You are on page 1of 9

1440 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 21, NO.

4, JULY 2013

Optimal and Fault-Tolerant Torque Control of Servo Motors


Subject to Voltage and Current Limits
Farhad Aghili

Abstract— This brief presents an optimal and ripple-free and hence a sinusoidal commutation can result in torque ripple.
torque control of a brushless servo motor with any back-emk It has been proved that suppressing the torque ripple of the
waveform that minimizes power dissipation subject to voltage motor drive of a servo system can significantly improve the
and current limits of the motor’s drivers/amplifiers. When one
or more phases reach the voltage and/or current limits, the system performance by reducing speed fluctuations [1], [2].
controller optimally reshapes the stator currents of the remaining Commercial high-performance electric motors reduce the pul-
phases for continuing accurate torque production. This allows sating torque by increasing the number of motor poles.
the motor to operate above the rated speed and torque that However, such motors tend to be expensive and bulky due
would be achieved without current reshaping. In the event that to construction and assembly of multiple coil windings.
an open-circuit or short-circuit of a winding occurs, the torque
controller can also isolate the faulty phase in order to generate Control approaches for accurate torque production in
torque as requested given the voltage and current constraints electric motors and their underlying models have been studied
of the healthy phases. Assuming the inductance of stator coils by several researchers [1]–[14]. It was assumed in these
is negligible allows the description of the phase voltage and works that the phase currents can be controlled accurately and
current limit requirements by a set of inequity constraints. It instantaneously, and therefore the currents can be treated as
follows by the derivation of a closed-form solution for the optimal
phase currents at given angular position, velocity, and desired the control inputs. Then, the waveforms of the motor phase
torque—rendering the control algorithm suitable for real-time currents are adequately preshaped so that the generated torque
implementation. Experimental results illustrate the capability is equal to the requested torque. However, when the motor’s
of the controller to achieve precise torque production during drivers have fixed-rate current and voltage limits, some of them
voltage/current saturation of the motor’s drivers or a phase may not be able to deliver the current inputs dictated by the
failure.
electronic commutator that may occur when the motor operates
Index Terms— Constrained optimization, current saturation, at high torque or speed. Consequently, the performance of the
fault-tolerant, optimal control, permanent magnet machines, torque production may significantly deteriorate as a result of
torque control of electric motors, torque ripple, voltage
saturation. phase current distortions caused by either voltage or current
saturation. Flux weakening allows a machine to operate above
the base speed in constant-power, high-speed region when
I. I NTRODUCTION there is a fixed inverter voltage and current [15]. Below the
rated speed, all of the stator currents can be used to produce
B RUSHLESS dc motors are commonly used as the drives
of servo systems in a wide range of industrial applica-
tions from robotics and automation to aerospace and military.
torque. Above the rated speed, a part of the stator current must
be used to oppose the permanent magnet (PM) flux while the
Accurate and ripple-free torque control of brushless motors remaining portion is used to produce torque. Several authors
is essential for precision control of such servo systems. have addressed flux weakening in PM machines [16]–[19].
In brushless motors, the electric power is distributed by an However, this technique can deal with electric motors with
electronically controlled commutation system, instead of a prefect sinusoidal back-electromotive force (EMF) waveform
mechanical commutator found in brushed dc motors. The and, in addition, phase current limits are not taken into
conventional electronic commutator incorporates a feedback account.
from the rotor angular position into a control system, which This brief presents a closed-form solution for optimal
excites the stator coils of the motor in a specific order in excitation currents for accurate torque control of brushless
order to rotate the magnetic field generated by the coils motors with any waveform that minimizes power dissipation
to be followed along by the rotor. Conventional drivers of subject to currents and voltage limits of the motor’s drivers.
brushless motors produce sinusoidal current waveforms for When the motor terminal voltages and/or phase currents reach
smooth motor operation. However, nonideal motors do not their saturation levels, the controller automatically reshapes
have a perfectly sinusoidally distributed magneto-motive force, the excitation currents in such a way that the motor generates
torque as requested. This optimal management of motor’s
Manuscript received April 27, 2011; revised February 9, 2012; accepted excitation currents can significantly increase the rated speed
May 24, 2012. Manuscript received in final form May 24, 2012. Date of pub- and torque of the motor in the face of the voltage and current
lication June 19, 2012; date of current version June 14, 2013. Recommended
by Associate Editor T. Parisini. limits of the drivers. In addition, the torque controller can
The author is with Canadian Space Agency, Saint-Hubert, QC J3Y 8Y9, be used as a remedial strategy to compensate for a phase
Canada (e-mail: farhad.aghili@asc-csa.gc.ca). failure, by optimally reshaping the currents of the remaining
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. healthy phases for accurate torque production. This brief is
Digital Object Identifier 10.1109/TCST.2012.2202118 organized as follows. Section II presents modeling of a brush-
1063-6536/$31.00 © 2012 IEEE

