealt with in Sections 4 and 5, respectively. A lab smooth, critically damped responses we have to choose the
servo set-up with responses for the control cycle correspond- open-loop gain K corresponding to s2 . From the root-locus
ing to the upper limit is considered in Section 6. Continuous breakpoint condition dGopen /ds = 0 one obtains
control is tested as well by applying the shortest control cycle
available in the set-up PLC. Alternative PID design using IMC 27 3
s2 = −3α, K= α, s1 = − α. (5)
is presented in Appendix. 4 4
The pole s1 follows from the division of the closed-loop char-
2. CONTINUOUS PID CONTROLLER FOR DOUBLE acteristic polynomial s3 + K(s + α)2 by (s − s2 )2 .
INTEGRATOR
We begin with a review of the design from [10]. a)

2.1. Root-locus
Consider the control system of Fig. 1 involving a double inte-
grator, PID controller, and reference filter F. The double integra-
tor ko /s2 represents in practice a connection of servo-drive, ser-
0
vomotor and mechanical load resulting in the effective gain ko .

yr ko y
F PID
− s2

Fig. 1. PID control system for double integrator 0

b)
Let the controller be described by
kI
GPID (s) = kP + + kD s (1)
s
with the gains kP , kI , kD of corresponding PID actions. As
0
demonstrated later in Section 4, such a controller may be con-
verted into classical servo structure composed of PI velocity
loop and P position loop, provided that its transfer function has
real zeros. This in turn requires the ratio TI /TD of the integral
time TI = kP /kI to the derivative time TD = kD /kP to be no
smaller than 4. If TI /TD = 4, as in the familiar Ziegler-Nichols 0
tuning rules [13], the PID transfer function can be written as
Fig. 2. Root-locus plot for the PID control system with: a) double real
(s + α)2 zero, b) complex zeros (× – open-loop pole, ◦ – open-loop zero,
GPID (s) = kR (2a)  – closed-loop pole)
s

2 Bull. Pol. Acad. Sci. Tech. Sci., vol. 70, no. 1, p. e139957, 2022
 Tuning servo controllers by multiple pole placement

Let the system dynamic be specified by a required settling If natural frequency ωn is the preferred specification param-
time ts . Since ts is determined by the dominant pole s1 close to eter to calculate the settings, the corresponding settling time
−α (which speeds up the transient), so taking ts ∼= 3/|s1 | gives follows from ts ∼ = 4/ωn [5] (for unity damping ratio).
the modulus of the controller zero as The experimental step response of the designed PID system
4 strictly corresponding to P-PI is shown in Section 6, as well
α= . (6) as the response when the derivative action involves additional
ts
filtering.
Using it in (5) and (2b) yields the following PID settings
2.4. Frequency characteristics
216 432 27 To compare the robustness of the pole placement and root-
kP = 2 , kI = 3 , kD = . (7)
ts ko ts ko ts ko locus designs one can check sensitivity with respect to the servo
gain ko which may be inaccurately determined. Sensitivities
2.2. Pole placement
|dGclosed /dko | of the closed-loop transfer function for ko = 1,
To get a triple closed-loop time constant λ as in IMC (Ap-
λ = 1 (ts = 8) in relevant frequency range are shown in Fig. 3.
pendix) we remove the condition TI /TD ≥ 4 required by the
As seen, the pole placement design is slightly more sensitive to
transformation of PID to P-PI. The following pole placement
ko than the root-locus.
equation for the denominator of the closed-loop transfer func-
tion holds 1
pole placement
s3 + ko kD s2 + kP s + kI = (s + p)3 , p = 1/λ .


(8) 0.8
root-locus

Equating coefficients of the polynomials on both sides gives

0.6
3 1 3
kP = 2 , kI = 3 , kD = (9)
λ ko λ ko λ ko 0.4

(compare Appendix). Here the ratio TI /TD = kP 2 /(kI kD ) equals 0.2

3, which implies complex zeros of the new controller. So its
transfer function can be written as
0
0.1 1 10
(s + αc )(s + α c ) 3
GPID (s) = K , K= , (10a)
s λ ko
Fig. 3. Sensitivities of closed-loop transfer functions
with the complex zero with respect to ko
 
