This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY 1

Speed Control of Permanent Magnet Synchronous Motor With Uncertain

Parameters and Unknown Disturbance
Abdullah A. Alfehaid , Elias G. Strangas, Life Senior Member, IEEE, and Hassan K. Khalil , Life Fellow, IEEE

Abstract— Control of the speed as well as shaping the speed in applications where high performance and high precision are
transient response of a surface-mounted permanent magnet required. Therefore, various more-advanced control techniques
synchronous motor (PMSM) is achieved using the method of feed- have been applied, such as sliding-mode control, feedback
back linearization and extended high-gain observer. To recover
the performance of feedback linearization, an extended high-gain linearization, fuzzy control, and model predictive control.
observer is utilized to estimate both the speed of the motor Because high performance has to be achieved in the presence
and the disturbance present in the system. The observer is of unknown load and uncertain parameters, it was determined
designed based on a reduced model of the PMSM, which is early on that the performance of these techniques can be
realized through the application of singular perturbation theory. improved by the use of disturbance observers that estimate the
The motor parameters are assumed uncertain and we only
assume knowledge of their nominal values. The external load load as well as the disturbance due to parameters uncertainty.
torque is also assumed to be unknown and time-varying, but A wide range of different disturbance estimation techniques
bounded. Stability analysis of the output feedback system is given. can be found in [4]–[6] and the references therein.
Experimental results confirm the performance and robustness The literature on the use of disturbance observers in
of the proposed controller and compare it to the cascaded the speed control of PMSMs is rich. Reference [7] surveys
proportional integral (PI) speed controller.
the literature through 2016. More recent work is included
Index Terms— Disturbance rejection, extended high-gain in the group of [8]–[16] and the reference therein. These
observer, feedback linearization, permanent magnet synchronous papers, which are published in the leading journals in the
motor (PMSM), singular perturbation, speed control.
field, are chosen to represent different control approaches.
References [8]–[10] use sliding-mode control with additional
I. I NTRODUCTION
features to reduce chattering, such as dynamic reaching laws

I N INDUSTRY, particularly in servo applications such as

computer numerical control (CNC) machines and robotic
systems, permanent magnet synchronous motors (PMSMs)
[8], [9] and adaptive terminal sliding-mode [10]. The observer
used with sliding-mode control could be an extended-state
observer [9] or a sliding-mode one [8], [10]. References [11]
have already been largely employed and rapidly replacing and [12] use model predictive control. Reference [13] uses
induction and dc motors. There are several advantages of feedback linearization with an extended high-gain observer
using PMSMs over dc and induction motors such as efficiency, that extends the state with multiple states. Reference [14] uses
high-power density, lightweight, maintenance-free, and small linear quadratic regulator (LQR) with an observer that assumes
size [1]. However, the mathematical model of a PMSM is periodic disturbance. Reference [15] uses a fuzzy controller
nonlinear and thus not easy to control. In addition, temperature with a fuzzy observer. Finally, [16] designs a disturbances
changes and variation in operating points could greatly affect observer that accounts for multiple disturbances that include
the parameters of PMSMs, for example, the stator winding cogging torques, load torques, friction torques, measurement
resistance can vary by as much as 200% of its nominal value error effects, dead-time effects, and parameter perturbations.
and the rotor flux linkage can vary by as much as 20% of Most of these papers present experimental/simulation results
its nominal value [2]. Moreover, the load torque in many that show superior performance to traditional PI controllers.
industrial applications is unknown, which adds more com- Reference [13] is the closest to our current work because it
plication to the control problem. Addressing these challenges uses feedback linearization but our observer is of lower order
require robust control techniques, which have led to a variety and our controller shapes the transient response.
of control approaches that have been successfully applied. The contribution of this brief paper is based on multiple
In industry, proportional integral (PI) controllers have been features that previous papers in the literature might have some
largely used in motor drive systems [3]. It is one of the of them but there is no paper that has all of them. These
simplest control techniques that offers good performance. features are:
However, PI controllers are not a good choice for speed control 1) The controller uses position measurement rather than
speed measurement.
Manuscript received March 18, 2020; revised July 14, 2020; accepted
September 7, 2020. Manuscript received in final form September 22, 2020. 2) The controller is easier to implement when compared
This work was supported in part by Qassim University and in part by the with sliding-mode, fuzzy, or model predictive control.
National Science Foundation under Grant ECCS-1128476. Recommended by 3) The disturbance is not assumed to be constant, slowly
Associate Editor A. Serrani. (Corresponding author: Abdullah A. Alfehaid.)
The authors are with the Department of Electrical and Computer Engineer- varying, or generated by a known model.
ing, Michigan State University, East Lansing, MI 48824-1226 USA (e-mail: 4) The observer is simple to implement because it is based
[email protected]; [email protected]; [email protected]). on an extended-state model.
Color versions of one or more figures in this article are available at
https://doi.org/10.1109/TCST.2020.3026569. 5) The observer’s order is the lowest possible because it
Digital Object Identifier 10.1109/TCST.2020.3026569 is based on a reduced-order model that accounts only
1063-6536 © 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: Auckland University of Technology. Downloaded on December 24,2020 at 17:10:29 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

2 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY

for the mechanical dynamics and because the state is

extended by only one variable.
6) The control method shapes the transient response as well
as the steady-state behavior.
7) Theoretical justification is established by rigorous analy-
sis of the closed-loop system.
This work is an extension of our preliminary work [17]. The
contributions over [17] are more details of the development of
the control method, new experimental results that include com-
parison with the cascaded PI controller, and, most importantly,
theoretical justification of the method by rigorous analysis of
the closed-loop system.