Authorized licensed use limited to: Universidad Nacional de Colombia (UNAL). Downloaded on March 24,2021 at 23:56:33 UTC from IEEE Xplore. Restrictions apply.
AGHILI: OPTIMAL AND FAULT-TOLERANT TORQUE CONTROL OF SERVO MOTORS 1441

less motor subject to current and voltage limits of its phases. (5) must be subject to the following set of 2 p inequality
A closed-form solution for the torque control problem to min- constraints:
imize the power dissipation subject to the phase current and
voltage constraints is given in Section III. Finally, experimental |i k | ≤ i max (6a)
results demonstrating the performance of the controller when |v k | ≤ v max ∀k = 1, . . . , p (6b)
the motor operates above rated torque and speed leading to
current and voltage saturations are reported in Section IV. where v k is the kth phase voltage, and i max and v max are,
respectively, the current and voltage limits of the motor’s
driver. In the following development, we will show that the
II. M OTOR M ODEL W ITH C URRENT AND VOLTAGE L IMITS
above set of 2 p inequalities can be equivalently reduced to
The electromagnetic torque developed by a single phase is a set of p inequalities, if the motor inductance is assumed
a function of the phase current and the rotor angle, while negligible.
cogging torque is only a function of the rotor angle. If a motor The voltage of the winding terminal of an electric motor
operates in a linear magnetic regime, then the torque τ of the is the superposition of the back-EMF voltage and the ohmic
motor with p phases is given by voltage drop, if the inductance of the stator coils is negligible
[20]–[22]. That is
τ (θ, i) = φ T (θ )i + τcog (θ ) (1)
where τcog is the cogging torque, θ is the angular position of v k = Ri k + ek , k = 1, . . . , p (7)
the rotor, and vectors φ ∈ R p and i ∈ R p are defined as
where ek is the back-EMF of the phases, which is equal to the
φ(θ ) = [φ1 (θ ), . . . , φ p (θ )] T rotor speed, ω, times the back-EMF shape functions, φk (θ )
i = [i 1 , . . . , i p ]T . ek = φk (θ )ω
Here, i k and φk (θ ) are the instantaneous current and back-EMF see the Appendix. Using the relation between the terminal
shape function of the kth phase. In rotary electric motors, the voltages and the phase currents in (7), we can rewrite (6b) as
shape functions are periodic functions of rotor angle. Since   v
successive phase windings are shifted by 2π/ p, the kth torque  ω  max
i k + φk (θ ) − ≤0 ∀k = 1, . . . , p (8)
shape function can be constructed as R R
  which are equivalent to
2π(k − 1)
φk (θ ) = φ qθ + (2)
p −v max − ωφk (θ ) v max − ωφk (θ )
≤ ik ≤ , k = 1, . . . , p.
where q is the number of motor poles. Furthermore, since R R
φ is a periodic function with spacial frequency 2π/q, it can (9)
be effectively approximated through the truncated complex Using (6a) in (9) yields
Fourier series as
−v max − ωφk (θ )

N
> −i max
φ(θ ) = cn e j nqθ
(3) R
v max − ωφk (θ )
n=−N < i max ∀θ ∈ R, k = 1, . . . , p.
√ R
where j = −1, {c1 , . . . , c N } are the Fourier coefficients,
and N can be chosen arbitrary large. Similarly, the cogging Therefore, the condition for the existence of a solution for (6a)
torque can be approximated by another finite Fourier series as and (9) can be expressed by
v max + Ri max

N
max |φk (θ )| < . (10)
τcog (θ ) = bn e j nqθ (4) θ,k ω
n=−N
In other words, if (10) is satisfied, then (6a) and (9) can be
where {b1, . . . , b N } are the corresponding Fourier coefficients. combined into the following forms:
The torque control problem is: given a desired motor
torque τd , solve the torque equation (1) in terms of the i k ≤ i k ≤ ī k (11)
phase currents, i. Given a scalar torque set point, mapping
where the lower- and upper-bound current limits are
(1) permits infinitely many phase current waveforms. Since
the continuous mechanical power output of electrical motors  −v max − ωφk (θ ) 
is limited primarily by heat generated from internal copper i k (θ, ω) = max − i max ,
R
losses, it makes sense to use the freedom in the phase current  v max − ωφk (θ ) 
solutions to minimize power losses ī k (θ, ω) = min i max , . (12)
R
Ploss = Ri2 (5) Finally, the current and voltage limit requirements can be
expressed by the following set of inequity constraints:
where R is the resistance of the phases. However, due to
voltage and current limits of the drivers, minimization of |i k − k (θ, ω)| − sk (θ, ω) ≤ 0 ∀k = 1, . . . , p (13)

