ACCOUNTING OF DC-DC POWER CONVERTER DYNAMICS IN DC

MOTOR VELOCITY ADAPTIVE CONTROL

H. El Fadil
Département LA2I, Ecole Mohammadia d’Ingénieurs, Agdal, Rabat, 10000
Maroc (Tel: +212-65-27-68-45; e-mail: elfadilhassan@yahoo.fr).

F. Giri
GREYC, Université de Caen, Bd Marechal Juin, B.P 8156, 14032, Caen
France (e-mail: giri@greyc.ensicaen.fr)

Abstract: This paper deals with the problem of velocity dynamics non taken in account in the design of the
control of a DC motor taking account the dynamics of the regulator can generate a reduction of the phase margin
DC-DC power converter, supposed here of Buck type. We
will develop for the global system, constituted by combined and in turn produces the instability of the system.
DC-motor and Buck-converter, a model of fourth order. On In this paper, we propose an adaptive control of DC
the basis of this model a regulator is designed using the motor velocity tacking account the dynamics of DC-
backstepping technique. The control purpose is, on one DC power converter of the Buck type. The converter
hand, asymptotic stability of the closed-loop system and, on
the other hand, perfect tracking of the reference signal (the input (duty ratio) is designed so that a smooth
machine speed). Both non adaptive and adaptive versions trajectory is followed by the motor angular velocity.
are designed and shown to yield quite interesting The proposed controller, based on the backstepping
performances. A theoretical analysis shows that both approach, is shown to achieve a good stability and
controllers meet their objectives. These results are
confirmed by experimental results which, besides, show that
perfect tracking objectives under constant but
the adaptive version deals better with load torque changes. unknown load torque.
The paper is organized as follows: in section 2 the
Key words: DC-motor, Buck converter, Backstepping, combined DC-motor and Buck converter model is
Lyapunov stability.
described. Section 3 and Section 4 are devoted to the
controller theoretical design; the controller stability
1. Introduction and tracking performance are illustrated in section 5. A
conclusion and reference list end the paper.
DC machines are extensively used in many
industrial applications such as servo control and
2. The combined DC Motor-Buck Converter
tractions tasks due to their effectiveness, robustness
Model
and the traditional relative ease in the devising of
appropriate feedback control schemes [4], specially Consider a permanent magnet motor with its armature
those of the PI and PID types ([3], [5], [8]). circuit loaded to a DC-DC power converter of Buck type as
Therefore, the problem of velocity control of DC- shown in Figure 1. Such a configuration constitutes a single
quadrant angular velocity control configuration for a DC
motors has been extensively dealt in the specialized
drive. The mathematical model of the composite system is
literature during the last years ([1], [2], [6],[7]). given by:
However, in the most works, the dynamics of DC
power converter has generally been ignored when
analyzing the resulting closed-loop systems. Indeed,
the converter is generally considered transparent as
being assimilated to the proportional gain. Moreover,
this simplification can be tolerated for the high
powers, but not for the low and the medium powers.
Such a simplification can only limit performances of
the regulator and can even compromise the global Fig. 1: The combined DC-motor- “Buck” converter model
stability of the system in closed loop. Indeed, a

1
 di 3. Non Adaptive Controller Design
L = −v + Eu
dt The aim is to directly enforce the average angular
dv
C = i − ia velocity x1 to track a given reference signal ω r (t ) ,
dt (1)
di when the model parameters are perfectly known. The
Lm a = v − Rm ia − K eω
dt reference signal and its four first derivatives are
dω assumed to be known, bounded, and piecewise
J = − fω + K m ia − TL
dt continuous. Following closely the backstepping design
[10], the controller is designed in four steps, since the
Where i is the converter input current, v is the control input appear after four derivatives of x1 . The
converter output voltage, ia is the dc motor armature
control law is summarized in Table 1.
circuit current and ω is the angular velocity of the
motor shaft, which may be subject to a constant but
TABLE 1: NON ADAPTIVE CONTROLLER
unknown load torque T L . The control input is
represented by the variable u , which represents a
Error variables:
switch position tacking value on the discrete set {0,
z1 = x1 − ωr (4)
1}.
The model (1) is useful to build-up an accurate z2 = Kx2 / J − α1 (5)
simulator for the combined DC-motor and buck z3 = Kx3 / (JLm ) − α 2 (6)
converter. However, it cannot be based upon to design z4 = Kx4 / (JLmC ) − α 3 (7)
a continuous control law as it involves a binary control Stabilizing functions
input, namely u . To overcome this difficulty, it is α1 = fx1 / J + TL / J + ω& r − c1z1 (8)
usually resorted to the averaging process over cutting
intervals [11]. This process is shown to give rise to (
α 2 = K / (JLm ) − ( f / J ) x1
2
) 2

