2 A[ (4)
i&.herea, and ad are the quadrature and direct axis cur-
rmts, L , and Ld are transformed winding inductances, where J is the mechanical inertia and D is the viscous
A! is the winding resistance, npis the number of perma- damping factor. The load torque function (TL) when
nent magnet pole pairs, w is the rotor angular velocity, the motor is rotating a rigid link in the vertical plane
firTzis a positive constant, and U, and U d represent the is given by'
transformed winding voltages. The torque produced by
the motlor is given by
That is, 71 is bounded by a first-order polynomial in
its arguments and N I is an unknown positive constant.
The reference model to be tracked is given by
where Krl is a positive constant. In practical cases,
the constant KT2is usually much larger than Krl. The
actual phase currents (and voltages) may be obtained
by inverting the Park transformation.
When magnetic saturation is present, the Park trans- where "r" is a reference signal. To facilitate fur-
fcrmation may not be applied in general. It was shown ther development, we use the notation T' = i,,f =
in [l]that in the presence of magnetic saturation, the [e - 6 d w - WdIT and consider 7' to be the input to the
quantities L,, Ld, Krl and Kr2 may be modeled as system. Using our robust control design methodology
piecewise continuous functions of the phase currents in [8,9], the desired torque profile TA is designed as
different intervals of currents. Although this model ac-
counts for the effects of saturation accurately, it is some-
what cumbersome for control design purposes. There-
fore, to be able to take advantage of the Park trans-
formation, an alternative approach was suggested in
[6] wherein the parameters L,, Ld, K r l , and Krz in where ,8 is the adaptation gain to counter the effect of
thLe motor dynamics (1)-(2)are modeled as piecewise the uncertainties. The adaptation law is
constants in different intervals of current. Such an ap-
pioach enables us to retain the dynamics given by (1)
and (2) while including the effects of magnetic satura-
tion in the motor model. The desired approximation
tc the magnetic saturation characteristics may be ob- where P is the positive definite, symmetric solution to
tained by choosing sufficiently small current intervals. the Algebraic Ricatti Equation
In the following section, we describe the application
of the robust adaptive control scheme for a class of non-
linear systems [8,9] to the BLDC motor and to multi-
P A + ATP - 26P [y ][ 0 1 ]P+Q=O (9)
link manipulators actuated by BLDC motors. The elec-
tr3mechanical dynamics of the motor fall into the class with Q being a positive definite, symmetric matrix and
of systems known as the Special Strict Feedback Form 6 is given by
[8,9], to which our design procedure is applicable. The
proposed voltage level controller is robust t o both para-
metric and dynamic uncertainties and can also reject
unknown, bounded disturbances entering the system where the parameter p is chosen later.
(although we do not show it here). Finally, simula-
tion results and conclusions are presented in sections The stability properties are analyzed using the can-
I\' and V respectively. didate Lyapunov function
2
111. Control Design
v,
~
= zTpz"+rO(&p*) (11)
lWe consider the load torque and friction terms as dynamic
In this section, the control design methodology is out- uncertainties originating in the mechanical subsystem of the
lined for the BLDC motor. For brevity, we first start plant.
289
where ?!,* is the desired value of the adaptation gain
and ro is a positive number. Differentiating along the
trajectories of the system, we obtain an upper bound
for VI as
2
Vi L -(LiQ(Q) - ~-~<1~)11z11~
- go(P - P*)
- [4P*(1+ p ) - ( S + 2)] IIBTPZl12 TOP*^
+ 112BTPzII 131
I
+N (12)
where 7 is the torque tracking error and
Cl = ll[ p +Ill (13)
where x is a known function of its arguments and is
given by
we need t o ensure that the desired currents zqd and i d d where iis the estimate of c required to counter the
produce the required torque. To this end, the following effect of the disturbances and I71 is a positive number.
commutation scheme is employed: Differentiating along the trajectories of the system, we
have
idd =0 , iqd = TA (16)
and p in equation (10) is chosen such that -1 < p 5
v2 = vl + Giq + '&id + 2r1(i- c)?
2
% - 1. (A crude estimate of the quantity is suf- % 5 - ( L i Q ( Q ) - C-1<i2)11Z112 -go@ - P*)
ficient to choose the correct p.) The torque tracking
error is given by
- [4P*(1+ p ) - (c
- + 2)] IIBTPfl12+ a0/3*~
i
+N [Vq -i q R - n p L d i d w - K T 2 W - Lq6qd
L q
1
1- + c)t.
I I
290
whlere the control gains Ki, i = 1 , . . . ,15 are all positive where z1 = 8, the vector of joint angles and z2 = 4,
x
with as given by (24), p as given by (27)' and ,8 given D(0) is the n x n positive definite symmetric inertia
by the adaptation law (8). The adaptation law for iis matrix, C(0,e),G(0), and F ( d ) are the n x 1 vectors
given by representing the Coriolis and centrifugal forces, gravity
n-1
loading and friction respectively. From physical consid-
erations, the unknown vector ib(x1, z 2 )may be bounded
by a second-order polynomial in the states as follows:
-r o1i. (30)
11@(~1,z2)ll F $0 + $111~11+ $211z1I2 (38)
The desired control gain p* and the design freedom s
are chosen along the derivation in [lo]: where $0,$1, and $2 are unknown constants. The ref-
erence model to be tracked is given by
id = A Z d +B[KZd + T ] (39)
where A and B are as in (37). Choosing p as
-r;la0j (42)
where P is the solution t o the algebraic Ricatti equa-
+
tion (9), and 8 = a ( 1 p). Once the desired torque
profile is obtained, the final voltage level controllers are
then designed using backstepping and the robust con-
trol methodology to track the desired currents given
by the commutation law (16). Due to the nature of
the bound on the uncertainties (38), additional terms
Therefore, the solutions converge to the compact set of the form $;ll,Zll4 accumulate in the derivative of the
augmented Lyapunov function
[:;I
be written as resultant voltage level controller is able to compensate
for parametric and dynamic uncertainties in both the
= [ o0 In
o][;:]+[:]x mechanical subsystem (such as inertia, friction, gravity
and Coriolis effects) as well as in the electrical sub-
system (such as winding inductances and resistances).
The details are not presented here since it is rather
straightforward to derive from earlier results; however,
it is notationally cumbersome.
29 1
IV. Simulation Results
Units
KT2
L,
Ld mH
RI, R2 -0.1
0 2 4 6 6 10 12
l e (sermds)
Tnm
14 16 18
The reference model to be tracked is given by Figure 2: Tracking error using the robust adaptive con-
trol.
7T
ed(t) = - sin(t)(l - exp(-0.1t2)) (44)
2
I . . , , , , , , ,
4'
0
"
2 4
"
6 8
"
10 12
"
14 16
'
18 20
1
Time (semnds)
292
5, , , , , , , , . , ,
0 2 4 6 6 10 12 I4 (6 I6 20
4r , , , , rm I, d
, , , , ,
293