II. M ATHEMATICAL M ODEL

The mathematical model of a surface-mounted PMSM in
the synchronous reference frame, the d–q coordinates, is pre- Fig. 1. Block diagram of the proposed control algorithm.
scribed by the following system of equations [18, Ch.9]: A. Current Loops
di d The fast-current loops are made possible by the smallness
L = −Ri d + n p Lωi q + u d (1)
dt of the electrical time constant τe = (L/R) and the use of PI
di q controllers for i d and i q . The current PI controllers are used
L = −Ri q − n p Lωi d − km ω + u q (2) here to provide means to regulate the currents and to further
dt
dω increase the time-scale separation between the dynamics of
J = km i q − Bω − TL (3)
dt the electrical and mechanical subsystems. The design of the
dθ current PI controllers starts by defining the current tracking
=ω (4)
dt errors as
where i d is the direct current, i q is the quadrature current, u d is ed = i dref − i d (5)
the direct input voltage, u q is the quadrature input voltage, ω eq = i qref − i q (6)
is the mechanical rotor speed, θ is the rotor position, TL is the
external load to the motor, R is the stator winding resistance, where ed and eq are the direct and quadrature errors, i dref and
L is the stator inductance, n p is the number of pole pairs, km i qref are the direct and quadrature current reference signals,
is the torque constant, B is the coefficient of viscous friction, where i dref is constant. The control inputs u d and u q are chosen
and J is the moment of inertia of the rotor. These equations as
demonstrate the electrical and mechanical relationships that u d = k p ed + x d , u q = k p eq + x q (7)
govern the dynamics of the PMSM. In (3), it is clear that i q
is the only torque producing current. Hence, the direct current with
 t
i d is usually regulated to zero to increase the efficiency of
x d = ki ed (t´)d t´ (8)
the system. However, the forthcoming analysis allows for a
0 t
constant reference for i d that could be different than zero.
x q = ki eq (t´)d t´ (9)
0
III. C ONTROL S TRATEGY
where t´ is a dummy integration variable, kp is the proportional
The goal is to design a feedback controller that can: gain, and ki is the integral gain. Substituting u d and u q
1) Regulate the speed of the PMSM to a reference signal into (1) and (2) and using (5) and (6), we obtain the following
ωref in the presence of both bounded external load TL equations:
and parameters uncertainty. ded R 1
2) Shape the transient response of the speed. τ = i d − ed − τ n p ω(i qref − eq ) − xd
dt (R + kp ) ref (R + kp )
The proposed control algorithm consists of three main (10)
parts, fast inner current loops, speed and disturbance estima- deq di qref R km
tion using the measured position via an extended high-gain τ =τ + i q − eq + ω
dt dt (R + kp ) ref (R + kp )
observer, and speed shaping and regulation via feedback
1
linearization. The block diagram of the control algorithm is + τ n p ω(i dref − ed ) − xq (11)
shown in Fig. 1. The inner current loops will be designed (R + kp )
purposely to be fast which will further increase the time-scale where τ = (L/(R + kp )) is the time constant of the current
separation between the electrical and mechanical subsystems. tracking errors; τ can be made much smaller than τe by
This allows the system to be reduced via singular perturba- increasing the proportional gain kp of the PI controller. This
tion theory. As a result, a reduced-order extended high-gain will help in reducing the model of the PMSM and thus
observer is achieved, which is easier to implement. reducing the order of the extended high-gain observer.

Authorized licensed use limited to: Auckland University of Technology. Downloaded on December 24,2020 at 17:10:29 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

ALFEHAID et al.: SPEED CONTROL OF PMSM WITH UNCERTAIN PARAMETERS AND UNKNOWN DISTURBANCE 3

B. PMSM Model Reduction The closed-loop system formed of (15), (16), (19), and (21)
Now, by the proper choice of the current controller gains kp is given by
and ki , ed , and eq are made fast and they reach quasi-steady- ξ̇ = A1 ξ (22)
state much faster than other state variables in the system. This
induces a two-time-scale system, with fast and slow dynamics, where ξ = [x d − Ri dref , x q − x̄ q , eω ]T ,
x̄ q is the steady sate of
which gives us an advantage and invites the use of the singular x q and satisfies the equation
   