average versions (of the above model) involving as a (

+ KRm / (JLm ) + fK / J )x 2
2
control input the mean value of u which is nothing − fTL / J 2
+ c12 z1 − c1 z 2 + ω&&r − z1 + c2 z2 (9)
other than the duty ratio µ . The average model turns
α 3 = −(b1 f / J + b2 K / Lm )x1
out to be the following:
+ (b1 K / J − b2 Rm / Lm + K / (JLm C ))x2
x&1 = −( f / J )x1 + (K / J )x2 − TL / J + b2 x3 / Lm − b1TL / J + − c13 + 2c1 + c2 z1 ( )
x&2 = −(K / Lm )x1 − (Rm / Lm )x2 + x3 / Lm + ( c12 + c 22 )
+ c1 c 2 − 2 z 2 − (c1 + c 2 + c 3 )z 3 + ω
&&&r (10)
(2) where
x&3 = x4 / C − x2 / C
b1 = K 2 / (JLm ) − ( f / J )2 (11)
x&4 = − x3 / L + µE / L b2 = KRm / (JLm ) + fK / J 2
(12)
Control law:
where we denote by x1 , x2 , x3 and x4 the average
µ = (JLm LC / (KE )){(b3 f / J − b4 K / Lm )x1
values of the angular velocity ω , the dc motor
− (b3 K / J + b4 Rm / Lm + b2 / (Lm C ))x2
armature circuit current ia , the converter output
+ (K / (JLm LC ) + b4 / Lm )x3 + b2 x4 / (Lm C ) + b3TL / J
voltage v and the converter input current i ,
respectively. ( )
+ c14 − 3c12 − c 22 − 2c1 c 2 + 2 z1
Furthermore, the switching policy is generated + (− c − c − c c − c c + 3c + 4c + c )z
3
1
3
2
2
1 2
2
2 1 1 2 3 2

+ (c + c + c + c c + c c + c c − 3)z
according to so-called Pulse Width Modulation 2 2 2
1 2 3 1 2 1 3 2 3 3
(PWM) principle described as follows:
− (c1 + c 2 + c 3 )z 4 + ω r( 4 ) + c 4 z 4 }
⎧⎪1 pour tk ≤ t ≤ tk + µ (tk )T (13)
u (t ) = ⎨ (3) where
⎪⎩0 pour tk + µ (tk )T ≤ t ≤ tk + T b3 = b1 f / J + b2 K / Lm (14)
b4 = b1 K / J − b2 Rm / Lm + K / (JLm C ) (15)
where tk represents a sampling instant and the
parameter T > 0 is the fixed sampling period

