Tuning the leading roots of a second order DC

servomotor with proportional retarded control

R. Villafuerte ∗ and S. Mondié ∗∗
∗ Department of Applied mathematics of Papaloapan University (UNPA)
Av. ferrocarril s/n Ciudad Universitaria Loma Bonita, Oax. 68400 México.
e-mail: [email protected], [email protected].
∗∗ Departament of Automatic Control CINVESTAV-IPN
A.P. 14-740 07360, México DF.
e-mail: [email protected].

Abstract: The stabilization of a second order system through a proportional delay controller insuring
a specified closed loop exponential decay s is studied. The analysis in the frequency domain allows
the determination of the s -stabilizability regions of the controller. The locus of the dominant roots is
analyzed in detail and the characterization of some key loci, including the maximum achievable decay
is obtained. The tuning of a DC servomotor experimental setup illustrates the results.
keyword:Second order system, s -stable region, proportional retarded control, DC servomotor.

1. INTRODUCTION A drawback of such a proportional-delay control strategy is

that the control design and response shaping tools available for
The second order system of the form achieving prescribed closed loop specifications of delay free
q¨ (t) + 2d n q˙ (t) + n 2 q (t) = bu(t) (1) systems cannot be used. The main reason is that the analysis
of the roots of the closed loop characteristic equation, the
where n > 0, d > 0 and b > 0, is the first choice model quasipolynomial
for a wide range of physical processes such as servomotors,
linearized oscillators, etc. p(s, k p , kr , h) = s2 + 2d n s + n 2 + bk p − bkr e−hs (5)
with infinite number of roots, is now a more complex task.
The state space description given by
As in the case of delay free systems, when the system has no
0 1 0 zeros, the shaping of the response depends first on the distance
   
x(t) = x(t) + u(t), of the rightmost eigenvalue(s) of the closed loop system to the
−n 2 −2d n b
imaginary axis and second, on the nature of these dominant
q (t) = ( 1 0 ) x(t), roots, real or complex, which define the oscillatory nature of
is clearly controllable hence one can assign arbitrarily the the response.
closed-loop spectrum with a full state feedback, a proportional
derivative feedback of the form This response shaping is closely related to the stabilization
with prescribed exponential decay, named s -stabilizability.
u(t) = −k1 q (t) − k2q˙ (t). (2) This problem has been addressed in the framework of Linear
Here, we want to avoid the design of an observer. We also want Matrix Inequalities, Mondié and Kharitonov (2005), Fridman
to skip the use of time derivative measuring devices, digital and Shaked (2003) but the results are usually poor because the
such as backward difference or analog such as tachometers, solution is very conservative, and when it exists, there is no
because of noise amplification. insight on which roots are the dominant ones, hence the fine
tuning of the response is not possible.
An alternative is the use of the delayed output in the control
loop. The stabilizing effect of delays in feedback has been The aim of this contribution is to provide a detailed frequency
studied in depth in the case of second order systems Abdallah domain analysis of the s -stabilizability the proportional-
et al. (1993) as well as more general classes of oscillatory delayed control law (3) in closed loop with system (1). The
systems Moreno and Michiels (2005), Suh and Bien (1979), main definitions and tools for the analysis are introduced in
Karafyllis (2008). An intuitive idea is that the substitution of section 2. In section 3, the s -stabilizability boundaries and
the approximation of the derivative regions are determined and a complete analysis of the dominant
roots for a given s is performed in section 4. Furthermore,
q (t) − q (t − h)
q˙ (t) ≈ for small h, the maximum exponential decay strategy is characterized by
h straightforward formulae in section 5. In section 6, an experi-
into (2) gives a control law of the form mental setup of a DC-servomotor is employed to illustrate the
u(t) = −k pq (t) + kr q (t − h) (3) results. The contribution ends with some concluding remarks.
where h is the delay, and
2. PRELIMINARY RESULTS
k p = k1 + k2 /h, and kr = k2 /h. (4)
Notice that h is now a design parameter that can be used for In this section, we introduce important properties and defini-
tuning. tions.
Lemma 1. Bellman and Cooke (1963), Datko (1978), For any whose roots are given by (10). Clearly, real roots of ps corre-
s ∈ R, there are only a finite number of roots with real part spond to the square root of positive l 1,2 .
greater than s . Consider the quasi-polinomial (5) and let Proposition 6. The parametric equations for root crossings of
a 0 = sup Re{s j } : p(s j , k p , kr , h) = 0, s j ∈ C . (6) the imaginary axis at purely imaginary pairs are

j=1,...,¥
1 −w 2 + (s 2 + n 2 − 2d ns + bk p)
 
Then for any s ≥ a 0 , there exists a constant L > 0 such that the h(w ) = cot−1
solution of (1)-(3) satisfies the inequality w −2(d n − s )w
p
|q (t)| ≤ Les t kj kh, (7) + n , n = 0, 1 . . . , (12)
w
where j is the initial condition and kj kh = maxt ∈[−h,0] kj (t )k. 2w (d n − s )
kr (w , h) = − s h . (13)
Definition 2. We say that the triplet (k p , kr , h) s -stabilizes sys- be sin(hw )
tem (1)-(3) if
a 0 ≤ − s , s ∈ R+ , Proof. Substituting s = jw and e− jw h = cos(w h) − j sin(w h)
where a 0 is given in (6). into (8) yields
Re{ps ( jw , k p , kr , h)} = −w 2 + (s 2 + n 2 − 2d ns + bk p)
As it is well known, the change of variable s −→ (s − s ) in the
frequency domain or y(t) = es t x(t) in the time domain reduces − bkr ehs cos(w h) = 0, (14)
the analysis of the s -stability to the stability of the transformed Im{ps ( jw , k p , kr , h)} = 2(d n − s )w + bkr e hs
sin(w h) = 0.
quasipolynomial (15)
ps (s, k p , kr , h) = s2 + 2(d n − s )s + (s 2 + n 2 − 2d ns + bk p) and the result follows by simple algebraic manipulations.
− bkr ehs e−hs . (8)
Remark 3. The decay of the autonomous system (u ≡ 0) is d n . 3. REGIONS OF s -STABILIZABILITY
The analysis presented in this paper is restricted to the case of
closed loop exponential decay s > d n which corresponds to an We now present graphical and analytical results concerning the
improved system response. s -stabilizability properties of system (1).
The equation (9) and the parametric equations (12, 13) define
The quasipolynomial (8) defines completely the s −stabilizability all boundaries corresponding to root crossings of the imaginary
and roots dominance of system (1). As the roots behavior in the axis. The ones that delimit s −stabilizability regions in the
complex plane is continuous with respect to continuous changes (k p , kr , h) tridimensional space are indeed the ones of interest. It
of the parameters Neimark (1949), loss of s -stabilizability is possible to perform a formal analysis of the stability regions
necessarily occur when the quasipolynomial has roots either at of the quasipolynomial (8) which is close to the widely studied
s = 0 or at a pair of pure imaginary roots at s = ± jw . Next, we form q(s) + p(s)e−sh where q(s) and p(s) are polynomials such
characterize the root crossings of the imaginary axis which are that deg(p(s)) < deg(q(s)) (see Cooke and Van Den Driessche
candidate loci for the stability/instability boundaries. (1986), Michiels and Niculescu (2007), Cooke and Grossman
Proposition 4. Roots crossing loci of the quasipolynomial (8) (1982), Abdallah et al. (1993) among others). Although the
of the imaginary axis at s = 0 satisfy quasipolynomial (8) is such that p(s) also depends on h it is
s 2 + n 2 − 2nd s + bk p clear that the methods can be extended. Here, given that the
kr = . (9) case is quite simple we can characterize the s −stabilizability
behs
using D-Partition techniques, Neimark (1949), by testing a
Proof. The result follows straightforwardly by substituting s = 0 single point of each region by using, for example, the argument
into (8): principle.
ps (0, k p , kr , h) = s 2 + n 2 − 2nd s + bk p − bkr ehs = 0 The typical delay dependent stability/instability phenomenon
of the s −stabilizability regions depicted in the bi-dimensional
Proposition 5. Roots crossing loci of the imaginary space (kr , h) for a fixed gain k p are sketched in Figure 1. The
p axis of the
quasipolynomial (8) at s = ± jw occur at w = l 1,2 for real parameter values are b = 1, n = 2.3968, d = 0.00055 and
and positive l 1,2 of the form k p = 5. It appears that s −stabilizability can be achieved not
only in the first region but also in regions corresponding to
l 1,2 = n 2 1 − d 2 + bk p − (s − d n )2


larger delays. However, the achievable decay s decreases as h
grows and the stability region narrows significantly, resulting in
q
∓ (bkr ehs )2 − 4 (s − d n )2 (n 2 (1 − d 2) + bk p). (10)
fragile controllers. The fact that he s −stabilizability regions in
Proof. Setting the space (kr , h) grows as the gain k p does is shown in Figure 3.
2
ps ( jw , k p , kr , h) = −w + 2 j(d n − s )w
The main region of interest for our analysis is shown in Figure 3
+ (s + n 2 − 2d ns + bk p) − bkr ehs e− jw h = 0
2
where the level curves corresponding to different s are depicted
and taking modulus yields in the bi-dimensional space (kr , h).
w 4 −2w 2 m + bk p − 2 (s − d n )2 +(m + bk p)2 −b2 kr2 e2hs = 0
 
One observe that for a given s −stabilizability specification, the
(11) same exponential decay is achieved at all points of the bound-
where m = s 2 − 2s d n + n 2 . By introducing the change of aries. The upper boundary corresponds to loci of p(s, k p , kr , h)
variable l = w 2 , we get a quadratic polynomial in l with at least one dominant real root at −s , while the lower
boundary corresponds to pairs of complex conjugate roots at
l 2 − 2l m + bk p − 2 (s − d n )2 + (m + bk p)2 − b2 kr2 e2hs = 0
 
−s ± jw . We also observe that the level curves collapse into a
 4.1 Key loci on the boundary

Locus 1: Double root at −s

The crossing of the branch corresponding to the root w 1 with
the boundary (s 2 + n 2 − 2d ns + bk p ) − bkr ehs = 0 occurs
when w 1 −→ 0. In this case there is a double root at s hence
d p (s,k ,k ,h) 
ps (0, k p , kr , h) = 0 and s dsp r  , therefore
s=0
(s 2
+ n 2 − 2d ns + bk p) = bkr ehs (16)
and
2(d n − s )
kr = − . (17)
hbehs
Combining (16) and (17), one obtains that
−2(d n − s )
Fig. 1. s -stable region of system (1)-(3)  s ∈ (0, 0.1),  s ∈ h= 2 . (18)
(s + n 2 − 2d ns + bk p)
(0.1, 0.5), s ∈ (0.5, 1),  s ∈ (1, 2),  s ∈ (2, 3.2).
Thus, given s > 0 and k p > 0 the parameters h and kr follow
from (18) and (17) respectively.
Locus 2: locus where w 1 = w 2
This locus is when the imaginary part of the active root of the
quasipolynomial turns from w 1 to w 2 . One can see from (10)
that l 1 ≤ l 2 . Clearly, the maximum possible value for l 1 occurs
when
q
(bkr ehs ) = 4 (s − d n )2 (n 2 (1 − d 2 ) + bk p).
In this case,
q
w 1 = w 2 = n 2 (1 − d 2) + bk p − (s − d n )2
and the corresponding delay h and gain kr can be obtained from
the parametric equations (12, 13).
Locus 3: Pair of complex conjugate roots at −s + jw 2 and
Fig. 2. s -stable region of system (1)-(3) for s = 3.2 and k p ∈ a real root at −s
[0, 3]. When the imaginary part of the active root is w 2 , then locus 3
occurs when
(s 2 + n 2 − 2d ns + bk p) − bkr ehs = 0
Then it follows that
l 2 = n 2 1 − d 2 + bk p − (s − d n )2


q
+ (bkr ehs )2 − 4 (s − d n )2 (n 2 (1 − d 2) + bk p), (19)
= 2 −s 2 + 2s d n − 2d 2 n 2 + n 2 + bk p (20)

and q
w 2 = 2 (−s 2 + 2s d n − 2d 2 n 2 + n 2 + bk p) (21)
The corresponding h and kr . follow from the parametric equa-
tions (12, 13).
Locus 4: Simple root at -s and a double real root at -a
As one can see in Figure 4, this locus has a special role: it
Fig. 3. s -stable region of system (1)-(3)  s ∈ (0, 0.5),  s ∈ corresponds to the case where ps (s, k p , kr , h) has a simple root
(0.5, 1), s ∈ (1, 1.5),  s ∈ (1.5, 2),  s ∈ (2, 3.2). at -s hence ps (0, k p , kr , h) = 0, or
single point that correspond to the unique maximum achievable (s 2
+ n 2 − 2d ns + bk p) = bkr ehs (22)
decay s ∗ that will be characterized in section 5.
and a double real  root at some −a hence pa (0, k p , kr , h) = 0
4. DOMINANT ROOTS ANALYSIS d pa (s,k p ,kr ,h) 
and ds  , therefore
s=0
As the system response is determined by the location of the
dominant roots, we provide next the analytical characterization (a 2
+ n 2 − 2d na + bk p) = bkr eha (23)
of some key loci of the boundary for which the exponential and
decay −s is achieved. 2(d n − a ) + hbkreha = 0 (24)
Combining (22), (23) and (24) it follows that a satisfies 2(d n − s ∗ )
h∗ = − , (29)
2 2
2(d n −a )(a −s )
− 2 2 n 2 + (s ∗ )2 − 2d ns ∗ + bk
p
(s + n − 2d ns + bk p)e n +a −2d na +bk p
2(d n − s ∗ )
= (a 2
+ n 2 − 2d na + bk p). (25) kr∗ = − ∗ ∗ . (30)
bh∗ es h
This implicit expression corresponds to the crossing of an Proof. The larger exponential decay s ∗ occurs into the first
exponential and of a quadratic in the variable a . It has a solution stability region. This region collapses into a single point when
at s and at some a 6= s > 0 that can be computed by using s increases. In this case, there is a triple root at −s , hence
numerical methods. ps (s, k p , kr , 0) = 0 and its first and second derivative with
Thus, given s > 0 and k p > 0 we obtain a from (25) and the respect to s are also null, namely,
parameters h and kr follow from (s 2
+ n 2 − 2d ns + bk p) = bkr ehs , (31)
−2(d n − a ) 2(d n − s ) + hbkre hs
=0 (32)
h= 2 , (26)
(a + n 2 − 2d na + bk p) and
2(d n − a ) h2 bkr ehs = 2. (33)
kr = − . (27)
hbeha It follows from (31) and (32) that
Remark 7. The characterization of the above loci combined −2(d n − s )
with the parametric equations allow to sketch straightforwardly h= 2 . (34)
(s + n 2 − 2d ns + bk p)
the first sigma-stability region as follows:
Moreover, (31) and (33) imply
Upper boundary: For the selected k p and s , sketch in the (h, kr )
plane 2
h2 = 2 . (35)
s 2 + n 2 − 2nd s + bk p (s + n 2 − 2d ns + bk p)
kr = , Substituting (34) into (35) yields
behs i
d n −s ) (s − d n )2 = (1 − d 2 )n 2 + bk p
h
where h ∈ (s 2 +−2(
n 2 −2d ns +bk p )
, h(w 2 ) . Here h(.) and w 2 are
and (28) is obtained. Finally, (29) and (30) follow from (31) and
defined in (13) and (21), respectively. (32).
Lower boundary: For the selected k p and s , sketch in the (h, kr )
plane 6. PR TUNING OF AN EXPERIMENTAL DC
2w (d n − s ) SERVOMOTOR
kr (w ) = − s h
be sin(h(w )w )
In this section, we illustrate with the experimental setup of a
with DC servomotor the results of the previous theoretical analysis.
h(w ) =
The servomechanism employed for the experiments consists
−w 2 + (s 2 + n 2 − 2d ns + bk p )
1 of a DC brushed motor controlled through a Copley Controls
  
 cot−1 , w ∈ [e , w e ];

w −2(d n − s )w

= 2 + (s 2 + n 2 − 2d ns
power amplifier, model 413, configured in current mode. A BEI
1 −w + bk p ) p
 
 cot−1

 + , w ∈ [w e , w 2 ]. optical encoder directly coupled to the motor shaft gives angu-
w −2(d n − s )w w
lar position measurements. Resolution of the optical encoder
for w ∈ [e , w 2 ), e > 0 and w e = min{w 2 , s 2
+ n 2 − 2d ns + is 2500 pulses per revolution. A Q8 card from Quanser Con-
bk p } sulting endowed with inputs for optical encoders performs data
acquisition. The electronics associated to these inputs multiply
For a full picture of the dominant roots behavior, we present by 4 the encoder resolution. In this way, one motor turn cor-
in Figure 4 the evolution of the location of the dominant roots responds to 10 000-encoder pulses. A factor of 10, 000 scales
as one travels on the s -stabilizability boundary s = 0.5 of down the angular position measurements. The card also has 12
Figure 3 when the system parameters are b = 1, n = 2.3968 bits digital-to-analog converters with an output voltage range of
, d = 0.00055 and the proportional gain is k p = 5. ±10 volts. All the programming is done using the Mathworks
MatLab Simulink environment under the Wincon real-time pro-
gram. The software runs on a Personal Computer using an Intel
5. MAXIMUM EXPONENTIAL DECAY STRATEGY Core 2 quad processor, and the Q8 board is allocated in a PCI
slot inside this computer.
Next, we characterize the maximum achievable exponential
decay s ∗ for a fixed k p ≥ 0 and we compute the corresponding 6.1 Servomechanism model
retarded gain kr∗ and delay h∗ .
Lemma 8. Let the proportional gain of the controller k p ≥ 0 be We consider the following second order model for the ser-
given. Then, the maximum exponential decay s ∗ of the closed vomechanism
loop system (1-3) that can be achieved by designing the pair J q̈(t) + f q̇(t) = t (t) = ku(t) (36)
(kr , h) is where q is the angular position, t (t) the input torque, u the
control input voltage, J the motor and load inertia, f the
q
s ∗ = d n + n 2 (1 − d 2) + bk p. (28)
viscous friction and k the amplifier gain. A motor, a power
Moreover, the values of the delay gain kr∗ and of the delay h∗ amplifier and a position sensor compose the servomechanism.
that s −stabilize the system (1-3) with the exponential decay The power amplifier is set to current mode; therefore, the
s ∗ are electromagnetic torque is proportional to the input voltage
Fig. 4. Root locus of the dominant roots of the quasipolynomial (5).

low from equations (29) and (30). The system response and the
control signal are shown in Figure 6 and 7, respectively.

Fig. 5. DC Servomotor experimental setup.

applied to the amplifier. This approach works for DC and AC
brushless servomotors.
Observe that equation (36) can be written as
Fig. 6. Angular position q(t) for (h∗ , kr∗ , k p ) =
q̈(t) = −aq̇(t) + bu(t) (37) (0.0449, 11.6526, 16).
where a = f /J, b = k/J are positive parameters. In the above
described platform, the estimated parameter values are a = Next, we display in Figure 8 and 9 the system response and con-
0.45, b = 31. Notice that a preliminary proportional feedback trol law for the loci 1,2,3 and 4. The corresponding parameters
leads to the form (1) studied in the previous sections. h and kr are listed on Table 1. In all cases, the exponential decay
is s = 4.5 and the proportional gain is k p = 16. The location of
6.2 Experimental results the roots on the s boundary determines the shape of the system
response.
We now illustrate how the choice of the gains kr and k p and
Remark 9. It appears that the above theoretical analysis leads
of the delay h in the control law u(t) = kr q (t − h) − k p q (t)
to a straightforward tuning formula for the proportional delay
determines the system response. A proportional gain k p = 16
such that the system operates in the region were the control control of second order systems which are the first election
model for a wide array of physical processes. The maximum
voltage does not go to saturation is selected.
achievable decay for a given proportional gain is given by (28)
The maximum achievable exponential closed loop decay s ∗ = and the corresponding parameters are readily computed from
22.4949 is obtained straightforwardly from the formula (28). (29) and (30). This seems to be an advantage compared with a
The corresponding parameters (h∗ , kr∗ ) = (0.0449, 11.6526) fol- proportional derivative controller that requires the tuning of a
Fig. 7. Control signal for (h∗ , kr∗ , k p ) = (0.0449, 11.6526, 16). Fig. 9. Input forces with control (3) and parameter (h, kr , k p , s )
given in Table 1.
Locus h kr
1 0.01663 15.39 ACKNOWLEDGEMENTS
2 0.0683 4.574
3 0.0839 11.37 This work was supported by CONACyT, Project 61076 and
4 0.03931 13.92 Scholar 206997. The authors thank José de Jesús Meza Serrano
5 0.0449 11.653
for his technical support.
Table 1. Parameters (h, kr ) of control (3) for s =
4.5 and k p = 16 given. REFERENCES
Abdallah, C., Dorato, P., Benitez-Read, J., and R., B. (1993).
Delayed positive feedback can stabilize oscillatory system.
Proceedings of the American Control Conference, 3106-
3107.
Bellman, R. and Cooke, K. (1963). Differential-Difference
Equations. Academic Press, Inc, New York, USA.
Cooke, K. and Grossman, Z. (1982). Discrete delay,
dristributed delay and stability switches. Journal of math-
ematical analysis and applications, 86:592-627.
Cooke, K. and Van Den Driessche, P. (1986). On zeroes of some
transcendental equations. Funkcialaj Ekvacioj, 29:77-90.
Datko, R. (1978). A procedure for determination of the ex-
ponential stability of certain differential difference equation.
Quart. Appl. Math., 1:279-292.
Fridman, E. and Shaked, U. (2003). Delay-dependen stability
and H-infinity control: constant and time-varying delay. Int.
J. Control, 76:48-60.
Karafyllis, I. (2008). Robust global stabilization by means of
Fig. 8. Angular position q(t) for (h, kr , k p , s ) given en Table 1. discrete-delay output feedback. Systems and Control Letters,
987-995.
filter or of an observer along with the tuning of the control law.
Michiels, W. and Niculescu, S. (2007). Stability and Stabi-
The delay physical implementation can be done with memory
lization of Time-Delay Systems. Society for Industrial and
elements.
Applied Mathematics.
Mondié, S. and Kharitonov, V. (2005). Exponential estimates
7. CONCLUDING REMARKS for time delay systems, a LMI approach. IEEE Trans. on
Autom. Contr., 50:268-273.
In this contribution a graphical and analytical study of the Moreno, V.K.S.N.J. and Michiels, W. (2005). Static output
s −stabilization of a second order system through proportional feedback stabilization: Necessary conditions for multiple
delayed control law is performed. A comprehensive charac- delay controllers. Transactions on Automatic Control, 82-86.
terization of the regions and boundaries of s −stabilization is Neimark, J. (1949). D-subdivisions and spaces of quasi-
performed and a detailed analysis of the dominant roots of the polynomials. Prikl. Mat. Meh., 13:349-380.
closed loop system is presented. This leads to the characteriza- Suh, I. and Bien, Z. (1979). Proportional minus delay con-
tion of the maximum achievable exponential decay strategy for troller. IEEE Transactions on Automatic Control, AC-
which simple formulae are available. An experimental setup of 24:370-372.
a DC servomotor illustrates the results and the simplicity of the
implementation.