1 1
−αc = − 1+ j√ . (10b)
2λ 3 3. DISCRETE PID CONTROLLER
3.1. Basic derivation
To illustrate the difference from the previous design, root-
If a PLC implementing a continuous control algorithm cannot
locus plot of the new system for varying K is shown in Fig. 2b.
sample fast enough, in particular two hundred times per settling
By construction, the triple pole s3 of the closed-loop system
time for P-PI [10], the continuous design should be transformed
equals −1/λ and is reached for K in (10a). Unlike the previous
into equivalent discrete form. So let ∆ denote a discretization
plot (Fig. 2a), where the double zero −α stops the real branch
step (control cycle). Discrete time transfer function of the dou-
at s1 , here the complex zeros open up the way for this branch to
ble integrator ko /s2 has the following form [1, 5]
meet the two circular branches at s3 .
∆2 z + 1
2.3. Tunings Go (z) = ko . (13a)
Assume as before that a required settling time ts and the servo 2 (z − 1)2
gain ko are given. For the triple time constant λ we may write Likewise the PID controller becomes
ts ∼
= 8λ , so λ ∼
= ts /8. Now according to (9)
z kD z − 1 k1 z2 −k2 z+k3
192 512 24 GPID (z) = kP + kI ∆ + = , (13b)
kP = 2 , kI = 3 , kD = . (11) z−1 ∆ z z(z−1)
ts ko ts ko ts ko
where
So the new settings are only slightly different from the previous
ones in (7), hence very similar behavior may be expected. It is kD kD kD
k1 = kP + kI ∆ + , k2 = kP + 2 , k3 = . (13c)
also natural to assume the reference filter as ∆ ∆ ∆
Re αc The open-loop transfer function takes the form
F(s) = , (12)
s + Re αc
(z + 1)(K1 z2 − K2 z + K3 )
Gopen (z) = (14a)
where Re αc = 1/(2λ ) = 4/ts , i.e. the same as α in (6). z(z − 1)3

Bull. Pol. Acad. Sci. Tech. Sci., vol. 70, no. 1, p. e139957, 2022 3
 A. Bożek and L. Trybus

with
∆2
Ki = ko ki (14b)
2
for i = 1, 2, 3. Therefore the denominator of the closed-loop
transfer function becomes 0

f (z) = z(z − 1)3 + (z + 1)(K1 z2 − K2 z + K3 ). (15)

Following the continuous case, we want f (z) to have a triple

root
r = e−p∆ (16) 1
Fig. 4. Root-locus for discrete PID
with p = 1/λ .
It is known that if a function f (z) has a root r of multiplicity
3.3. Discretization step
k, then
The discrete form of the controller requires evaluation of the
0 00
f (r) = f (r) = f (r) = . . . = f (k−1)
(r) = 0. (17) step ∆ for which the design is feasible. The admissible ∆ is
determined by the lower limit condition r ≥ r4 . Using (16) for
Applying these conditions to f (z) for k = 3 one obtains a sys- r4 = 0.682 yields ∆ ≤ −(ln r4 )/p or ∆ ≤ 0.383λ . Since λ ∼ =
tem of three equations linear with respect to K1 , K2 and K3 , ts /10 for the quadruple time constant limit, we get
having the solution [14]
ts
∆≤ , (21)
26
K1 = C(3r3 + 8r2 + 5r − 4),
K2 = C(3r4 + 12r3 + 14r2 − 4r − 1), (18a) which is the requirement for discrete implementation providing
the settling time ts .
K3 = Cr3 (r2 + 4r + 7), Analogous requirement for the discrete P-PI (or, equiva-
lently, PID with a reference filter) having a double real zero
where yields ∆ ≤ ts /45 [10]. Thus the benefit from the pole placement
1−r
C= . (18b) design rests in almost twice longer limit on the discretization
(r + 1)3 step than in the former case.

3.4. Tunings and responses

3.2. Interval for r
For a feasible value of r (see below), root-locus plot with re- Given a required settling time ts , gain ko , and step ∆ one should
spect to the open-loop gain is shown in Fig. 4. The right part determine the corresponding value of r from (16) with p = 1/λ ,
of the plot with the triple breakpoint z3 = r corresponds to con- λ = ts /8, i.e. r = exp(−8∆/ts ). If the condition r ≥ r4 (= 0.682)
tinuous case (Fig. 2b), whereas the left one with the pole z1 does not hold, larger ts must be chosen. Then the parameters K1 ,
reflects the increased order of Gopen (z) from (14a) compared to K2 , K3 need to be calculated according to (18) and, finally, the
its continuous counterpart. The analytic expression for z1 may target settings obtained by the formulas
be obtained from the division
2(K2 −2K3 ) 2(K1 −K2 +K3 ) 2K3
kP = , kI = , kD = (22)
f (z) ko ∆2 ko ∆3 ko ∆
K1 (r), K2 (r), K3 (r)
= z − z1 , (19)
(z − r)3 derived from (13c) and (14b). Analytic properties of the func-
tions (18) assert that K2 −2K3 > 0, K1 −K2 +K3 > 0 and K3 > 0
which gives for r ∈ [r4 , 1). Hence (22) provide realizable settings of the dis-
crete PID controller.
(1 − r)(r2 + 4r + 7) K3
z1 = = 3. (20) In practical applications it could be often more convenient to
(r + 1)3 r calculate the settings from
With decreasing r the right and left part of the plot (Fig. 4) ρP ρI ρD
converge and meet at some value of r for which the equation kP = , kI = , kD = , (23)
950ko ∆2 9400ko ∆3 230ko ∆
z1 (r) = r √
holds. The only real solution of this equation gives
r = r4 = 4 8 − 1 ∼= 0.682, i.e. the unique quadruple real pole after reading out the values ρP , ρI , and ρD from the nomograms
of the closed-loop transfer function. Analytic properties of the in Fig. 5 for a given r. The nomograms have been obtained by
expression (20) assert that if r ∈ [r4 , 1) then z1 ∈ (0, r4 ], hence combining (18) and (22).
the system is stable and its dynamic is dominated by the triple Simulated step responses for the discrete PID and the P-PI
pole r or the quadruple if r = r4 . from [10] are compared in Fig. 6. The controller settings are