2
Proposition 1 particular, the tracking error z1 is bounded but not
Consider the closed loop system consisting of the vanishing. Therefore adaptive versions of the above
subsystem (2) and the control law (13). If the angular controllers turn out to be interesting alternatives.
velocity reference ω r and its four first derivatives are To cope with such a model uncertainty the new
known and bounded, then the closed loop undergoes, controller will be given a learning capacity. More
in the ( z1 , z 2 , z 3 , z 4 )-coordinates, the following specifically, the controller to be designed should
equation: involve an on-line estimation of the unknown
parameter
⎡ z& ⎤ ⎡− c 1 0 0 ⎤ ⎡ z1 ⎤
⎢ 1⎥ ⎢ 1 ⎥⎢ ⎥ θ = TL / J (19)
⎢ z& 2 ⎥ = ⎢ − 1 − c2 1 0 ⎥⎢z2 ⎥
(16)
⎢ z& ⎥ ⎢ 0 −1 − c3 1 ⎥⎢z3 ⎥
⎢ 3⎥ ⎢ ⎥⎢ ⎥ The obtained estimate is denoted θ̂ , it follows that
⎣⎢ z& 4 ⎦⎥ ⎣⎢ 0 0 −1 − c 4 ⎦⎥ ⎣⎢ z 4 ⎦⎥
~
θ = θˆ + θ (20)
where the design parameters ( c1 , c 2 , c3 , c 4 ) are
positives and freely chosen. Consequently, the error ~
where θ is the estimation error.
system (16) is globally asymptotically stable. It Following the tuning functions backstepping design
follows that: [10], then the control law and the parameter update
i) All signals in closed loop are bounded, law, designed in four steps are summarized in table 2.
ii) The tracking error z1 = x1 − ω r vanishes. □
TABLE 2: TUNING FUNCTION DESIGN
Proof. Let us consider the following Lyapunov Error variables
function candidate: z1 = x1 − ω r (21)
4
z 2 = Kx 2 / J − α1 (22)
V= ∑ 0.5z
i =1
2
i (17)
z3 = Kx3 / (JLm ) − α 2 (23)
z 4 = Kx 4 / (JLm C ) − α 3 (24)
its time derivative, using Table 1, is given by Stabilizing functions
α1 = −c1 z1 + fx1 / J − w1θˆ + ω& r (25)
4
α 2 = − z1 − c 2 z 2 − ψ 2 − w1γτ 2
V& = − ∑c z
i =1
2
i i ≤0 (18)
α 3 = − z 2 − c3 z 3 − ψ 3 − λ3γτ 3 + υ 3
(26)
(27)
Tuning functions
This shows that the ( z1 , z 2 , z 3 , z 4 )-system is globally τ 1 = w1 z1 (28)
asymptotically stable. From Table 1 we see that, since τ i = τ i −1 + wi z i i = 2,...,4 (29)
the α i ’s are smooth function, the state variables x i can Regressor constants
be expressed as a smooth function of z1 ,L , z i . Hence, w1 = −1 (30)
the state vector x(t ) is globally asymptotically stable. w2 = c1w1 + f / J (31)
Furthermore, from (17) and (18), we can see that (
w3 = b1 − w1 c12 − 1 − γw12 + w2 a 3) (32)
V → 0 as t → ∞ , hence, all variable errors z i , i=1…4, w4 = b3 + a 4 w1 + a 5 w2 + a 6 w3 (33)
vanish asymptotically. This ends the proof of Adaptive control law
Proposition 1. JLm LC
µ= {− z 3 − c 4 z 4 −ψ 4 − λ 4 γτ 4 + υ 4 } (34)
KE
4. Adaptive Controller Version Parameter update law
&
Controllers of section 3 guarantee perform well only θˆ = γWz (35)
when the model (2) is perfectly known. This
where z = [z1 , z 2 , z 3 , z 4 ]
T
and W = [w1 , w2 , w3 , w4 ]
particularly means that the load torque T L is constant
and time-invariant. When this is not the case, the
controllers may still provide an acceptable behavior, in