Authorized licensed use limited to: Universidad Nacional de Colombia (UNAL). Downloaded on March 24,2021 at 23:56:33 UTC from IEEE Xplore. Restrictions apply.
1442 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 21, NO. 4, JULY 2013

where Now, defining the function


1 L = f + λh + μT g (17)
k (θ, ω) = (ī k + i k )
2
1 where f = i T i, g = [g1, g2 , . . . , g p ]T , λ ∈ R, and
sk (θ, ω) = (ī k − i k ). (14) μ = [μ1 , μ2 , . . . , μ p ]T . Let i ∗ provide a local minimum of
2
f (i) satisfying the equality and inequality constraints (15b)
The inequality constraints (13) imposed on the phase currents
and (15c). Assume that column vectors ∇ i g|i=i ∗ are linearly
are equivalent to both voltage and current limits (6). Notice
independent. Then according to the Kuhn–Tucker theorem
that parameters k and sk are not constants rather they are
[23], there exist μk ≥ 0 ∀k = 1, . . . , p such that
functions of mechanical states θ and ω.
∇i L = 0 (18a)
III. O PTIMAL R ESHAPING OF P HASE C URRENTS μk gk (i k∗ ) = 0 ∀k = 1, . . . , p. (18b)
In this section, the optimal value of instantaneous value of Denoting sgn(·) as the sign function, we can show that
each phase current i k will be computed in terms of the design 
sgn(i 1 − 1 ) sgn(i p −  p )
variables φk (θ ), (θ, ω), sk (θ, ω), and τ (θ, ω) at any given ∇ i g = diag ,...,
position θ and speed ω. However, in the following analysis, s1 sp
we drop the position and velocity arguments of the variables in which the columns are linearly independent. The only pitfall
for the simplicity of notation. is i k = 0, where the sign function is indefinite. We assume
that the optimal solutions i k∗ are nonzero because φk = 0. This
assumption will be relaxed later. Substituting f , h, and g from
A. Quadratic Programming
(15) into (18) yields
The derivations in this section present the optimal phase
currents 2sk i k∗ + λsk φk + μk sgn(i k∗ − k ) = 0 (19a)
i ∗ = [i 1∗ , . . . , i ∗p ]T μk (|i k∗ − k | − sk ) = 0. (19b)

which generate the desired torque τd and minimize the power Equations (19a) and (19b) together with (15b) constitute a
losses (5) subject to the constraints (6). By setting τ = τd set of 2 p + 1 nonlinear equations with 2 p + 1 unknowns i ∗ ,
in (1) and using the equivalent inequality constraints (13), λ, and μ to be solved in the following. Since μk gk (i k∗ ) = 0
the problem of finding optimal instantaneous currents can be while μk ≥ 0 and gk (i k∗ ) ≤ 0, we can say that μk = 0 if
equivalently formulated as the following quadratic program- |i k − k | < sk , while μk > 0 if |i k − k | = sk . Therefore,
ming problems: (19a) can be written in the following compact forms:
1
min i T i (15a) Tk (i k∗ − ρk ) = − λφk − k ∀k = 1, . . . , p. (20)
T
2
s.t. h = φ i + τcog − τd = 0 (15b) The mapping Tk : Dk → R, in which
|i 1 − 1 |
g1 = −1≤0 Dk (x) = {x ∈ R : |x − k | ≤ sk }
s1
.. is defined by
. 
|i p −  p | x, |x − k | < sk
gp = − 1 ≤ 0. (15c) Tk (x) = μk (21)
sp x+ 2sk sgn(x − k ), |x − k | = sk .
Note that all the instantaneous variables in the above equality It is apparent that the mapping is invertible on D, that is there
and inequality constants, i.e., φk , k , and sk , are at given rotor exists a function T −1 (x) such that Tk−1 (Tk (x)) = x ∀x ∈ D.
angular position, θ , and velocity ω. Since all the functions In other words, the variable i k∗ in (20) can be determined
are convex, any local minimum is a global minimum as well. uniquely if the right-hand side of the equation is given. The
Now, we seek the minimum point i ∗ satisfying the equality and inverse of the mapping is the saturation function, i.e., Tk−1 (·) ≡
inequality constraints. Before we pay attention to the general sk sat(·/sk ) where

solution, it is beneficial to exclude the trivial solution, i k∗ = 0. ⎨ 1, x >1
If the kth torque shape function is zero, that phase contributes sat(x) = x, −1 ≤ x ≤ 1 (22)
no torque regardless of its current. Hence ⎩
−1, x < −1.
φk = 0 ⇒ i k∗ = 0 ∀k = 1, . . . , p (16) In view of the function definition, we can rewrite (20) as
 
∗ −0.5λφk − k
immediately specifies the optimal phase currents at the i k = k + sk sat ∀k = 1, . . . , p. (23)
crossing point. By excluding the trivial solution, we deal with sk
a smaller set of variables and number of equations in our Upon substitution of the optimal phase currents from (23) into
optimization programming. Therefore, we have to find the the torque equation (15b), we arrive at
optimal solution corresponding to the nonzero part. Hereafter, p  
−0.5λφk − k
without loss of generality, we assume that all torque shape φk sk sat = τd − τcog − τ (24)
functions are nonzero. sk
k=1

Authorized licensed use limited to: Universidad Nacional de Colombia (UNAL). Downloaded on March 24,2021 at 23:56:33 UTC from IEEE Xplore. Restrictions apply.
AGHILI: OPTIMAL AND FAULT-TOLERANT TORQUE CONTROL OF SERVO MOTORS 1443

30 Amplifier
i∗1 Stator

θ = 100 (deg) Optimal


20
τd i∗2
Controller
M
vk ≤ vmax i∗3
10
ik ≤ imax τ
Mechanical
ψ(λ)

0 Load
ω d
dt
φ(θ) θ
−10
Fourier
τcog (θ)
series
−20

Fig. 2. Architecture of the optimal torque controller.