4 Bull. Pol. Acad. Sci. Tech. Sci., vol. 70, no. 1, p. e139957, 2022
 Tuning servo controllers by multiple pole placement

50 a)
45 yr ko y
PIP PIV
− − s2
40 v
35
s
30
25 b)
20
yr ko y
F R
15 − s2
10
5 Fig. 7. a) PI-PI servo control system, b) single-loop equivalent

0
0.7 0.75 0.8 0.85 0.9 0.95
single one of Fig. 7b, where [10]
r
Fig. 5. Nomograms for the discrete PID controller settings (kPV s + kIV )(s2 + kP s + kI )
R(s) = , (25a)
s2
calculated for the discretization step ∆ = 0.015 s as in labora- kP s + kI
F(s) = 2 . (25b)
tory experiments in Section 6. It implies the approximate lower s + kP s + kI
bounds on settling time equal to 26∆ ∼ = 0.4 s and 45∆ ∼= 0.7 s Before going into design issues note that by setting kI to 0
for the PID and P-PI, respectively. One can try to ignore the we get the classical P-PI structure represented by
condition ∆ ≤ ts /45 for P-PI to obtain a system as fast as for
PID, but then the step response becomes oscillatory (Fig. 6). (kPV s + kIV )(s + kP )
R(s) kI =0
= ,
s
(26)
kP
1 F(s) k =0 = .
I s + kP
0.8 So R(s) for kI = 0 is just the PID controller of Fig. 1. As seen
from above, such a controller may have real zeros only. If kP =
0.6 kIV /kPV , it has a double zero equal to −α in (2a).
y

0.4 4.2. Pole placement

For notational convenience, let us rewrite (25a) as
PID+F, ts = 0.4
0.2 (s + α)(s2 + β s + γ)
P-PI [10], ts = 0.7
R(s) = K (27a)
P-PI [10], ts = 0.4 s2
0
0 0.2 0.4 0.6 0.8 1 with
t [s]
kIV
Fig. 6. Step responses for the considered discrete PID and the P-PI K = kPV , α= , β = kP , γ = kI . (27b)
kPV
proposed in [10]
The open-loop transfer function Gopen (s) = R(s)ko /s2 is now of
4th order, so a quadruple pole may be required.
To simplify development we introduce normalization with
4. CONTINUOUS PI-PI CONTROL respect to α, obtaining the open-loop transfer function
4.1. Control structures
By applying PI control both for velocity and position loops we (s0 + 1)(s0 2 + β 0 s0 + γ 0 )
get the system of Fig. 7a, where Gopen (s) = K 0 , (28a)
s0 4
kIV with
PIV (s) = kPV + ,
s s Kko
(24) β γ
kI s0 = , K0 = , β0 = , γ0 = . (28b)
PIP (s) = kP + . α α α α2
s
Characteristic polynomial of the closed-loop transfer function
Additional integral in the position loop improves the accu- becomes
racy of servos tracking complicated trajectories and suppress-
ing varying disturbances. The two loops may be replaced by a f (s0 ) = s0 4 + K 0 (s0 + 1)(s0 2 + β 0 s0 + γ 0 ). (29)

Bull. Pol. Acad. Sci. Tech. Sci., vol. 70, no. 1, p. e139957, 2022 5
 A. Bożek and L. Trybus

We want this polynomial to have a quadruple root s04 , so a)

4
f (s0 ) = s0 − s04 . (30)

Equating coefficients of the expanded forms of (29) and (30)

yields the simple solution 0

K 0 = 8, β 0 = 2, γ 0 = 2, s04 = −2. (31)