3
The functions and constants introduced in the above and its four first derivatives are known and bounded,
Table 2 have the following expressions then the closed loop undergoes, in the following
equations:
ψ 2 = −b1 x1 − b2 x 2 + fθˆ / J − c12 z1 + c1 z 2 − ω&&r (36)
~
z& = Az z + W T θ (56)
ψ 3 = b3 x1 + b4 x2 − b2 x3 / Lm &
θˆ = γWz (57)
+ b1θˆ + a1 z1 + a 2 z 2 + a 3 z 3 − ω
&&&r (37)
where Az is a skew symmetric matrix defined as
ψ 4 = −(b3 f / J + b4 K / Lm )x1
follows
+ (b3 K / J − b4 Rm / Lm + b2 / (Lm C ))x2
⎡− c1 1 0 0 ⎤
+ (b4 / Lm − K / (JLm LC ))x3 − b2 x4 / (Lm C ) − b3θ̂ ⎢ −1 − c2 1 + σ 23 σ 24 ⎥⎥
Az = ⎢ (58)
[
+ − c1 a 4 − (
a 5 1 + γw12 ) − a λ γw ] z
6 3 1 1
⎢ 0
⎢
− 1 − σ 23 − c3 1 + σ 34 ⎥
⎥
⎣ 0 − σ 24 − 1 − σ 34 − c4 ⎦
+ [a 4 − a 5 (c 2 + γw1 w2 ) − a 6 (1 + σ 23 + λ3γw2 )] z 2
Consequently, the error system (56) is globally
+ [a 5 (1 + σ 23 − γw1 w3 ) − a 6 (c 3 + λ3γw3 )] z 3
asymptotically stable. It follows that:
+ a 6 z 4 − ω r(4 ) (38) i) All signals in closed loop are bounded,
ii) The tracking error z1 = x1 − ω r vanishes.
λ3 = f / J + w1a3 (39) iii) The estimate parameter θ̂ converges toward
λ 4 = b1 + a 5 w1 + a 6 λ3 (40) uncertain and constant parameter θ
b1 = K / (JLm ) − ( f / J )
2 2
(41) □
b2 = KRm / (JLm ) + fK / J 2
(42) Proof. Consider the following Lyapunov function
candidate:
b3 = b1 f / J + b2 K / Lm (43)
b4 = − K / (JLm C ) − b1 K / J + b2 Rm / Lm (44) ( ~
V = 0. 5 z T z + θ 2 / γ ) (59)
( )
a1 = c1 c12 − 1 − γw12 − a 3 1 + γw12 ( ) (45)
a2 = −[c 2
1 − 1 − γw12 + a 3 (c 2 + γw1 w2 ) ] (46) its time derivative, using Table 2, is given by
a 3 = c1 + c 2 + γw1 w2 (47) 4
a 4 = a1 + λ3γw1 (48) V& = − ∑c z 2
i i ≤0 (60)
a 5 = 1 + a 2 + λ3γw2 + σ 23 (49) i =1

a 6 = c3 + a 3 + λ3γw3 (50)
i) From (59) and (60) we can see that the
υ 3 = −σ 23 z 2
σ 23 = w1 γw3
(51)
(52)
~
( )
equilibrium z ,θ = 0 is globally asymptotically
stable and in turn the state vector x(t ) is globally
υ 4 = −σ 24 z 2 − σ 34 z 3 (53)
asymptotically stable.
σ 24 = γw1 w4 (54)
ii) From LaSalle’s Invariance Theorem [10], it
σ 34 = λ 3γw4 (55) ~
( )
further follows that the state z ,θ converges to
the largest invariant set M of (56)-(57) contained
where c1 , c 2 , c3 , c 4 and γ are positive design
parameters freely chosen.
{( )
~
}
in E = z ,θ ∈ IR 5 / z = 0 , that is, in the set where
The main result of this subsection is summarized in the V& = 0 . This means, in particular, that z (t ) → 0 as
following proposition t→∞.
iii) On the invariant set M , we have z ≡ 0 and z& ≡ 0 .
Proposition 2 Setting z = 0 and z& = 0 in (56) and (57) we obtain
Consider the closed loop system (2) subject to ~
θ& = 0 and W T θ = 0 . As the regressor vector is
uncertain load torque T L and the controller composed composed of constant parameters it follows that
~
of the adaptive control law (34) and the parameter we get θ = 0 on M , which implies that
update law (35). If the angular velocity reference ω r