−30
−40 −30 −20 −10 0 10 20 30 40
λ
subject to current and voltage limits of the drivers may proceed
Fig. 1. Piecewise linear function of the Lagrangian multiplier. by the following steps.
1) Given angular position θ , calculate the shape functions,
φk , and cogging torque, τcog , form their Fourier coeffi-
where

p cients according to (2)–(4).
τ = φjj (25) 2) Given angular velocity ω, compute sk , k , and τ from
j =1 the current and voltage limits, i max and v max , according
to (12), (14), and (25).
is given at any position θ . Note that τ does not have any phys-
3) Calculate the optimal currents from (29). Go to
ical meaning but can be interpreted as the torque associated
step 1).
with nonzero mean values of the phase current limits. Since
λ is the only unknown variable in (24), finding the optimal
values of the phase currents boils down to solve the algebraic B. Fault-Tolerant
equation (24) for the Lagrangian multiplier. Apparently, the The optimal solution given in the previous section can be
left-hand side of (24) is a piecewise linear function of λ. Thus, readily used as a remedial strategy to recover from a phase
(24) can be concisely written as failure. The idea is that when an open-circuit or short-circuit
ψ(λ) = τd − τcog − τ . (26) fault occurs, the faulty phase is isolated and the currents
of the remaining healthy phases are optimally reshaped for
As an illustration, the piecewise linear function for a motor accurate torque production. It is clear from (16) that when the
system, which will be described in Section IV, is depicted in instantaneous back-EMF of a phase takes a zero value then the
Fig. 1 for a particular motor angle. The slope of the piecewise corresponding phase current becomes trivially zero. Therefore,
linear function abruptly decreases each time one of the phases the faulty phase can be easily isolated by setting the value
saturates and when all phases saturate the slope become zero, of its back-emk shape function to zero in the optimization
in that case there is no solution for λ. Therefore, the piecewise formulation. Now, suppose that the motor’s shape functions for
linear function is invertible only if the following conditions are the optimization programming (15) are modified according to
satisfied: 
φk , for normal phase
φk = (30)
ψmin + τcog − τ ≤ τd ≤ ψmax + τcog − τ . (27) 0, for faulty phase.
Then, the corresponding Lagrangian multiplier can be Then, the optimal solution (29) ensures that the instanta-
obtained by neous currents of faulty phase is zero, while the remaining
λ = ψ −1 (τd − τcog − τ ). (28) phases generate the desired torque satisfying the optimality
conditions.
Finally, substitution of (28) into (23) yields the optimal current
of each phase as a function of the instantaneous values of the
IV. E XPERIMENT
motor angle and speed by
In order to evaluate the performance of the optimal torque
i k∗ (θ, ω) = k (θ, ω) controller, experiments were conducted on a three-phase syn-
 
−0.5ψ −1 (τ̄ )φk (θ ) − k (θ, ω) chronous motor with nine pole pairs. The experimental setup is
+ sk (θ, ω)sat (29)
sk (θ, ω) illustrated in Fig. 3. Three independent current servo amplifiers
where (Advanced Motion Control 30A20AC) control the motor’s
excitation currents as specified by the torque controller. The
τ̄ = τd − τcog (θ ) − τ (θ, ω).
electric motor and a hydraulic rack and pinion rotary motor
The optimal torque controller is schematically illustrated are mounted on the rigid structure of a dynamometer. The
in Fig. 2. In summary, the optimal torque control algorithm hydraulic motor’s shaft is connected to that of the electric

Authorized licensed use limited to: Universidad Nacional de Colombia (UNAL). Downloaded on March 24,2021 at 23:56:33 UTC from IEEE Xplore. Restrictions apply.
1444 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 21, NO. 4, JULY 2013

2
Brushless Torque phase 1
motor transducer Hydraulic motor phase 2
1.5 phase 3

0.5

φ (Nm/Amp)
0

−0.5

−1

−1.5

Fig. 3. Motor prototype mounted on the dynamometer.