One may comment that the solution can also be obtained from
the equations f (s04 ) = f 0 (s04 ) = f 00 (s04 ) = f 000 (s04 ) = 0 (see (17)
in Section 3). Roots of the polynomial s0 2 + 2s0 + 2, i.e. normal- 0
ized zeros of the controller, are −1 ± j1. b)
2
4.3. Tunings
As before, assume that a settling time ts and gain ko are given. 1
For the quadruple pole s4 = s04 α = −2α one may write ts ∼ =
10/|s4 |, hence 0
5
α∼ = . (32) -1
ts
-2
Values of the remaining parameters K, β , γ follow from (28b)
and those of the PI controllers from (27b). Finally, one obtains -3 -2 -1 0
the settings for the pair of controllers
Fig. 8. Root-locus plot for PI-PI structure with: a) three real zeros
10 50 (β double), b) two complex zeros and a real one
kP = , kI = 2 ,
ts ts
(33)
40 200 5. DISCRETE PI-PI CONTROL
kPV = , kIV = 2 .
ts ko ts ko We follow the steps of Section 3 with some extensions.
Step and ramp responses for such settings are presented in Sec- 5.1. Basic derivation
tion 6. Discrete implementation of the PI-PI system from Fig. 7a con-
sists of the double integrator (13a) and the controllers
4.4. Idea of the previous design [10]
Although not necessary, the term s2 + kP s + kI in (25a) has been z
GPI p (z) = kP + kI ∆ ,
assumed to have a double root −β , analogously as in the P-PI z−1
(36)
case. So the controller transfer function becomes z
GPIv (z) = kPV + kIV ∆ ,
z−1
(s + α)(s + β )2
R(s) = K . (34) whith the operator s replaced by (z − 1)/(∆z). As before, the
s2
two loops are equivalent to the single one of Fig. 7b with [10]
The corresponding root-locus plot shown in Fig. 8a has two
breakpoints sb1 and sb2 . Critically damped responses are ob- (z − α)(z − β )(z − γ)
R(z) = kR , (37a)
tained when the two breakpoints are reached simultaneously. ∆z(z − 1)2
This condition yields the following solution
∆z [(kP + kI ∆)z − kP ]
F(z) = , (37b)
16α (kI ∆2 + kP ∆ + 1)(z − α)(z − β )
K= , β = 2α . (35)
ko
where
Settling time is determined by the pair of dominant poles in the
breakpoint kR = (kPV + kIV ∆)(kI ∆2 + kP ∆ + 1),
√ 1
αβ = ,
sb1 = −2(2 − 2)α ∼
= −1.17α. 2
kI ∆ + kP ∆ + 1
kP ∆ + 2 (38)
α +β = ,
In the pole placement case, instead of the two breakpoints, kI ∆2 + kP ∆ + 1
we have one quadruple breakpoint s04 as in Fig. 8b, where the kPV
internal and external branches meet. γ= .
kPV + kIV ∆

6 Bull. Pol. Acad. Sci. Tech. Sci., vol. 70, no. 1, p. e139957, 2022
 Tuning servo controllers by multiple pole placement