4
 M = {(0,0)} and , in particular, we have θˆ → θ as
t →∞.
We thus established the proof of Proposition 2
5. Experimental Results
The backstepping no adaptive and adaptive
controllers shown in Table 1 and Table 2 has been
applied to the combined DC-Motor-“Buck” Converter
according to the experimental setting of figure 2,
where x denotes the state vector.
controller design parameters have the following
values: c1 = 600 ; c 2 = 700 ; c 3 = 400; c 4 = 500 ;
γ = 1x10 −11 . The corresponding performances are
illustrated by Fig.4. This shows that, despite the load
torque uncertainty, the controller behavior is quite
satisfactory. It is worth noting that such a good
behavior is preserved when facing different variations
of the load torque. As can be expected, the estimation
parameter θ̂ converges asymptotically to the uncertain
value θ as depicted in Fig.4.
Fig.5 shows that the motor velocity tracks perfectly its
varying reference. The desired trajectory ω r is a signal
switching between 40 and 50rad/s.
Rotor speed (rd.s-1) Inductor current i (A)
6
60
4
40

20 2

0 0
0 0.1 0.2 0.3 0 0.1 0.2 0.3
Fig. 2: Experimental bench for DC-motor velocity control Load torque TL (N.m) Duty ratio µ

0.08
1
Table 3 lists the numerical values for the parameters 0.06 0.8

the combined system studied in this paper. 0.04

0.6
0.4
0.02
0.2

TABLE 3: PARAMETERS OF THE COMBINED DC- 0

0 0.1 0.2 0.3
0
0 0.1 0.2 0.3
MOTOR-BUCK CONVERTER time (s) time (s)

L = 20x10-3[H] C = 400x10-6[F] Fig.3: Backstepping non adaptive controller performances

K e = 0.046[V.s/rad] K m = 0.046[N-m/A]
Rotor speed (rd.s-1) teta
J = 7.06x10-5[kgxm2] f = 8.42x10-4 60 1000
[Nm.s/rad] 40

L m =2.63x10-3 [H] R m = 2.0[Ω] 20

500

E = 12[V] T L = 0.05 [N-m] 0 0

0 0.1 0.2 0.3 0 0.1 0.2 0.3

Duty ratio µ estimate of teta

A. Non adaptive controller 1

Fig.3. illustrates the behavior of the controller (13) 0.8 1000

in presence of a constant reference ω r = 60rad/s and a 0.6

0.4 500

variable load torque. The value of the load torque T L 0.2

0 0
was step changed for a time period of 0.1 seconds to 0 0.1 0.2
time (s)
0.3 0 0.1 0.2
time (s)
0.3

50% value of its original one. The relevant design Fig.4: Backstepping adaptive controller performances
parameters have the following values: c1 = 1x103;
c 2 = 1.5x103; c 3 = 400; c 4 = 500. It is seen that motor
Rotor speed (rd.s-1) Inductor current i (A)

50 3
velocity perfectly tracks its reference only when the 40
30 2
load torque is equal to its nominal value. Besides, 20
1
when the load torque deviate around its nominal value,
10
0
0
it can be seen that the motor velocity deviate, in turn, 0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8

Speed reference Duty ratio µ

around its reference.
40 0.8

B. Adaptive controller version 20 0.6

We consider now the adaptive controller (34)-(35). 0 0.4

0.2
The reference signal value and the load torque
-20
0
0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8
variation are the same as in simulation A. The adaptive time (s) time (s)

5
 Fig.5: Tracking behavior of the adaptive controller illustrates that the adaptive controller provide excellent
C. Importance of converter dynamics asymptotic stability, a perfect tracking behavior, and a
We are going to illustrate here the importance to good compensation of load torque changes.
take in account, when designing the controller, the Rotor speed (rd.s-1)
80
dynamics of the converter. When the converter
60
dynamic is not taken in account, the output voltage of
the converter is v = µE which leads, after averaging, to 40

the following second order model 20

x& 1 = −( f / J )x1 + (K / J )x 2 − T L / J 0 0.05 0.1 0.15 0.2 0.25 0.3

(61)
x& 2 = −(K / L m )x1 − (R m / L m )x 2 + (E / L m )µ
Armature current (A)
6

where we denote by x1 and x 2 , the average values of 4

the angular velocity ω and the dc motor armature 2

circuit current i a , respectively.
Following closely the backstepping design procedure 0
0 0.05 0.1 0.15 0.2 0.25 0.3
in [10], one gets the following regulator time (s)

Fig.6: Tracking behavior of the controller (62) associated,

{(
µ = (JLm / (KE )) K 2 / (JLm ) − f 2 / J 2 ) successively, with the global system (2) and the simplified
system (61).
(
+ KRm / (JLm ) + fK / J 2
) (solid line: simplified system ; dotted line: global system)

− fTL / J +2
(c12 )
− 1 z 1 − (c1 + c 2 )z 2 + ω&&r
where z1 and z 2 are given by (4) and (5),
respectively.
Regulator (62) has been applied successively to the
system (2) and to the simplified model (61). The
relevant design parameters have been given the
following values: c1 = 8x103, c 2 = 104. Fig.6 illustrates
the tracking behaviour of two systems in presence of a
constant reference ω r = 70rad/s. As it can be seen
from the figure, if the regulator (62) perfectly
stabilises the simplified system (61), it is not the same
way for the system (2). This clearly shows that
neglecting the converter dynamics in the regulator
design may lead to drastic deterioration of the closed-
loop performances.
6. Conclusion
In this paper, we have dealt with the problem of DC-
motor velocity control tacking account the dynamics
of a switching power converter. The average controller
design is elaborated via the use of the backstepping
approach. In the case of perfectly known converter
model, the control objective can be ensured using a
backstepping non adaptive controller (13). In the case
of unknown load torque, an adaptive version of the
backstepping controller ((34)-(35)) has been developed
to achieve the control objective. Simulation results
} (62)
References
1. J. Linares-Flores and H. Sira-Ramírez, “DC motor
velocity control through a DC-to-DC power converter”,
Proceedings of 43rd IEEE Conference On decision and
Control, Atlantis, paradise island, Bahamas, December
14-17, 2004, pp. 5297-5302.
2. Gwo-Ruey Yu, Ming-Hung Tseng and Yuan-Kai Lin,
“Optimal positioning control of a DC servo motor using
sliding mode”, Proceedings of the 2004 IEEE
International Conference on Control Applications,
Taipei, Taiwan, September 2-4, 2004, pp. 272-277.
3. M. I. Jahmeerbacus, M.K.Oolun, C. Bhurtun, and
K.M.S. Soyjaudah, “Speed Sosorless Control of a
Converter-fed DC motor”, IEEE Trans. On Ind.
Elect.,Vol. 43, no.4, pp.492-497, 1996.
4. S. E. Lyshevski, Electromechanical System, Electric
Machines, and Applied Mechatronics, Boca raton, FL:
CRC Press, 2000.
5. A. J. Millán, M.I. Giménez de Guzmán, M. Torres, “ A
Cloose Loop DC Motor Speed Control Simulation
System Using Spice”, IEEE International Caracas
Conference on devices, Circuits and Systems, pp. 261-
265, 1995.
6. J.Tze. Huang and Y. Huei Chou, “ Adaptive control of
a shunt DC motor with persistent excitation”,
Proceedings of the 16th IFAC World Congress, Prague,
Czech Republic, July 4-8, 2005.
7. Michael J. Burridge and Logicon RDA, “An Improved
Nonlinear Control design for Series DC Motors”,

6
 Proceedings of the American Control Conference,
Albuquerque, New Mexico, June 1997, pp. 1529-1533.
8. M. F. Rahman, D. Patterson, A. Cheok, and R. Betts,
Motor Drives. Power Electronics Handbook, Academic
Press, 2001.
9. Ahmed Hussein, Kataro Hirasawa and Jinglu Hu,
“Online Identification and Control of a PV-Supplied
DC Motor using Universal Learning Networks”,
ESANN’2003 proceedings-European Symposium on
Artificial Neural networks, Bruges (Belgium), 23-25
April 2003, pp.173-178.
10. Krstić M., I. Kanellakopoulos and P. V. Kokotović,
“Nonlinear and adaptive control design”, John Willy &
Sons, NY, 1995.
11. Krein P.T., J. Bentsman, R.M. Bass, and B. Lesieutre,
“On the use of averaging for analysis of power
electronic system”, IEEE Transaction on Power
Electronics, Vol. 5, n°2, 1990.