perturbation method [19] to reduce the model and then design d x̄ q ki dωref
the extended high-gain observer. = R ψ̄ 0, x̄ q , ωref , , TL +km ωref − x̄ q
dt (R+kp) dt
The quasi-steady-state of the fast variables ed and eq ,
obtained by setting τ = 0 in (10) and (11), is and
⎡ ⎤
ki
⎢− R + k p
1 0 0
ed = (Ri dref − x d ) (12) ⎥
R + kp ⎢   ⎥
⎢ ⎥
A1 = ⎢ ki ki B ⎥.
1 ⎢ 0 − − 2
km − R J kω − ⎥
eq = (Ri qref + km ω − x q ). (13) ⎣ kp km kp J ⎦
R + kp
0 0 −kω
Substituting (6) into (3), we obtain the equation
(23)
dω km B 1
= (i qref − eq ) − ω − TL (14) The matrix A1 is Hurwitz.
dt J J J
where i qref is viewed as the control input. Now, we use (8),
D. Feedback Linearization Under Output Feedback
(9), (12)–(14) to arrive at the following slow dynamics of the
system: Since the external load TL is assumed unknown and only the
nominal parameters of the PMSM are assumed to be known,
dxd ki
= (Ri dref − x d ) (15) (17) is rewritten as
dt R + kp
dω
dxq ki = α̂i qref − γ̂ ω + μ̂x q + σ (24)
= (Ri qref + km ω − x q ) (16) dt
dt R + kp
where α̂, γ̂ , and μ̂ are the nominal values of α, γ , and μ, and
dω 1
= αi qref − γ ω + μx q − TL (17) the disturbance σ is defined by
dt J
1
where σ = (α − α̂)i qref − (γ − γ̂ )ω + (μ − μ̂)x q − TL .
2
J
km kp km B km
α= , γ = + , and μ = . The assumption here is that the rotor position is directly
J (R+kp) J (R+kp) J J (R+kp) measured and the speed of the PMSM is not available for
measurement. Therefore, the measured rotor position is used
C. Feedback Linearization Under State Feedback to drive an extended high-gain observer that estimates both
The method of feedback linearization is used here to regu- the speed ω and the disturbance σ . The extended high-gain
late the speed of the PMSM to a reference signal ωref . It also observer, formed using (4) and (24), is given by
provides means to shape the transient response of the speed. d θ̂ ρ1
The speed tracking error is defined as = ω̂ + (θ − θ̂ ) (25)
dt ε
eω = ωref − ω (18) d ω̂ ρ2
= α̂i qref − γ̂ ω̂ + μ̂x q + σ̂ + 2 (θ − θ̂ ) (26)
dt ε
where ωref is the speed reference signal. Using (17) and (18), d σ̂ ρ3
we obtain the following: = 3 (θ − θ̂ ) (27)
dt ε
deω dωref 1
= + γ ωref − αi qref − γ eω − μx q + TL . (19) where θ̂ , ω̂, and σ̂ are the estimates of θ , ω, and σ , respec-
dt dt J tively, ρ1 , ρ2 , and ρ3 are chosen such that the roots of the
It is desired to match the transient response of the speed polynomial
tracking error to that of the target system
s 3 + ρ1 s 2 + ρ2 s + ρ3 (28)
deω
= −kω eω (20) are real and negative, and ε > 0 is a small parameter which
dt
controls the observer’s speed of convergence. If we have not
where kω > 0. If the speed ω were available for measurement
used the singular perturbation method to reduce the model, the
and the external load TL were exactly known, the state
order of the extended high-gain observer would have been 4,
feedback control law would have been given by
  which would be harder to implement.
1 dωref 1 The speed tracking error, which is formed using (18)
i qref = + γ ωref + (kω − γ )eω − μx q + TL
α dt J and (24), is given by
 
dωref deω dωref
 ψ̄ eω , x q , ωref , , TL . (21) = + γ̂ ωref − α̂i qref − γ̂ eω − μ̂x q − σ.
dt dt dt

Authorized licensed use limited to: Auckland University of Technology. Downloaded on December 24,2020 at 17:10:29 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

4 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY

Since the objective here is to drive the speed tracking error taken as
trajectory to match that of the target system (20), the output
ξ1 = x d − Ri dref , ξ2 = x q − x̄ q , ξ3 = eω
feedback control law is taken as
  1 1
1 dωref η1 = 2 (θ − θ̂), η2 = (ω − ω̂)
i qref = + γ̂ ωref + (kω − γ̂ )êω − μ̂x q − σ̂ ε ε
α̂ dt ⎛  ⎞
  ψ êω , x q , σ̂ , ωref , dωdtref
 ψ êω , x q , σ̂ , ωref ,
dωref
(29) η3 = (α − α̂)Mφε ⎝ ⎠
dt M
1
where êω is the speed tracking error estimate, defined by êω = − (γ − γ̂ )ω + (μ − μ̂)x q − TL − σ̂
ωref − ω̂. It can be verified that J
1
    z1 = id − (kpi dref + x d )
dωref dωref R + kp
ψ eω , x q , σ, ωref , = ψ̄ eω , x q , ωref , , TL . ⎡ ⎛  ⎞
dt dt ψ ê , x , σ̂ , ω , dωref
1 ω q ref dt
(30) z2 = iq − ⎣kp Mφε ⎝ ⎠
R + kp M
To protect the system from the peaking phenomenon of high- ⎤
gain observers [19], the control law (29) is saturated outside
the compact set 1 = {ξ T P1 ξ ≤ c1 }, where P1 = P1T > 0 is − km ω + x q ⎦
the solution of the Lyapunov equation P1 A1 + A1T P 1 = −Q 1
⎡⎤ ⎡ ⎤
for some Q 1 = Q 1T > 0, c1 > 0 is chosen such that ξ(0) is ξ1 η1  
z
in the interior of 1 , and ξ = ⎣ ξ2 ⎦ , η = ⎣ η2 ⎦ , z = 1
  z2
ξ3 η3
dωref
max ψ̄ eω , x q , ωref , , TL < i qmax (31) where φε is an odd function defined by
ξ∈ 1 dt ⎧
⎪
⎪ y, for 0 ≤ y ≤ 1
where i qmax is the limit of |i q |. The above inequality is possible ⎪
⎨ y−1 y2 − 1
if φε (y) = y + − , for 1 ≤ y ≤ 1 + ε
  ⎪
⎪ ε ε 2ε
dωref ⎪1 + ,
⎩ for y ≥ 1 + ε
ψ̄ 0, x̄ q , ωref , , TL < i qmax 2
dt
which is a continuously differentiable nondecreasing function
which restricts ωref and TL . The control law is then given by with a locally Lipschitz derivative and bounded uniformly in
⎛  ⎞ ε on any bounded interval of ε. Furthermore, the function φε
ψ êω , x q , σ̂ , ωref , dωdtref satisfies |sat(y) − φ ε (y)| ≤ ε/2 and |φε (y)| ≤ 1∀y ∈ R.
i qref = Msat ⎝ ⎠ (32)
M For ξ ∈ 1 , the closed-loop system is described by
ξ̇ = A1 ξ + E 1 z + B1 f 1 (·)+ε B2 f 2 (·) (34)
where M = i qmax .
εη̇ = A2 η − B3 ρ3 η1 + ε[B3 f 3 (·) + B4 f 4 (·)] + B4 bz 2 (35)
τ ρ3  kp
τ ż = −z − φ (·)B5 η1 − ε B5 f 2 (·) + τ g(·) (36)
IV. C LOSED -L OOP A NALYSIS ε α̂ ε R + kp
Theorem 1: Consider the closed-loop system formed of where A1 is defined by (23)
the PMSM model (1)–(4), the PI current controllers (7), ⎡ ⎤ ⎡ ⎤
−ki 0 0
the extended high-gain observer (25)–(27), and the speed ⎢ 0 −ki ⎥ 1 ⎢ ⎥
controller (32). Assume that E1 = ⎢ ⎥ , B1 = ⎢ ki R ⎥
⎣ km ⎦ R + kp ⎣ km kp ⎦
1) ωref , (dωref /dt), and (d 2 ωref /dt 2 ) are bounded; 0 − −
2) TL and (d TL /dt) are bounded; ⎡ J ⎤ J
0
3) |((α − α̂)/α̂)| < 1; kp ⎢ ⎥
4) ξ(0) is in the interior of 1 and the initial states i d (0), B2 = ⎢ −k i ⎥
R + kp ⎣ km ⎦
i q (0), θ̂ (0), ω̂(0), and σ̂ (0) are bounded. −
⎡ J⎤ ⎡ ⎤ ⎡ ⎤
Then, there exist positive constants λ1 and λ2 such that for
−ρ1 1 0 0 0
all ε ≤ λ1 and τ/ε ≤ λ2 , the trajectories of the closed-loop
A2 = ⎣ −ρ 2 0 1 ⎦ , B3 = ⎣ 0 ⎦ , B4 = ⎣ 1 ⎦
system are bounded for all t ≥ 0, and
−ρ3 0 0 1 0
τ  
|eω (t) − eω (t)| → 0 as ε → 0 and → 0 for all t ≥ 0. B5 =
0
, b=
k m
ε 1 J
(33) ⎛  ⎞
  ψ ê , x , σ̂ , ω , dωref
g (·) α− α̂  ⎝ ω q ref dt
⎠.
Proof: The closed-loop system is represented as a three- g(·) = 1 , and  = φε
g2 (·) α̂ M
time-scale singularly perturbed system. The state variables are

Authorized licensed use limited to: Auckland University of Technology. Downloaded on December 24,2020 at 17:10:29 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

ALFEHAID et al.: SPEED CONTROL OF PMSM WITH UNCERTAIN PARAMETERS AND UNKNOWN DISTURBANCE 5

The functions f 1 and f 2 are globally bounded, and f 1 vanishes ωref and TL are constant, it can be shown that the system has
at η = 0. The functions f 3 and g(·) satisfy a bound of the form an equilibrium point at (ξ̄ , η̄, z̄), where
ka + kb η + kc z, while f 4 satisfies a bound of the form
ka + kb η where ka , kb , and kc are positive constants. ξ̄1 = −τ (R + kp )n p ωref (Bωref + TL )/km
For τ ε 1, the system (34)–(36) is singularly ξ̄2 = τ (R + kp )n p ωref i dref , ξ̄3 = η̄1 = η̄2 = 0
perturbed with three-time-scales. In the fastest time-scale, the η̄3 = τ km n p ωref i dref /J
boundary layer model is τ ż = −z, which is obtained by setting z̄ 1 = τ n p ωref (Bωref + TL )/km , and z̄ 2 = −τ n p ωref i dref .
(τ/ε) = 0, τ = 0, and ε = 0 on the right-hand side
of (36). The quasi-steady-state of this model is z = 0. In the At this equilibrium point eω = ξ3 = 0. We now show that the
intermediate time-scale, the boundary layer model is equilibrium point is exponentially stable and every trajectory
in converges to it as time tends to infinity. Toward that end,
εη̇ = A2 η − B3 ρ3 η1 (37)
we shift the equilibrium point to the origin by the change of
which is obtained by setting ε = 0 and z = 0 on the right-hand variables
side of the η̇-equation. This system can be represented as a
ξ̃ = ξ − ξ̄ , η̃ = η − η̄, z̃ = z − z̄.
negative feedback connection of the transfer function
ρ3 The transformed system is given by
(εs) =
(εs) + ρ1 (εs)2 + ρ2 (εs) + ρ3
3
ξ̃˙ = A1 ξ̃ + E 1 z̃ + B1 f˜1 (·)
and the time-varying gain (·). Since |((α − α̂)/α̂)| < 1
εη̃˙ = H η̃ + B4 b z̃ 2 + ε[B3 f˜3 (·) + B4 f˜4 (·)]
and |φε (·)| ≤ 1, || < 1. Because the poles of the transfer τ ρ3
function (εs) are real and negative, maxω |( j εω)| ≤ 1. The τ z̃˙ = −z̃ + B5 η̃1 + τ g̃(·)
ε α̂
circle criterion [19] shows that the origin of (37) is globally
exponentially stable. By applying a loop transformation and where H = A2 − ρ3 ((α − α̂)/α̂)B3 C3 , in which C3 = [1 0 0].
using the Kalman–Yakubovich–Popov lemma [19, Lemma As shown in the proof of Theorem 1, H is Hurwitz for
6.3], we obtain a quadratic Lyapunov function V2 (η) = η T P2 η |((α − α̂)/α̂)| < 1. The functions f˜ and g̃ are obtained
whose derivative with respect to (37) is bounded from above from f and g by subtracting their values at the equilibrium
by −(λ/ε)V2 (η) for some positive constant λ, independent of point. A composite Lyapunov function for this three-time-scale
ε. In the slowest time-scale, the slow model is ξ̇ = A1 ξ , which singularly perturbed system is taken as
is obtained by setting ε = 0 and η = 0 on the right-hand side 1
of (34). V = ξ̃ T P1 ξ̃ + η̃ T P3 η̃ + z̃ T z̃
2
Consider the set
where P3 is the positive definite symmetric solution of the
= 1 × {W (η, z) ≤ ε2 c2 } Lyapunov equation P3 H + H T P3 = −I . The time derivative
of V satisfies the inequality
where W (η, z) = η T P2 η + (1/2)z T z. Similar to arguments ⎡ ⎤
used in the analysis of high-gain observer [19], it can be shown k1 − k2 − k3
⎡ ⎤T     ⎡ ⎤
that, by choosing c2 > 0 large enough, the set is positively ξ̃  ⎢ ⎢−k2 1 k6 ⎥⎥ ξ̃ 
− k − k +
invariant for sufficiently small ε and (τ/ε). This is done by V̇ ≤ − ⎣η̃⎦ ⎢ ⎢ ε
4 5
ε ⎥ ⎣η̃⎦
⎣     ⎥
⎦
showing that V̇1 < 0 on the boundary of 1 , and Ẇ < 0 on z̃ k6 1 z̃
the boundary W (η, z) = ε2 c2 . −k3 − k5 + − k7
ε τ
At the initial time, (η(0), z(0)) could be outside the set
{W (η, z) ≤ ε2 c2 } but would move rapidly toward the where k1 to k7 are positive constants independent of τ and
set and will reach it within an interval [0, T (ε)], where ε. The matrix of this quadratic form is positive definite for
limε→0 T (ε) = 0. Because the initial state ξ(0) is in the interior sufficiently small ε and (τ/ε). Hence, the origin is exponen-
of 1 , choosing ε small enough ensures that ξ do not leave tially stable. Moreover, the forgoing inequality is valid in ;
hence, all trajectories in converge to the origin as t tends
1 , and by the end of this interval (ξ, η, z) would be in the
positively invariant set . to infinity.
The limit (33) follows from the continuous dependence
V. C ASCADED PI S PEED C ONTROLLER
of the solutions of differential equations on parameters [19,
Theorem 9.1] and exponential stability of the subsystem Since this brief paper introduces a new method to control
ξ̇ = A1 ξ . the speed of PMSMs, it is natural to compare it with the
Theorem 2: Under the assumptions of Theorem 1, sup- most widely used speed control method in industrial and
pose ωref and TL are constant. Then, there exist positive commercial electrical drive systems.
constants λ3 and λ4 such that for ε < λ3 and (τ/ε) < Nearly all speed control methods of electrical drive systems
λ4 , lim t →∞ eω (t) = 0. that are used currently in the industry utilize cascaded PI
Proof: We have already shown in the proof of Theo- controllers [3]. There are several versions and variations of the
rem 1 that the trajectories of the system enter the positively cascaded PI speed control of electric motors that can be found
 
η in [1]–[3] and [20]. Here, we have chosen to compare our
invariant set wherein   = O(ε). Inside , the saturation method with the versions of [3] and [20] since they are most
z
is not active; hence, Mφε (ψ/M) = Msat(ψ/M) = ψ. When recent and both claim that their method is the most used speed

Authorized licensed use limited to: Auckland University of Technology. Downloaded on December 24,2020 at 17:10:29 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

6 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY

TABLE I
PARAMETERS OF PMSM

Fig. 2. Block diagram of the PI speed controller. conducted to show the speed tracking ability of the proposed
controller. The second experiment is performed to test the
control method in electrical drive systems. The main difference robustness of the proposed controller against external load.
between the versions of the PI speed controller in [3] and [20], For comparison purposes, these experiments were exactly
and the versions presented in [1] and [2] is that in [3] and [20] repeated using the cascaded PI speed controller. The nominal
the cross-coupling between the current equations is canceled parameters of the used surface-mounted PMSM are shown in
by using nonlinear feedback through the control inputs u d and Table I.
u q leaving the mathematical model of the PMSM linear. The PI controllers of the inner current loops are designed
All of these references except [20] assumed the speed is such that the currents i d and i q are relatively fast. The
directly measured when presenting the cascaded PI controller proportional and integral gains of the PI current controllers
and a speed observer was not discussed. In [20], a speed that are found to satisfy the criterion above are: kp = 20 and
estimator is described to have a position differentiator in series ki = 2500. In all cases, i dref = 0.
with a low-pass filter. Thus, the speed estimator constitutes a There are two aspects that must be considered when
high-pass filter from the measured position to the estimated designing the parameter ε of the extended high-gain observer.
speed. Fig. 2 shows the complete cascaded PI speed controller First, the assumption τ ε 1 must hold. Second, ε
that is used here for comparison. should be chosen so that the best compromise between fast
The development of the cascaded PI controller is straight- convergence and minimal noise amplification is achieved. With
forward. The direct and quadrature voltages are given by τ = 2.15 ∗ 10−4 , the value of ε that satisfies both criteria was
u d = kp ed + x d − n p L ω̂i q found to be 0.001. In addition, the roots of the polynomial (28)
of the extended high-gain observer are all assigned at −1.
u q = kp eq + x q + n p L ω̂i d + km ω̂
Therefore, ρ1 = 3, ρ2 = 3, and ρ3 = 1. The gain kω = 60.
where ed , eq , x d , and x q are defined as in (5)–(9), respectively. The disturbance estimate was passed through a first-order low-
The quadrature current reference i qref is the output of the PI pass filter before using it in the controller.
speed control loop and is given by There were over 80 experiment runs conducted to fine-tune
 t the cascaded PI speed controller. Only about 30% of these
i qref = h p (ωref − ω̂) + h i (ωref (t´) − ω̂(t´))d t´ experiment runs were considered. The remaining 70% were
0
discarded because they were too oscillatory, they have slow
where h p and h i are the proportional and integral gains of responses, or an unacceptable performance when an external
the PI speed controller, respectively. Furthermore, the speed load was applied. Only three pairs of the speed PI gains
estimate ω̂ is defined by were selected out of the considered 30%. They were selected
s based on overshoot and disturbance rejection, so, for example,
ω̂ = [θ ]
hos + 1 out of all transient responses that have an overshoot ≤2%
where h o is the speed estimator time constant. It can be seen we selected the one that has the best disturbance rejection.
that decreasing h o decreases the simultaneous speed estimation A summary of the selected tuning gains is shown in Table II.
error but on the other hand increases quantization noise of the The estimator time constant h o = 0.0032 was found to provide
optical encoder. fast convergence and an acceptable noise amplification.
It should be noted that the decoupling of the current equa- The control algorithm is implemented using the National
tions is exact only when the true parameters of the machine Instruments’ Real-Time system with 10 kHz sampling fre-
are accurately known. quency. Hall effect sensors are used to measure the phase
currents of the motor. The position of the motor’s shaft is
VI. E XPERIMENTAL R ESULTS measured with a 2500PPR incremental encoder that is directly
The performance of the proposed control method is eval- mounted at the end of the motor’s shaft. The other end of the
uated through two experiments. The first experiment is shaft is directly connected to an induction motor with a shaft

Authorized licensed use limited to: Auckland University of Technology. Downloaded on December 24,2020 at 17:10:29 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

ALFEHAID et al.: SPEED CONTROL OF PMSM WITH UNCERTAIN PARAMETERS AND UNKNOWN DISTURBANCE 7

TABLE II
PI S PEED C ONTROLLER T UNING PARAMETERS

Fig. 4. (a) Experimental speed of PMSM. (b) Experimental speed deviation

from reference speed.

Fig. 3. Photograph of the experimental setup.

coupler. The induction motor is used to apply external load

to the PMSM. All the reported speeds from the experiment
are estimated using a separate high-gain observer that is
independent of the feedback loop. Fig. 3 shows a photograph
of the experimental setup.

A. Experiment I
In this case, the nominal parameters, which are shown in
Table I, are used in the proposed controller and no external
Fig. 5. (a) Speed of PMSM when the external load was applied. (b) Speed
load is applied to the motor. The motor in this case is at of PMSM when the external load was removed.
standstill when the following speed profile is applied:
⎧
⎪ 1 z t2,
⎪ for 0 ≤ t < t1
B. Experiment II
⎪
⎪ 1
⎪2
⎪ This case investigates the effectiveness of both the proposed
⎨a f t + C 1 for t1 ≤ t < t2
ωref = control method and the PI speed controller when the PMSM
⎪
⎪ 1 is externally loaded with a step disturbance as well as a
⎪
⎪ − z t 2 + z 2 t + C2 , for t2 ≤ t < t3
⎪ 2 1
⎪ time-varying disturbance. The PMSM, in both of these cases,
⎩
C3 , for t3 ≤ t is regulated at a constant speed of 100 rad/s. Then, the external
where z 1 = 310 719, z 2 = 21 554, a f = 1554, C1 = −3.884, load is applied.
C2 = −647.6, C3 = 100, t1 = 0.005, t2 = 0.0644, and Fig. 5(a) shows the speed of the motor ω when the proposed
t3 = 0.0694. The speed profile is designed to be relatively controller is used, and the speed of the motor ωPI when the PI
fast and the maximum required acceleration does not exceed speed controller is used with different pairs of PI gains when
the motor’s capability. the load is applied to the PMSM. The load is a step of about
Fig. 4(a) shows the speed reference signal ωref , the speed 2N · m which was applied at t = 10 s and removed at t =
of the motor ω when the proposed controller is used, and 20 s. It can be seen that at the moment the load was applied,
the speed of the motor ωPI when the cascaded PI controller the speed of the motor in all cases dropped then recovered
is used with different pairs of PI gains. It can be seen that and the speed was maintained at 100 rad/s. This behavior
the proposed controller was able to control the speed of the is expected from the proposed controller because there is an
motor to closely track the reference speed while the PI speed integral action that results when the proposed controller is
controller was not able to perform as well as the proposed used in the presence of constant disturbance which leads to
controller. Fig. 4(b) shows the speed tracking error between the zero steady-state error. Similarly, the PI speed controller relies
speed reference ωref and the motor speed for both controllers. on the integrator part of the PI controller to yield zero steady-
This figure further shows the performance difference between state error. There is however difference in how much the speed
the proposed controller and the cascaded PI speed controller. drops at the moment the load was applied and also how fast

Authorized licensed use limited to: Auckland University of Technology. Downloaded on December 24,2020 at 17:10:29 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

8 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY

more tuning effort than the proposed control method. Second,

the proposed controller performed much better tracking a
speed reference signal than the PI speed controller. Third, the
proposed controller showed more robustness to external load
than the PI speed controller.
Future extensions of this work will study the field-
weakening problem and will extend the controller to the
sensorless case.

R EFERENCES
Fig. 6. Speed of PMSM when the time-varying external load was applied.
[1] W. Leonhard, Control of Electrical Drives, 3rd, ed. Springer 2001.
[2] R. Krishnan, Electric Motor Drives: Modeling, Analysis, and Control.
the speed is recovered. One can see that the step disturbance Upper Saddle River, NJ, USA: Prentice-Hall, 2001.
caused the speed of the motor to drop about 2.5% when the [3] L. Wang, S. Chai, D. Yoo, L. Gan, and K. Ng, PID and Predic-
tive Control of Electrical Drives and Power Converters Using MAT-
proposed controller is used while it caused the speed to drop LAB/Simulink. Singapore: Wiley, 2015.
about 5% when the PI speed controller is used. It can also be [4] Z. Gao, “On the centrality of disturbance rejection in automatic control,”
seen that the proposed controller was able to recover the speed ISA Trans., vol. 53, no. 4, pp. 850–857, Jul. 2014.
[5] S. Li, J. Yang, W.-H. Chen, and X. Chen, Disturbance Observer-Based
relatively fast while the PI controller requires high integrator Control: Methods and Applications. Boca Raton, FL, USA: CRC Press,
gain to achieve similar recovery time. Similar behavior was 2014.
observed at the moment when the external load was removed [6] L. B. Freidovich and H. K. Khalil, “Performance recovery of feedback-
linearization-based designs,” IEEE Trans. Autom. Control, vol. 53,
which is shown in Fig. 5(b). no. 10, pp. 2324–2334, Nov. 2008.
Fig. 6 shows the speed of the motor ω when the proposed [7] J. Yang, W. Chen, S. Li, L. Guo, and Y. Yan, “Disturbance/uncertainty
controller is used, and the speed of the motor ωPI when the PI estimation and attenuation techniques in PMSM drives—A survey,”
IEEE Trans. Ind. Electron., vol. 64, no. 4, pp. 3273–3285, Apr. 2017.
speed controller is used with different pairs of PI gains when a [8] X. Zhang, L. Sun, K. Zhao, and L. Sun, “Nonlinear speed control for
time-varying load is applied to the PMSM. The load is about PMSM system using sliding-mode control and disturbance compensation
1 + 0.75 sin(50 ∗ (t − 10)) which was applied at t = 10 s. techniques,” IEEE Trans. Power Electron., vol. 28, no. 3, pp. 1358–1365,
Mar. 2013.
It can be seen that the proposed controller was able to keep [9] Y. Wang, Y. Feng, X. Zhang, and J. Liang, “A new reaching law
the peak-to-peak steady-state error less than 3.5 rad/s while the for antidisturbance sliding-mode control of PMSM speed regulation
smallest peak-to-peak steady-state error resulting from using system,” IEEE Trans. Power Electron., vol. 35, no. 4, pp. 4117–4126,
Apr. 2020.
the PI speed controller is about 5.5 rad/s. [10] A. K. Junejo, W. Xu, C. Mu, M. M. Ismail, and Y. Liu, “Adaptive
Both experiments show the effectiveness of the proposed speed control of PMSM drive system based a new sliding-mode reaching
control method in the presence of external disturbance. The law,” IEEE Trans. Power Electron., vol. 35, no. 11, pp. 12110–12121,
Nov. 2020, doi: 10.1109/TPEL.2020.2986893.
robustness of the proposed control method against external dis- [11] R. Errouissi, M. Ouhrouche, W.-H. Chen, and A. M. Trzynadlowski,
turbance is due to estimating the disturbance and canceling it “Robust cascaded nonlinear predictive control of a permanent magnet
in the control law. On the other hand, the PI speed controller’s synchronous motor with antiwindup compensator,” IEEE Trans. Ind.
Electron., vol. 59, no. 8, pp. 3078–3088, Aug. 2012.
robustness depends on dominating the effect of disturbance [12] X. Zhang, Y. Cheng, Z. Zhao, and Y. He, “Robust model predictive direct
and thus requiring very high controller gains which affect the speed control for SPMSM drives based on full parameter disturbances
transient response. This poses a compromise between transient and load observer,” IEEE Trans. Power Electron., vol. 35, no. 8,
pp. 8361–8373, Aug. 2020.
response and robustness against external load which does not [13] H. Sira-Ramirez, J. Linares-Flores, C. Garcia-Rodriguez, and
exist in the proposed controller. M. A. Contreras-Ordaz, “On the control of the permanent magnet
synchronous motor: An active disturbance rejection control approach,”
VII. C ONCLUSION IEEE Trans. Control Syst. Technol., vol. 22, no. 5, pp. 2056–2063,
Sep. 2014.
A high-performance control method has been introduced to [14] C. Xia, N. Liu, Z. Zhou, Y. Yan, and T. Shi, “Steady-state performance
regulate and shape the transient response of the speed of a improvement for LQR-based PMSM drives,” IEEE Trans. Power Elec-
tron., vol. 33, no. 12, pp. 10622–10632, Dec. 2018.
PMSM to match that of a target system. We use an extended [15] H. H. Choi, N. T. Vu, and J. Jung, “Design and implementation of a
high-gain observer, which is driven by the measured rotor Takagi–Sugeno fuzzy speed regulator for a permanent magnet synchro-
position, to estimate both the speed of the motor and the nous motor,” IEEE Trans. Ind. Electron., vol. 59, no. 8, pp. 3069–3077,
Aug. 2012.
disturbance. Then, these estimates are used in a feedback [16] Y. Yan, J. Yang, Z. Sun, C. Zhang, S. Li, and H. Yu, “Robust speed regu-
linearization law to shape and regulate the speed of the lation for PMSM servo system with multiple sources of disturbances via
motor. The extended high-gain observer is designed based an augmented disturbance observer,” IEEE/ASME Trans. Mechatronics,
vol. 23, no. 2, pp. 769–780, Apr. 2018.
on a reduced model of the system. The model is reduced by [17] A. A. Alfehaid, E. G. Strangas, and H. K. Khalil, “Speed control
creating fast-current loops which allow us to utilize singular of permanent magnet synchronous motor using extended high-gain
perturbation theory. observer,” in Proc. Amer. Control Conf. (ACC), Boston, MA, USA,
Jul. 2016, pp. 2205–2210.
Performance and robustness of the proposed control method [18] J. Chiasson, Modeling and high-Performance Control of Electric
are confirmed by experimentally comparing it to the PI speed Machines. Hoboken, NJ, USA: Wiley, 2005.
controller. The experimental results showed that the proposed [19] H. K. Khalil, Nonlinear Systems. Upper Saddle River, NJ, USA:
Prentice-Hall, 2002.
control method is superior to the PI speed controller in three [20] N. P. Quang and J.-A. Dittrich, Vector Control of Three-Phase AC
different ways. First, the PI speed controller required much Machines, 2nd ed. Berlin, Germany: Springer-Verlag, 2015.

Authorized licensed use limited to: Auckland University of Technology. Downloaded on December 24,2020 at 17:10:29 UTC from IEEE Xplore. Restrictions apply.