In [10] it was assumed that α, β , γ are real (R). However, since 5.2. Intervals for r and ∆
the formulas above express the product and sum of the zeros α, The general shape of root-locus plot (not shown) is very similar
β , they remain correct also for complex (C) case provided that to the one of Fig. 4, but with the right part replaced by discrete
version of Fig. 8b. The fifth single pole z1 is obtained by substi-
β = α ∈ C, γ ∈ R. (39) tuting (42) into (41) and dividing the resulting f (z) by (z − r)4 .
This yields
We want the closed-loop system with a quadruple pole z4 = r.
Let us rewrite R(z) in the form similar to PID in (13b), (1 − r)(r + 3)(r2 + 2r + 5) K4
i.e. with the numerator linear with respect to the coefficients z1 = = 4. (45)
(r + 1)4 r
k1 z3 − k2 z2 + k3 z − k4
R(z) = . (40) As before,
√ the equation z1 (r) = r gives the unique real solution
z(z − 1)2 r5 = 5 16 − 1 ∼= 0.741 that represents the quintuple pole of the
closed-loop system. Moreover, properties of the expression (45)
For such R(z) the denominator of the closed-loop transfer func- imply that if r ∈ [r5 , 1) then z1 ∈ (0, r5 ], hence the system is
tion becomes stable with the dominant pole r.
This time the condition r ≥ r5 yields ∆ ≤ 0.3λ , so for ts ∼ =
f (z) = z(z − 1)4 + (z + 1)(K1 z3 − K2 z2 + K3 z − K4 ) (41)
12λ (quintuple λ ) we get
with Ki , i = 1, . . . , 4 defined in (14b). Applying the condi- ts
tion (17) with the multiplicity k = 4 to f (z) above, we get a ∆≤ . (46)
40
system of four equations linear with respect to K1 , . . . , K4 . Its
solution has the form [14] This is considerably better than ts /130 in [10] where the double
4 3 2 real zero prevented getting the quadruple pole (see Fig. 8a).
K1 = C(4r + 15r + 19r + 5r − 11),
K2 = C(6r5 + 30r4 + 55r3 + 35r2 − 25r − 5), 5.3. Tunings and responses
(42a)
K3 = C(4r6 + 20r5 + 44r4 + 45r3 − 11r2 − 5r − 1), Having kR , a, b, γ calculated from (43) and (44), the original pa-
rameters of the controllers are obtained by solving (38). Hence
K4 = Cr4 (r + 3)(r2 + 2r + 5),
b − 2a 1+a−b
where kP = , kI = ,
1−r a∆ a∆2
C= . (42b) (47)
(r + 1)4 a(1 − γ)kr
kPV = aγkr , kIV = .
∆
Comparing (37a) with (40) and taking into account (14b), we
get As in Section 3 we can combine the formulas (42), (43), (44),
2K1 (47) to get the following compact expressions for the settings
kR = (43)
ko ∆
ρP ρI
kP = , kI = ,
and α, β , γ as the roots of the equation K1 z3 −K2 z2 +K3 z−K4
= 300∆ 2000∆2
0. Assuming K1 6= 0, it is a cubic equation in a general form (48)
ρPV ρIV
whose real root γ, as well as the product a = αβ and the sum kPV = , kIV = .
110ko ∆ 930ko ∆2
b = α + β of the two other roots can be consecutively obtained
from the following expressions to satisfy the conditions (39) The coefficients ρP , ρI , ρPV , ρIV are read out from the nomo-
grams in Fig. 9 for a given r. The nomograms confirm that the
27K1 2 K4 − 9K1 K2 K3 + 2K2 3 PI-PI structure is realizable for the quadruple pole placement,
δ=
54K1 3 since all the settings are positive for r ∈ [r5 , 1).
Simulated step responses for the PI-PI controllers with the
p
27K1 2 K4 2 +(4K2 3 −18K1 K2 K3 )K4 +4K1 K3 3 −K2 2 K3 2
+ , settings developed in this section and those proposed in [10]
33/2 2K1 2
are compared in Fig. 10. The discretization step ∆ = 0.015 s
√ is used as in Section 3. Approximate lower bounds on settling
3K1 K3 − K2 2 K2
time are 40∆ ∼= 0.6 s and 130∆ ∼
3
γ= δ− √3
+ ∈ R, = 2.0 s for the new PI-PI settings
9 δ K1 2 3K1
(44) and the previous ones from [10], respectively. Again, one can
K4 K2 − γK1 try to ignore the condition ∆ ≤ ts /130 attached to the formulas
a = αβ = ∈ R, b = α +β = ∈ R,
γK1 K1 from [10] to have the settling time 0.6 s, but this leads to an
unstable system. The stability limit is reached for the settling
where δ is an auxiliary variable calculated first to get γ. time 0.75 s, as shown in Fig. 10.

Bull. Pol. Acad. Sci. Tech. Sci., vol. 70, no. 1, p. e139957, 2022 7
 A. Bożek and L. Trybus

50 6. EXPERIMENTAL RESULTS
45 A laboratory servo system shown in Fig. 11 has been set-up
40
to verify the designs. It consists of a ball screw linear actua-
tor, Estun AC servomotor and servo drive [15] with EtherCAT
35
bus, Beckhoff C6920 industrial PC [16] for PLC code execu-
30 tion, and standard PC with TwinCAT 3 engineering software
25 for programming. The continuous and discrete controllers have
20
been implemented in ST (Structured Text) language as function
blocks of the IEC 61131-3 standard [17]. All calculations de-
15
scribed by the formulas (11), (18), (22) for PID, and (33), (42),
10 (43), (44), (47) for PI-PI have been embedded in a PLC pro-
5 gram to compute target settings for given data. Although some
0
of the formulas look complicated, they can be expressed by ele-
0.75 0.8 0.85 0.9 0.95 mentary ST instructions, including EXPT function for the cube
r root in (44).
Fig. 9. Nomograms for the discrete PI-PI control structure settings The servo has been controlled using a combined reference
trajectory composed of step and ramp parts shown in Fig. 12.
Disturbance has been simulated by a signal added to controller
1 output.
The discrete control algorithms have been running with the
0.8 cycle ∆ = 15 ms. Their settings have been calculated for the
multiple poles corresponding to the shortest settling time, i.e.
0.6 for the design parameter r = r4 = 0.682 in the case of PID (Sec-
y

tion 3) and r = r5 = 0.741 in the case of PI-PI (Section 5). Con-

0.4 tinuous algorithms have been emulated by the Euler backward
approximation executed with the shortest cycle ∆min = 1 ms
0.2
PI-PI, ts = 0.6 available in the lab system.
PI-PI [10], ts = 2.0 Step responses of the originally designed discrete controllers
PI-PI [10], ts = 0.75
0
exhibit overshoots due to zeros of the transfer functions. Hence
0 0.5 1 1.5 2 the reference filters have to be added. Zeros of the PID con-
t [s] troller are the roots (complex) of the polynomial K1 z2 − K2 z +
Fig. 10. Step responses for the considered discrete PI-PI and the PI-PI K3 . Following the continuous filter (12), one can apply its 1st
proposed in [10] order discrete counterpart

industrial PC

servo drive

TwinCAT 3 software

servomotor

linear actuator

Fig. 11. Laboratory servo system

8 Bull. Pol. Acad. Sci. Tech. Sci., vol. 70, no. 1, p. e139957, 2022
 Tuning servo controllers by multiple pole placement

Fig. 12. Experimental results recorded by TwinCAT 3 software: green – reference position, blue – actual position, red – control output

(1 − z f )z a)
F1,PID (z) = , (49)
z−zf
0.1
where z f denotes the real part of the zeros, i.e. z f = 0.5K2 /K1 .
The full 2nd order filter 0.08

(K1 − K2 + K3 )z2
y [m]

0.06 yr
F2,PID (z) = , (50)
K1 z2 − K2 z + K3 PID+F1
0.04 PID+F2
slightly more complicated, cancels out both zeros. PI-PI+F1
The discrete PI-PI control structure has two real zeros 0.02 PI-PI+F2
P-PI [10]
kP kPV 0 PI-PI+F [10]
zfa = , zfb = , (51)
kP + kI ∆ kPV + kIV ∆ 0 0.5 1 1.5 2
t [s]
where z f a emerges in the connection of R(z) from (37a) with
F(z) from (37b) according to the diagram in Fig. 7b, and z f b b)
follows from (37a) and (38). One can eliminate the zero z f a by 0.1
the reference filter
0.08
(1 − z f a )z
F1,PI-PI (z) = (52)
z−zfa 0.06
y [m]

or, equivalently, by splitting the position PI controller to get

0.04
the I-P-PI structure proposed in [10]. If the filter (52) turns out
y
insufficient, one can use the following variant 0.02
r
PID+F1
(1 − z f a )(1 − z f b )z2 PI-PI+F1
F2,PI-PI (z) = (53) 0 PI-PI+F2
(z − z f a )(z − z f b )
0 0.2 0.4 0.6 0.8 1
that cancels both zeros. t [s]
Comparison of experimental step responses for the two con- Fig. 13. Step responses of control structures: a) discrete, b) continuous
trollers (PID, PI-PI) and related filters (F1, F2) is presented in
Fig. 13a. The configuration PID+F1 works correctly in general,
however, the motion is not smooth, with oscillations of velocity. The filters F2 in the structures PID+F2 and PI-PI+F2 cancel
There is no such behavior in the case of PID+F2. The structure all zeros affecting the shape of responses. Therefore the settling
PI-PI+F2 settles down aperiodically, whereas PI-PI+F1 exhibits times determined by (21) and (46) may be expected, namely
an overshoot of about 10%. Barely visible oscillations in the tsPID ≥ 26 · 0.015 s ∼
= 0.4 s and tsPI-PI ≥ 40 · 0.015 s = 0.6 s, for
vicinity of yr are due to static friction. Experimental responses PID+F2 and PI-PI+F2, respectively. As seen from Fig. 13a the
for the P-PI and PI-PI controllers tuned according to [10] are actual settling times are close to the estimates. Table 1 col-
also presented in Fig. 13a. Their settling times are longer, par- lects them and compares them with those obtained for the dou-
ticularly for the PI-PI, as shown earlier in Sections 3 and 5. ble zero design [10]. The approximate minimum settling times

Bull. Pol. Acad. Sci. Tech. Sci., vol. 70, no. 1, p. e139957, 2022 9
 A. Bożek and L. Trybus

Table 1 a)
Settling times of discrete control structures 0.08 y
r
PID
Multiple pole Double zero [10]
Controller PI-PI
0.06
MIN EX MIN EX PID+F1
PID+F2
PID / P-PI 26∆ 0.4 s 45∆ 0.7 s

y [m]
0.04 PI-PI+F1
PI-PI 40∆ 0.6 s 130∆ 2.0 s PI-PI+F2

0.02
(MIN) in terms of ∆ are given, along with the values obtained
in the experiments for ∆ = 15 ms (EX).
The continuous control has been emulated for the shortest 0
available cycle ∆min = 1 ms. The controllers are tuned for ts = -0.2 0 0.2 0.4 0.6 0.8
0.5 s, being actually the average of settling times obtained in the t [s]
discrete control (0.4 vs. 0.6 s). Experimental step responses are b)
presented in Fig. 13b. The responses of the structures without 0.08 y
r
reference filters exhibit overshoots, as in the case of discrete
PID
controllers. The control structure PID+F1 with the filter (12) PI-PI
0.06
completely eliminates the overshoot. For the PI-PI structure, PID+F1
the inertial filter PI-PI+F1
y [m]

0.04 PI-PI+F2
kI 1
F1,PI-PI (s) = = t (54)
kP s + kI s
s+1
5 0.02

is applied, which is equivalent to splitting the position PI into

0
I and P, as proposed in [10]. However, this filter leaves a few
percent overshoot. To eliminate the overshoot completely, the -0.2 0 0.2 0.4 0.6 0.8
following extended filter can be used t [s]
Fig. 14. Ramp tracking for control structures: a) discrete,
kI kIV 1 b) continuous
F2,PI-PI (s) = · = 2 . (55)
kP s + kI kPV s + kIV ts
s+1
5 0.01
All the responses are smooth and settle down according to the 0.005
assumed settling time 0.5 s.
error [m]

Note that for ∆min = 1 ms we may get the shortest settling 0

time ts = 26 ms ∼ = 30 ms of the discrete PID servo control pro-
-0.005
vided by the Beckhoff PLC or other modern PLCs.
The integral actions of the plant and controllers should pro- PID PI-PI
-0.01
vide exact tracking of higher-order reference trajectories. Of -1 0 1 2 3 4 5 6
course, the tracking will be error-free only without the reference t [s]
filters which, as shown for the ramps in Figs. 14a, b, introduce
unavoidable delays.
Control error responses of continuous structures due to dis-
disturbance

turbances acting at steady-state are compared in Fig. 15. Al-

though the PI-PI structure suppresses the step and ramp distur-
bance much better, it suffers from a visible limit cycle due to 0
friction. So in practice a tracking PI-PI controller may switch
into PID control mode while approaching the steady-state.
Guidelines on how to reduce to some extent the effect of friction -1 0 1 2 3 4 5 6
on servo behavior can be found in [2, 3]. t [s]
In practical implementations, so-called real PID algorithm is Fig. 15. Disturbance suppression in continuous control structures
most often used with the filtered derivative block, here in the
form kD s/(TD s/D + 1), TD = kD /kP , to suppress measurement
noise and high-frequency disturbances. In the experiments, two 2) real version as above with D = 5 (PIDR). Note that D = 5
implementations of the continuous PID with the reference fil- is a default value in a number of industrial controllers. As seen
ter (12) have been compared: 1) basic derivative kD s (PID), in Fig. 16, step responses of the position y almost overlap. How-

10 Bull. Pol. Acad. Sci. Tech. Sci., vol. 70, no. 1, p. e139957, 2022
 Tuning servo controllers by multiple pole placement

ever, the control signal u for the PIDR is a decaying oscillation APPENDIX
due to the increased order of the system and unchanged set- Following notation from [11], let G denote a plant transfer func-
tings. Since high-resolution encoder is used in the servo, the tion, G
e its approximation such that Ge=G e+ Ge− where G e+ in-
oscillations may be easily reduced by increasing D. Neverthe- cludes delays and right half-plane zeros, and has unity steady-
less, extension of the pole placement method on arbitrary D in state gain. To analyze the robustness of the approximation, an
the real PID remains a subject of future work. IMC controller R∗ is defined
1
0.1 R∗ = f, (56)
G−
e

where f denotes a low-pass filter that specifies a closed-loop

y [m]

0.05 e+ f ). The controller R∗ is

transfer function (actually equal to G
yr
designed for a feedback system involving G − G e as a place-
PID+F
holder plant. It can be converted into the standard system with
0 PIDR+F (D=5)
the original plant G and the controller R, where
-0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
t [s] R∗
R= . (57)
1 − R∗ G
e
PID+F
0.15 PIDR+F (D=5) In the case of the double integrator and perfect approxima-
tion we have G = ko /s2 = G e=G e− since G e+ = 1. Taking into
u [-1..1]

0.1
account integral action of the controller, assume the closed-loop
0.05 transfer function as
0 as2 + bs + 1
f= , (58)
-0.05 (λ s + 1)3
-0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
t [s] so with a triple time constant λ . Using (58) in (56), and
Fig. 16. Step responses of the servos with ideal and real continuous then (56) in (57) gives
PID controllers
1 s(as2 + bs + 1)
R= . (59)
ko λ 3 s2 + (3λ 2 − a)s + (3λ − b)
7. CONCLUSIONS
By taking a = 3λ 2 and b = 3λ the controller R becomes of PID
New tuning rules for PID and PI-PI servo controllers have been
type, namely
derived using multiple pole placement approach, as an alter-
native to the root-locus design [10]. The root-locus assumes 3 1 1 3
real zeros of the PID controller, so it can be converted into the R= + + s. (60)
λ 2 ko λ 3 ko s λ ko
classical P-PI two-loop structure. This is not the case for the
pole placement PID which has complex zeros, so it must be
implemented in a standard single loop. The new approach ad- ACKNOWLEDGEMENTS
mits twice longer control cycle for the discrete PID and four Discussions with Dr T. Żabiński and Prof. R. Leniowski are
times for PI-PI. This in turn implies that twice or four times gratefully acknowledged.
shorter settling times can be obtained for the same control cy-
cle. Shorter settling time may be important in the case of high-
performance servos, whereas longer control cycle suits heavily REFERENCES
loaded PLCs executing demanding tasks in parallel. Actually, [1] B. Siciliano and O. Khatib, Eds., Springer Handbook of
the pole placement approach has been motivated by an earlier Robotics. Berlin Heidelberg: Springer, 2008.
IMC solution for a double integrator. [2] G. Ellis, Ed., Control System Design Guide, 4th ed. Butterworth-
Nomograms for tuning the discrete PID and PI-PI controllers Heinemann, 2012.
given a settling time and control cycle are provided. They may [3] G.W. Younkin, Industrial Servo Control Systems, 2nd ed. New
be useful to obtain the settings quickly in practical applications. York: Marcel Dekker, 2002.
[4] S.-M. Yang and K.-W. Lin, “Automatic Control Loop Tuning for
However, the design formulas can also be directly implemented
Permanent-Magnet AC Servo Motor Drives,” IEEE Trans. Ind.
using an IEC 61131-3 engineering software. Electron., vol. 63, no. 3, pp. 1499–1506, 2016.
Lab experiments have confirmed the feasibility of the pole [5] G.F. Franklin, J.D. Powell, and A.F. Emami-Naeini, Feedback
placement design. In particular, the smooth shape and settling Control of Dynamic Systems, 7th ed. Reading: Addison-Wesley,
time of the step responses have been verified both for continu- 2019.
ous and discrete cases, as well as the ability of ramp tracking [6] L. Sciavicco and B. Siciliano, Modelling and Control of Robot
and disturbance suppression. Manipulators. London: Springer, 2000.

Bull. Pol. Acad. Sci. Tech. Sci., vol. 70, no. 1, p. e139957, 2022 11
 A. Bożek and L. Trybus

[7] T. Tarczewski, M. Skiwski, L.J. Niewiara, and L.M. Grzesiak, [12] C. Grimholt and S. Skogestad, “Optimal PI and PID control of
“High-performance PMSM servo-drive with constrained state first-order plus delay processes and evaluation of the original and
feedback position controller,” Bull. Pol. Acad. Sci. Tech. Sci., improved SIMC rules,” J. Process Control, vol. 70, pp. 36–46,
vol. 66, pp. 49–58, 2018. 2018.
[8] V. Rao and D. Bernstein, “Naive control of the double integra- [13] K.J. Åström and T. Hägglund, Advanced PID Control, Research
tor,” IEEE Control Syst. Mag., vol. 21, pp. 86–97, 2001. Triangle Park, 2005.
[9] P.B. Schmidt and R.D. Lorenz, “Design principles and imple- [14] “Maxima CAS homepage.” [Online]. Available: [Link]
mentation of acceleration feedback to improve performance of [Link]/.
DC drives,” IEEE Trans. Ind. Appl., vol. 28, no. 3, pp. 594–599, [15] “ESTUN Industrial Technology Europe.” [Online]. Available:
1992. [Link]
[10] T. Żabiński and L. Trybus, “Tuning P-PI and PI-PI controllers [16] “BECKHOFF New Automation Technology.” [Online]. Avail-
for electrical servos,” Bull. Pol. Acad. Sci. Tech. Sci., vol. 58, pp. able: [Link]
51–58, 2010. [17] EN 61131-3, Programmable controllers – Part 3: Programming
[11] D.E. Seborg, T.F. Edgar, D.A. Mellichamp, and F.J. Doyle, Pro- languages (IEC 61131-3:2013), International Standard, CEN-
cess Dynamics and Control, 4th ed. New York: Wiley, 2016. ELEC Std., 2013.

12 Bull. Pol. Acad. Sci. Tech. Sci., vol. 70, no. 1, p. e139957, 2022