−2
100 110 120 130 140 150 160
Motor angle (deg)

motor via a torque transducer (Himmelstein MCRT 2804TC) Fig. 4. Back-EMF waveform.
by means of two couplings, which relieve bending moments or
shear forces due to small axes misalignments. The speed of the
hydraulic motor is controlled by a pressure compensated flow every motor position θ , the current load cases i 1 s and the
j
control valve. The hydraulic pressure is set sufficiently high corresponding measured torques τ j (θ ) are related by
so that the hydraulic actuator can always regulate the angular ⎡1 ⎤ ⎡ 1 ⎤
speed regardless of the electric motor torque. In other words, i1 1 τ (θ )
the operating speed of the electric motor is independently set ⎢i 2 1⎥   ⎢ τ 2 (θ ) ⎥
⎢ 1 ⎥ φ1 (θ ) ⎢ ⎥
by the hydraulic actuator. The phase voltages are attenuated ⎢ .. .. ⎥ = ⎢ . ⎥. (32)
⎣ . . ⎦ τcog (θ ) ⎣ .. ⎦
by resistor branches and then sensed by Isolation Amplifiers
(model AD 210 from Analog Device). The Isolation Ampli- i 1u 1 τ u (θ ).
fier System eliminates the possibility of leakage paths and Then, φ1 (θ ) and τcog (θ ) can be obtained at every given
ground loops between the power servo amplifiers and the position by using the pseudo-inverse of the above matrix
data acquisition system, by providing complete transformer equation.
isolation. A multichannel data acquisition system acquires the Since the motor has nine pole pairs, the torque trajectory is
analog data at the sampling rate of 1 kHz. Errors between periodic in position with a fundamental spatial-frequency of
the measured rotor position, used by the torque controller 9 cpr (cycles/revolution) and thus the torque pattern repeats
for current commutation, and the true rotor position will every 40 degrees. The Fourier coefficients can by calculated
result in torque ripple. One source of this error is measured by the inverse Fourier series
quantization. In order to minimize the torque-ripple induced by
this quantization as much as possible, the motor uses a high- θo + 2π
 q
resolution encoder with 0.001° resolution for measurement of cn = φ1 (θ )e− j nqθ
angular position. θ=θo
θo + 2π
 q
A. Identification of Motor Parameters bm = τcog (θ )e− j mqθ .
Torque shape functions are measured by using the hydraulic θ=θo
dynamometer [24]. To this end, the torque trajectory data
The phase shape functions, φk , and the cogging torque are
versus position was registered during the rotation, while one
shown in Fig. 4, and the corresponding Fourier coefficients
phase was energized with a constant current. The motor is
are given in Table I.
with winding resistance
R = 2.54 (31)
B. Performance Test
and inductance L = 12.5 mH and hence electric time-constant The proposed optimal torque controller subject to phase
is μ = 0.0049 sec. The torque-angle data are registered current and voltage limits has been implemented on the three-
within almost one rotation while the phase current is kept phase motor. The objective of this section is to demonstrate
constant. But the current is incremented at the end of each that the optimal torque controller can deliver accurate torque
rotation stroke by 1 Amp until an ensemble of torque pro- production in the face of current and voltage saturation.
files belonging to the span of [−15, 15] Amp is obtained. It has been shown [2] that nonconstrained optimization of
By taking the average of two sequences of the motor torques motor torque leads to the following solution of phase currents:
corresponding to when the motor shaft is rotated clockwise
and counter-clockwise, the effect of the bearing friction torque φk (θ )
i k (θ ) = (τd − τcog ). (33)
in the motor torque measurement is compensated. Now at φ(θ )2

Authorized licensed use limited to: Universidad Nacional de Colombia (UNAL). Downloaded on March 24,2021 at 23:56:33 UTC from IEEE Xplore. Restrictions apply.
AGHILI: OPTIMAL AND FAULT-TOLERANT TORQUE CONTROL OF SERVO MOTORS 1445

0.5 4
phase 1
phase 2
0.4
3 phase 3
0.3
2
0.2

1
0.1
τcog (Nm)

(Amp)
0 0

−0.1
−1
−0.2
−2
−0.3

−0.4 −3

−0.5 −4
100 110 120 130 140 150 160 100 110 120 130 140 150 160
Motor angle (deg) Motor angle (deg)
11
Fig. 5. Waveform of the cogging torque. phase 1
phase 2
phase 3
TABLE I 10

H ARMONICS OF THE BACK -EMF, φ(θ ), AND C OGGING T ORQUE , τcog (θ )


9
Harmonics cn bn

s (Amp)
no. (nm/A) (nm)
8
1 0.2730 + 0.7270 j 0.0236 − 0.0476 j
2 −0.0002 − 0.0002 j −0.0010 + 0.0009 j
3 −0.0004 − 0.0004 j 0.0001 − 0.0007 j 7
4 −0.0000 + 0.0002 j 0.0001 − 0.0007 j
5 0.0002 + 0.0001 j 0.0006 + 0.0003 j
6 0.0000 − 0.0000 j 0.0000 + 0.0003 j 6
7 0.0003 + 0.0001 j −0.0008 − 0.0004 j
8 0.0000 − 0.0000 j 0.0000 − 0.0006 j
5
9 −0.0002 + 0.0000 j 0.0002 − 0.0001 j 100 110 120 130 140 150 160
10 0.0001 + 0.0000 j −0.0006 + 0.0001 j Motor angle (deg)
11 −0.0199 − 0.0021 j −0.0010 + 0.0009 j
12 0.0009 − 0.0008 j 0.0863 + 0.0122 j Fig. 6.  and s variables corresponding to the phase current and voltage
13 −0.0022 + 0.0004 j −0.0003 − 0.0002 j limits imax = 10 Amp, v max = 40 Volt, and ω = 21 rad/s.
14 0.0001 − 0.0001 j −0.0006 + 0.0001 j −6
15 0.0000 + 0.0001 j −0.0000 − 0.0003 j

−6.5

For a comparative result, the performances of both constrained


−7
optimization torque controller (29) and nonconstrained opti-
mization torque controller (33) are presented. It is worth noting
τ (Nm)

that, in the case of nonconstrained optimization torque con- −7.5

troller, the maximum torque is reached as soon as one phase


saturates. On the other hand, the constrained optimization −8

algorithm increases the torque contribution of the unsaturated


phases when one phase saturates, until, in the limit, all phases −8.5
are saturated.
The motor shaft is rotated by the hydraulic actuator at a −9
100 110 120 130 140 150 160
constant speed while the motor torque is monitored by the Motor angle (deg)
torque transducer. The maximum phase voltage and current
are specified as Fig. 7. Virtual torque.

i max = 10 Amp
Driver limits: (34) For the sake of illustration, the graphs of the current variables
v max = 40 Volt.
k and sk for the particular operating conditions (34) and (35)
In the first part of the experiment, the desired torque and are plotted in Figs. 6 and 7, respectively. It should be noted
the rotation speed of the hydraulic motor are set to that these waveforms are not fixed when the motor speed
 changes. Figs. 8 and 9 illustrate the phase current waveform
τd = 10 nm and the time-history of the subsequent terminal voltages,
Case I: (35)
ω = 21 rad/s. respectively. It is clear from the latter figure that the phase

Authorized licensed use limited to: Universidad Nacional de Colombia (UNAL). Downloaded on March 24,2021 at 23:56:33 UTC from IEEE Xplore. Restrictions apply.
1446 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 21, NO. 4, JULY 2013

5
phase 1 phase 1
phase 2 10 phase 2
4 phase 3
phase 3

Phase current (Amp)


Phase current (Amp)

0 0

−1

−2 −5

−3

−4
−10
−5
100 110 120 130 140 150 160 100 110 120 130 140 150 160
Motor angle (deg) Motor angle (deg)

Fig. 8. Case I: phase currents. Fig. 11. Case II: phase currents.

50
phase 1 50
phase 1
phase 2
40 phase 2
phase 3 40
phase 3
30
30
Terminal voltage (Volt)

Terminal voltage (Volt)


20
20

10
10

0
0

−10
−10

−20
−20

−30
−30

−40
−40

−50
100 110 120 130 140 150 160 −50
100 110 120 130 140 150 160
Motor angle (deg) Motor angle (deg)

Fig. 9. Case I: phase voltages. Fig. 12. Case II: phase voltages.
12
27

11
26
10

25
9
Torque (Nm)

Torque (Nm)

8 24

7
23
6 τd = 10 Nm τd = 25 Nm

ω = 21 rad/s 22 ω = 2 rad/s
5

4 21
sat. is taken into account
sat. is not taken into account sat. is taken into account
3 sat. is not taken into account
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 20
Time (s) 0 5 10 15 20 25 30 35 40
Time (s)
Fig. 10. Case I: motor torque with and without taking the phase current and
Fig. 13. Case II: motor torque with and without taking the phase current
voltage saturation into account.
and voltage saturation into account.

voltages reach their limits in this experiment, but the phase relatively low because the motor operates well below the rating
currents are far from the current limit. Note that phase voltage torque. Trajectories on the motor torque under the optimal
saturation is caused by high motor speed, which generates high torque controller with and without taking the saturation in
back-EMF voltage. However, the phase currents remain to account are shown in Fig. 10 by solid and dashed lines,

Authorized licensed use limited to: Universidad Nacional de Colombia (UNAL). Downloaded on March 24,2021 at 23:56:33 UTC from IEEE Xplore. Restrictions apply.
AGHILI: OPTIMAL AND FAULT-TOLERANT TORQUE CONTROL OF SERVO MOTORS 1447

30 50
phase 2
40 phase 3

20
30
constrainted optimization
Maximum torque (Nm)

Terminal voltage (Volt)


non−constrainted optimization
Case I operating point 20
10
Case II operating point
10

0 0

−10
−10
−20

−30
−20
−40

−30 −50
−30 −20 −10 0 10 20 30 100 110 120 130 140 150 160
ω (rad/s) Motor angle (deg)

Fig. 14. Maximum attainable motor torque with constrained and noncon- Fig. 16. Case III: phase voltages when phase 1 is open circuit.
strained optimization of motor torques. 12

phase 2 11
10 phase 3

10

5 9
Phase current (Amp)

Torque (Nm) 8

0 7

−5
5

4
with compensation
−10 without compensation
3
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
100 110 120 130 140 150 160 Time (s)
Motor angle (deg)
Fig. 17. Case III: motor torque when phase 1 is open circuit.
Fig. 15. Case III: phase currents when phase 1 is open circuit.

increases the torque contribution of the unsaturated phases


respectively. It is clear from the graphs that the proposed when one phase saturates. This is clearly demonstrated in
optimal controller is able to significantly reduce the torque Fig. 14 that depicts the maximum attainable motor torques
ripple induced by phase voltage saturation. corresponding to the solutions of the constrained optimization
At low velocity and high desired torque, it is most likely (29) and nonconstrained optimization (33). The graphs indi-
that the phases current saturation occur rather than the phase cate that the maximum torque capability is boosted by 20%
voltage saturation. As an illustration, in the second part of the when the phase saturation is considered in the phase current
experiment, the desired torque and the shaft speed are set to shape function.

τd = 25 nm
Case II: (36)
ω = 2 rad/s. D. Single-Phase Failure
The optimal torque controller can be readily used as reme-
It is apparent from the values of phase currents and voltages in
dial control strategy under operation of a single-phase failure.
Figs. 11 and 12 that the phase currents reach their limit while
In this experiment, the current circuit of the motor’s first phase
the phase voltages are not saturated. Again, it is clear from
is virtually broken by sending zero signal to the enable port
the trajectories of the motor torque under operating condition
of the corresponding power amplifier. Thus
(36) in Fig. 13 that the proposed optimal controller is able to
immunize the torque ripples induced by the current saturation. φ1 (θ ) ≡ 0 ∀θ ∈ R. (37)
The objective is to produce the same torque 10 nm as the three
C. Motor Torque-Velocity Characteristic phases by using only the remaining two phases
The control algorithm presented in previous section permits ⎧
torque sharing among phases when some phases saturate. ⎨ τd = 10 nm

This results in a considerable increase in the attainable maxi- Case III: ω = 15 rad/s (38)


mum motor torque because the torque controller automatically φ1 ≡ 0 (open-cuircuit).

Authorized licensed use limited to: Universidad Nacional de Colombia (UNAL). Downloaded on March 24,2021 at 23:56:33 UTC from IEEE Xplore. Restrictions apply.
1448 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 21, NO. 4, JULY 2013

The currents and voltages of the two normal phases are shown [3] Y. Murai, Y. Kawase, K. Ohashi, K. Nagatake, and K. Okuyama, “Torque
in Figs. 15 and 16. The waveforms of the phase voltages and ripple improvement for brushless DC miniature motors,” IEEE Trans.
Ind. Appl., vol. 25, no. 3, pp. 441–450, May–Jun. 1989.
currents are still periodic but highly distorted because of the [4] N. Matsui, T. Makino, and H. Satoh, “Autocompensation of torque
lack of one phase. The dashed line in Fig. 17 depicts motor ripple of direct drive motor by torque observer,” IEEE Trans. Ind. Appl.,
torques when one phase is open circuit. The dashed lines in the vol. 29, no. 1, pp. 187–194, Jan.–Feb. 1993.
[5] D. G. Taylor, “Nonlinear control of electric machines: An overview,”
figures show motor torque produced by a controller assuming IEEE Control Syst. Mag., vol. 14, no. 6, pp. 41–51, Dec.
all three phases are normal. This results in drastic torque 1994.
fluctuation because the healthy phases do not compensate [6] C. French and P. Acarnley, “Direct torque control of permanent magnet
drives,” IEEE Trans. Ind. Appl., vol. 32, no. 5, pp. 1080–1088, Sep.–Oct.
for lacking torque of the faulty phase. The solid line in the 1996.
figure shows that the motor still produces the constant desired [7] J.-K. Kang and S.-K. Sul, “New direct torque control of induction
torque when the torque controller is designed based on the motor for minimum torque ripple and constant switching frequency,”
IEEE Trans. Ind. Appl., vol. 35, no. 5, pp. 1076–1082, Sep.–Oct.
two healthy phases. 1999.
[8] Y. Wang, D. Cheng, C. Li, and Y. Ge, “Dissipative Hamiltonian
realization and energy-based L 2 -disturbance attenuation control of mul-
V. C ONCLUSION timachine power systems,” IEEE Trans. Autom. Control, vol. 48, no. 8,
A closed-form solution to the problem of optimal torque pp. 1428–1433, Aug. 2003.
[9] Z. Xu and M. F. Rahman, “A variable structure torque and flux controller
control of brushless motors minimizing the copper loss subject for a DTC IPM synchronous motor drive,” in Proc. IEEE 35th Annu.
to voltage and current limits of the drivers/amplifiers has Power Electron. Special. Conf., vol. 1. Jun. 2004, pp. 445–450.
been found. It has been shown that accurate and ripple-free [10] S. Ozturk and H. Toliyat, “Direct torque and indirect flux control of
brushless DC motor,” IEEE/ASME Trans. Mechatron., vol. 16, no. 2,
torque production is possible even if the voltage or current pp. 351–360, Apr. 2011.
of one or more phases reach their saturation level. In this [11] H. Grabner, W. Amrhein, S. Silber, and W. Gruber, “Nonlinear feedback
control approach, the excitation currents were reshaped when control of a bearingless brushless DC motor,” IEEE/ASME Trans.
Mechatron., vol. 15, no. 1, pp. 40–47, Feb. 2010.
the terminal voltage and/or phase current reach their saturation [12] C.-W. Tsai, C.-L. Lin, and C.-H. Huang, “Microbrushless DC motor
level in such a way that the motor can still generate torque as control design based on real-coded structural genetic algorithm,”
requested. Moreover, the developed control algorithm allowed IEEE/ASME Trans. Mechatron., vol. 16, no. 1, pp. 151–159, Feb. 2011.
[13] F. Aghili, “Fault-tolerant torque control of BLDC motors,” IEEE Trans.
the motor optimally generates precise torque when operating Power Electron., vol. 26, no. 2, pp. 355–363, Feb. 2011.
under a phase failure. Experimental results obtained from a [14] F. Aghili, “Optimal joint torque control subject to voltage and current
brushless servo-motor under the proposed torque controller limits of motor drivers,” in Proc. IEEE/RSJ Int. Conf. Intell. Robots
Syst., San Francisco, CA, Sep. 2011, pp. 3784–3791.
demonstrated accurate torque production under voltage/current [15] A. El-Refaie, D. Novotny, and T. Jahns, “A simple model for flux
saturation of the motor’s drivers or failure of one phase. weakening in surface PM synchronous machines using back-to-back
thyristors,” IEEE Power Electron. Lett., vol. 2, no. 2, pp. 54–57, Jun.
2004.
A PPENDIX [16] T. Sebastiangordon and G. R. Slemon, “Operating limits of inverter-
The voltage equations are given by driven permanent magnet motor drives,” IEEE Trans. Ind. Appl., vol. 23,
no. 2, pp. 327–333, Mar. 1987.
di k [17] T. M. Jahns, “Flux-weakening regime operation of an interior
vk = L + Ri k + ek permanent-magnet synchronous motor drive,” IEEE Trans. Ind. Appl.,
dt
vol. 23, no. 4, pp. 681–689, Jul. 1987.
where L is the self-inductance. The above equations in the [18] B. Sneyers, D. W. Novotny, and T. A. Lipo, “Field weakening in buried
Laplace domain can be written as permanent magnet AC motor drives,” IEEE Trans. Ind. Appl., vol. 21,
  no. 2, pp. 398–407, Mar. 1985.
1 [19] R. Schiferl and T. A. Lipo, “Power capability of salient pole per-
(v k − ek ) = Ri k (39) manent magnet synchronous motors in variable speed drive applica-
1 + μs tions,” IEEE Trans. Ind. Appl., vol. 26, no. 1, pp. 115–123, Jan.–Feb.
where μ = L/R is the electrical time-constant and s is the 1990.
[20] J. R. Hendershot and T. J. E. Miller, Design of Brushless
Laplace variable. For small time-constant μ  1, the fast Permanent-Magnet Motors. Oxford, U.K.: Magna Physics & Clarendon,
dynamics can be ignored and hence (39) can be effectively 1994.
[21] O. Moseler and R. Isermann, “Application of model-based fault detection
approximated by (7). to a brushless DC motor,” IEEE Trans. Ind. Electron., vol. 47, no. 5,
pp. 1015–1020, Oct. 2000.
R EFERENCES [22] J.-H. Lee, D.-H. Kim, and I.-H. Park, “Minimization of higher back-
EMF harmonics in permanent magnet motor using shape design sen-
[1] S. J. Park, H. W. Park, M. H. Lee, and F. Harashima, “A new sitivity with B-spline parameterization,” IEEE Trans. Magn., vol. 39,
approach for minimum-torque-ripple maximum-efficiency control of no. 3, pp. 1269–1272, May 2003.
BLDC motor,” IEEE Trans. Ind. Electron., vol. 47, no. 1, pp. 109–114, [23] H. W. Kuhn and A. W. Tucker, “Nonlinear programming,” in Proc. 2nd
Feb. 2000. Berkeley Symp. Math. Stat. Probab., 1951, pp. 481–492.
[2] F. Aghili, M. Buehler, and J. M. Hollerbach, “Experimental character- [24] F. Aghili, J. M. Hollerbach, and M. Buehler, “A modular and
ization and quadratic programming-based control of brushless-motors,” high-precision motion control system with an integrated motor,”
IEEE Trans. Control Syst. Technol., vol. 11, no. 1, pp. 139–146, Jan. IEEE/ASME Trans. Mechatron., vol. 12, no. 3, pp. 317–329, Jun.
2003. 2007.

Authorized licensed use limited to: Universidad Nacional de Colombia (UNAL). Downloaded on March 24,2021 at 23:56:33 UTC from IEEE Xplore. Restrictions apply.

You might also like