This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication.

Digital Implementation of an Adaptive Speed

Regulator for a PMSM
Han Ho Choi, Member, IEEE, Nga Thi-Thuy Vu, and Jin-Woo Jung, Member, IEEE

Abstract

We design an adaptive speed regulator for a permanent magnet synchronous motor (PMSM). The proposed
adaptive regulator does not require any information on the PMSM parameter and load torque values, thus it is
insensitive to model parameter and load torque variations. We implement the proposed adaptive speed regulator
system by using a TMS320F28335 floating point DSP. We give simulation and experimental results to verify that
our method can be successfully used to control a PMSM under model parameter and load torque variations.

Key Words- permanent magnet synchronous motor (PMSM), adaptive control system, disturbance, uncertainty.

I. I NTRODUCTION

A permanent magnet synchronous motor (PMSM) has been widely employed in servo applications such as
chip mount machines, robotics, and hard disk drives because it features low noise, low inertia, high efficiency,
and low maintenance cost. However, a PMSM can not be easily controlled because of the uncertainties such as
parameter variations and load torque variations. Therefore, the linear control methods such as PID control can
not guarantee high performance. To solve this problem, many researchers have proposed various design methods,
e.g., adaptive control [1]-[4], nonlinear feedback linearization control [5], fuzzy control [6]-[7]. Recently, several
authors [8]-[10] have proposed disturbance-observer-based PMSM control methods that can effectively suppress
load torque variations. However, most of the previous PMSM control design methods cannot guarantee stability
and convergence of speed error responses under inexact information on the PMSM parameters such as the stator
resistance, the stator inductance, the rotor inertia, the viscous friction coefficient, the magnetic flux, etc. