You are on page 1of 5

Optimal Tuning for a Double Integrator

Plus Dead Time


Igor Bélai, Pavol Bisták and Mikuláš Huba
Slovak University of Technology in Bratislava
Ilkovičova 3, 812 19 Bratislava, Slovakia
Email: pavol.bistak, mikulas.huba, igor.belai@stuba.sk

Abstract—This paper deals with an optimal disturbance ob- of a simple dead time. An extensive experimental research
server (DO) based filtered PID (DO-FPID) controller design [10], [11], [12] fully confirmed these theoretical results. But,
based on the performance portrait method solved for a PD it also showed that derivation of an optimal controller tuning
controller + double integrator + dead time loop configuration.
By virtue of an equivalence of a dead time and n-th order filters for the closed loop with the double integrator + dead time
used in DO-FPID control, the derived results may be used in based on the performance portrait method working with a
a broad range of tasks originating in a control of plants with performance portrait generated by the setpoint step responses
a dominant 2nd order dynamics, which will be illustrated by a may not guarantee shape related features and optimality of the
noise attenuation motivated design for a positional DC motor disturbance responses.
control.
This paper re-considers design of a filtered PD (FPD) and a
DO based filtered PID (DO-FPID) controllers for a dominant
I. I NTRODUCTION
2nd-order plant dynamics with the input disturbance di , with
Design of an optimal filter and its influence on the resulting x = (y, ẏ)0 , ẏ = dy/dt and y being the plant state, or output
closed loop robustness and performance has been recently
treated both in the traditional PID control [1], as well as in ÿ = Ks (ur + di ) − a1 ẏ − a0 y (1)
the disturbance observer (DO) based control [2], [3], [4], [5], Such a plant may be described by a “pole-zero form“ transfer
whereby the robustness and performance have been analyzed function
in the frequency domain. It has been showed that by increasing  
Y (s) Ks
the filter order the robustness decreases, although the per- F (s) = = 2 (2)
formance may increase, but without a clear recommendation Ur (s) di =0 s + a1 s + a0
regarding an optimal filter degree choice. Design of an optimal This dominant loop dynamics will further be extended by a
PD controller for the double integrator + dead time plant model dead time Td that may characterize not only an intrinsic loop
represented one of the first applications of the performance dead time, but also effect of intentionally introduced filters
portrait method [6] for an optimal and robust controller tuning. used for the loop robustification, as well as a noise attenuation.
The optimal controller design based on searching in a closed For quantitative evaluation of the speed of responses the
loop performance portrait generated over normed controller IAE (Integral of Absolute Error) will be used defined as
parameters for a tuning yielding minimal IAE (Integral of Z ∞
Absolute Error) values of the setpoint step responses and IAE = |e(t)| dt ; e = w − y ; w = setpoint (3)
satisfying additional shape related constraints put on the plant 0
input and output was firstly formulated in plane of the normed However, in difference to the relatively simple task met in
controller gains. Later, this task has been resolved by express- optimization of the DO based filtered PI control [13], where
ing the optimal tuning in terms of equivalent closed loop poles the tuning optimal for the setpoint responses guarantees also
[7]. These may be introduced as figures that after substituting optimal control for the disturbance steps, design of an optimal
for poles into a pole assignment controller derived for a plant filtered PD control for a double integrator+dead time plant
without time delays yield controller parameters optimal for has firstly to analyze separately an optimal tuning of both
the loop with a chosen time delay. Again, the search has been the setpoint step responses denoted by IAEs and of the
carried out on the performance portrait generated by analyzing disturbance step responses denoted IAEd . In [14] the authors
setpoint step responses. tried to solve a similar task by a modified minimization
In [8], [9] it has been shown that a design of a modified problem with the cost function chosen as
disturbance observer based filtered PID controller motivated
J = ws Js + wd Jd ; ws + wd = 1 ; ws ∈ h0, 1i
by achieving an improved noise attenuation, while keeping the (4)
Js = IAEs /IAEs,min ; Jd = IAEd /IAEd,min
best possible performance, may be simply carried out when
based on an equivalence of time delays. Thereby, effect of an More attention to balancing this PI control trade-off has been
n-th order filter used both in the outer (stabilizing controller) given recently also in [15], [16], [17], [14]. They observed
and in the inner (DO) loops may be equivalent to an effect that for the controller tuned to optimize the disturbance
response (regulatory mode), the setpoint tracking (servo mode) TV2(us) TV2(ud)
is deteriorated, and conversely. In [15] the authors formulated 0.8 0.8 0.3
0.1
the problem of optimally balanced tuning graphically in the 0.001
0.1

κD=KDKsTd
0.6 0.6
0.001
plane of IAEs and IAEd values of both responses and tried to
01 0.1
get optimal solution by minimizing the distance between the 0.4 0.0 0.4 0.3
0.3 01 1
1 0.0 23
actual point (IAEs , IAEd ) and an ideal (not reachable) point 0.2
23
0.2
0.1 10
0.05 0.1 0.15 0.2 0.05 0.1 0.15 0.2
with the coordinates composed of the minimal IAEs,min and
TV0(ys) TV1(yd)
IAEd,min achieved by a separate optimization of the setpoint 0.8 0.8 0.001
and disturbance responses 01 1
0.0 0. 3

1
κD=KDKsTd

0.
q 0.6 0. 0.6 0.3 1
1 2
J1 = (IAEs − IAEs,min )2 + (IAEd − IAEd,min )2 (5) 0.4
2
0.4 .00
1 3
01 .1 3 0
0.0 0
0.3 00..131 10
1 10 23 30
0.2 0.2
To get simple solutions that fit into the limited space of a 0.05 0.1 0.15 0.2 0.05 0.1 0.15 0.2

conference paper, this paper will avoid a deeper analysis by IAEs(ys) IAEd(yd)

considering the cost function 0.8 0.8

30
6

κD=KDKsTd
0.6 0.6
4
J = IAEs + IAEd (6) 8

15
6

15
0.4 6 0.4 15
8
It is of higher importance to include into the performance 8 15
30
30 30
600
10
0.2 0.2
optimization the tolerable integral deviations from ideal shapes 0.05 0.1 0.15
2
0.2 0.05 0.1 0.15
2
0.2
κP=KPKsTd κP=KPKsTd
of the setpoint and disturbance step responses at the plant input
and output [18], [13], [8], [10], [12] that may be formulated
in form of inequalities Figure 1. Performance portrait of the PDO FPID positional control with
different optimal tunings: red - min(IAEs + IAEd ), comlex poles (16);
T V0 (ys ) ≤ ys ; T V1 (yd ) ≤ yd blue - min(IAEs + IAEd ), real poles (20); green - min(IAEs ), complex
(7) poles (21); cyan - min(IAEs ), real poles (22)
T V2 (us ) ≤ us ; T V2 (ud ) ≤ ud
specified by the vector ys , yd , us , ud with the index s
corresponding to the setpoint and d corresponding to the In [10] the closed loop transfer functions of a PDO-FPID
disturbance response. Thereby, we will denote as strictly MO control gave the closed loop transfer functions
disturbance responses those characterized by T V1 (yd ) = 0.
Y (s) K P K s e Td s
When wishing to define an optimal control as something Fr (s) = R(s) = (s2 +a1 s)eTd s +Ks (KP +KD s)
unique, it should be characterized in the simples possible way, Y (s) Ks (eTd s −1)
(12)
Fi (s) = Ui (s) = (s2 +a1 s)eTd s +Ks (KP +KD s)
as for example
In the case of MO setpoint and 1P disturbance responses, when
 = ys = yd = us = ud → 0 (8) the control error does not change its sign, the IAE values
corresponding to unit input steps of particular inputs have been
But, in order to respect the always limited precision of control derived as
and also that of computer simulations, some ”sufficiently”
a1 +Ks KD −KP Ks Td Td
small IAEs = KP Ks ; IAEi = KP (13)
 = ys = yd = us = ud = 0.001 (9) To simplify the used formulas, one may introduce normed
coefficients
will be chosen instead. As an optimization result, optimal
parameters KP , KD of the PD control algorithm K P = KP Ks Td2 ; K D = KD Ks Td ; p = Td s (14)

ur = KP (r − y) − KD ẏ + uc ; uc = a0 r/Ks − di (10) An analytical derivation based on a multiple real dominant


pole po then yields optimal values
will be found and then expressed by means of the so called √
po = −2 + 2√
equivalent poles α1 and α2 related to KP , KD by the pole √
K P o = 2e−2+ √2 (5 2 − 7) ≈ 0.079 (15)
assignment control of a delay-free plant giving −2+ 2

K Do = 2e ( 2 − 1) ≈ 0.461
α1 α2 −a0
KP = Ks ; KD = − α1 +α
Ks
2 +a1
(11) Basic problem in dealing with a pure dead time PDO-FPID
system numerically is that one needs to work with a DO filter
II. L OOP ANALYSIS FOR A DOUBLE INTEGRATOR BY THE
with n ≥ 2 and a time constant Tn . Thus, the parameters that
PERFORMANCE PORTRAIT METHOD (PPM)
would be used as a basis for a general case may be derived
Generation of a performance portrait for an optimal con- just in a limit for Tn → 0. This means that one has to work
troller tuning by the performance portrait method has already with some sufficiently short time constants Tn << Td and to
been described in [6], [7], [18], [13]. try to extrapolate the situation with Tn = 0 [13].
Setpoint Step Responses − Plant Output Disturbance Step Responses − Plant Output
1 2
min(IAEs+IAEd), complex p.
0.9 1.8 min(IAEs+IAEd), real poles
min(IAEs), complex poles
0.8 1.6
min(IAEs), real poles
0.7 1.4

0.6 1.2
−−−> y(t)

−−−> y(t)
0.5 1

0.4 0.8

0.3 min(IAEs+IAEd), complex p. 0.6

min(IAEs+IAEd), real poles


0.2 0.4
min(IAEs), complex poles
0.1 0.2
min(IAEs), real poles
0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
−−−> t −−−> t

Setpoint Step Responses − Controller Output Disturbance Step Responses − Controller Output
0.15 0
min(IAEs+IAEd), complex p. min(IAEs+IAEd), complex p.
min(IAEs+IAEd), real poles −0.2 min(IAEs+IAEd), real poles
min(IAEs), complex poles min(IAEs), complex poles
0.1
−0.4
min(IAE ), real poles min(IAE ), real poles
s s

−0.6
0.05

−−−> u(t)
−−−> u(t)

−0.8

−1
0
−1.2

−1.4
−0.05

−1.6

−0.1 −1.8
0 5 10 15 20 25 30 0 5 10 15 20 25 30
−−−> t −−−> t

Figure 2. PDO FPID positional control - setpoint steps Figure 3. PDO FPID positional control - disturbance steps

A. Optimal PD controller parameters by the PPM equivalent poles

By solving the optimization problem with the cost function 0.280 ± j0.197
α1,2o = − (18)
(6) with the additional shape related constraints (7),(9) over a Td
2D performance portrait consisting of 51x51 points generated When wishing to work with real equivalent poles, one has to
with n = 2, Tn = 0.05 and a sampling period 0.001 for add to the optimization an additional constraint given as
K P ∈ (0.02, 0.2) and K D ∈ (0.2, 0.8) (Fig. 1) one gets
2
the optimal values of the normed controller parameters (14) K D − 4K P ≥ 0 (19)
slightly different from (15) as
The calculations then yield optimal parameters
K oP = 0.117 ; K oD = 0.560 (16)
α1,2 = −0.314/Td ; K P = 0.085 ; K D = 0.584 (20)
Sometimes, as e.g. in a constrained control [11], it may be Figs 1-3 then show comparison of the found controller gains
advantageous to used instead of the optimal controller gains and poles (16), (18), (20) and of the corresponding transient re-
the so called optimal equivalent closed loop poles that may sponses with optimization carried out by using a performance
be introduced as figures that after substitution into the pole portrait generated just by the setpoint step responses. It is to
assignment controller (11) yield the optimal controller gains note that by the optimization reported in [6], [7] it was possible
(16). In such a way one gets for the double integrator with to achieve faster transients due to the neglected shape related
a0 = a1 = 0 the equivalent poles constraints on the disturbance responses (see Fig. 1). In such
1
 q  a case one may get the optimal working points
2
α1,2 = − K D ± K D − 4K P (17)
2Td 0.280 ± 0.215
α1,2 = − ; K P = 0.124 ; K D = 0.560
Td
Thus, for the optimal gains (16) one gets complex conjugate (21)
or
Measurement: Rotor position
α1,2 = −0.352/Td ; K P = 0.103 ; K D = 0.644 (22)
1
However, the corresponding T V2 (ud ) values are already higher
n=2
then the tolerable value  = 0.001. n=3
0.8
n=4
III. I LLUSTRATIVE EXPERIMENT n=5

Position [rad]
n=6
0.6
This illustrative experiment deals with the DO-FPID posi-
tional control of HSM 150 DC motor. The controller tuning
uses the identified parameters: 0.4

J = 0.00012 [kgm2 ]; moment of inertia


B = 0.00016 Nm.s.rad−1 ; viscous friction 0.2
TGM = 0.00025 [s]; torque generator time constant
∆ φ = 6.283/10000; [rad]; position resolution 0
0 0.2 0.4 0.6 0.8 1
The rotor speed has been evaluated by differentiating the t [s]
position signal from an incremental encoder,which is known
Measurement: Reference torque
to be a source of the quantization noise with an amplitude 0.3
n=2
increasing by decreasing the sampling period. n=3
0.25
When considering the plant model (1) one gets Ks = n=4
n=5
1/J = 8333.3, a1 = B/J = 1.3333 and a0 = 0. All 0.2 n=6

internal delays (including the torque generator time constant


0.15
TGM , the electrical time constant and the sampling period T*m [Nm]
Ts = 0.5ms) will be approximated as the plant dead time 0.1
Tdp = TGM + Ts = 0.75ms.
0.05
The possibility to approximate dynamics of a positional
servo control by a double integrator is well known in practice 0
and may e.g. be found in [19]. One formal advantage of the
dead time approximation is that the impact of two dead time −0.05

terms corresponds to their sum. Thus, by allocating a second −0.1


0 0.2 0.4 0.6 0.8 1
dead time term Tdf to describe the impact of the filtration, t [s]
one may use a controller tuning corresponding to the total
dead time Figure 4. Measured position and torque step responses - unconstrained steps,
Td = Tdp + Tdf (23) Td = 5ms, sampling period Ts = 0.25ms, setpoint step 1 rad, disturbance
step 0.1N m
By means of the above introduced equivalent poles substituted
into the constrained PD controller [11] and of the equivalence
of the time delays [8], [9] it is the easy to calculate for a
chosen n the filter time constants Measurement: Rotor position (detail)
1.005
2(n + 1) − R p
Tn = √ Tdf ; R = 2n(n + 1) (24)
(2 − 2)(n + 1)(n + 2)
1
Thus, one may easily tune the disturbance observer based
constrained filtered PID control and to show effect of the 0.995
Position [rad]

newly derived equivalent poles and of the filter order n on per-


n=2
formance of the resulting setpoint and disturbance responses. 0.99 n=3
As it is obvious from the zoomed position signal in Figs 4- n=4
n=5
6, application of higher order filters allows to achieve smooth 0.985 n=6
transients with a possible overshooting not exceeding two
encoder increments (or less, see Fig. 6 for Td = 3ms) and 0.98
without increasing the torque ripple.
0.975
IV. C ONCLUSION 0 0.2 0.4 0.6 0.8 1
t [s]
This paper considered analysis of the optimal tuning of a
predictive disturbance observer based filtered PID controller Figure 5. Zoomed measured position - unconstrained steps
tuning for a double integrator and an application of the
achieved controller in the positional DC motor control.
Measurement: Rotor position, m = 1 [2] E. Sariyildiz and K. Ohnishi, “Design constraints of disturbance observer
5.005
in the presence of time delay,” in Mechatronics (ICM), 2013 IEEE
T = 3 [ms]
International Conference on, Feb 2013, pp. 69–74.
T = 5 [ms]
[3] E. Sariyildiz and K. Ohnishi, “Analysis the robustness of control systems
5 based on disturbance observer,” International Journal of Control, vol. 86,
no. 10, pp. 1733–1743, 2013.
[4] E. Sariyildiz and K. Ohnishi, “A guide to design disturbance observer,”
J. Dyn. Sys., Meas., Control, vol. 136, no. 2, 2013.
Position [rad]

4.995
[5] E. Sariyildiz and K. Ohnishi, “Performance and Robustness Trade-off in
Disturbance Observer Design,” in 39th Annual Conference of the IEEE
Industrial Electronics Society (IECON). Vienna, Austria: IEEE, 2013,
4.99 pp. 3679–3684.
[6] M. Huba, “Designing robust controller tuning for dead time systems,”
in Int. Conf. System Structure and Control. Ancona, Italy: IFAC, 2010.
4.985 [7] M. Huba, “Robust controller tuning for constrained double integrator,” in
Preprints of the Workshop Selected topics on constrained and nonlinear
control, M. Huba and S. Skogestad, Eds. STU Bratislava - NTNU
Trondheim, 2011, pp. 203–211.
0 1 2 3 4 5 6 7 8
[8] M. Huba, “Open flexible PD-controller design for different filtering
t [s] properties,” in 39th Annual Conference of the IEEE Industrial Elec-
tronics Society (IECON). Vienna, Austria: IEEE, 2013.
[9] M. Huba, “Modular PID-controller design with different filtering prop-
Figure 6. Zoomed position of a constrained control according to [11], Td = erties,” in 39th Annual Conference of the IEEE Industrial Electronics
3 − 5ms, sampling period Ts = 0.25ms, setpoint step 5 rad, disturbance Society (IECON). Vienna, Austria: IEEE, 2013.
step 0.08N m, n = 6 [10] M. Huba and I. Bélai, “Comparison of Two Approaches to a Positional
Servo Control,” in 15th Int. Carpathian Control Conference - ICCC,
Velké Karlovice, Czech Republic, 2014.
[11] M. Huba, I. Bélai, and P. Bisták, “Noise attenuation motivated controller
Numerical analysis of the optimal PD controller tuning by design. Part II: Position control,” in Speedam Symposium, Ischia, Italy,
the performance portrait method gives results that are very 2014, pp. 1331–1336.
close to the analytical controller tuning by the triple real [12] M. Huba and I. Bélai, “Experimental evaluation of a DO-FPID controller
with different filtering properties,” in IFAC World Congress, Cape Town,
dominant pole method. The derived optimal controller gains, South Africa, 2014.
or the optimal equivalent poles may be used for tuning both [13] M. Huba, “Comparing 2DOF PI and Predictive Disturbance Observer
the unconstrained and constrained controllers, whereby they Based Filtered PI Control,” Journal of Process Control, vol. 23, 10, pp.
1379–1400, 2013.
guarantee outstanding transients with a negligible overshooting [14] C. Grimholt and S. Skogestad, “Optimal PI-Control and Verifcation of
(below two increments of the incremental encoder) that seem the SIMC Tuning Rule,” IFAC Conf. Advances in PID Control PID’12,
already to be close to the physical performance limits of the vol. ThPl.1, 2012.
[15] O. Arrieta and R. Vilanova, “Simple PID Tuning Rules with Guaranteed
considered drive. Ms Robustness Achievement,” IFAC World Congress Milano, vol. 18,
1, 2011.
ACKNOWLEDGMENT [16] S. Alcántara, W. Zhang, C. Pedret, R. Vilanova, and S. Skogestad, “IMC-
like analytical design with S/SP mixed sensitivity consideration: Utility
This work has been partially supported by the grants APVV- in PID tuning guidance,” Journal of Process Control, vol. 21,6, pp. 976
0343-12 Computer aided robust nonlinear control design and – 985, 2011.
VEGA 1/0937/14 Advanced methods for nonlinear modeling [17] S. Alcántara, R. Vilanova, C. Pedret, and S. Skogestad, “A look into
robustness/performance and servo/regulation issues in PI tuning ,” in
and control of mechatronic systems. IFAC Conf. Advances in PID Control PID’12, Brescia, Italy, 2012.
[18] M. Huba, “Performance measures, performance limits and optimal PI
R EFERENCES control for the IPDT plant,” Journal of Process Control, vol. 23, 4, pp.
[1] V. R. Segovia, T. Hägglund, and K. Aström, “Measurement noise 500–515, 2013.
filtering for PID controllers,” Journal of Process Control, vol. 24, no. 4, [19] M. Huba, “Tuning of a Filtered Pole Assignment Controller for an
pp. 299 – 313, 2014. Integral Plant,” in 15th Int. Carpathian Control Conference - ICCC,
Velké Karlovice, Czech Republic, 2014.

You might also like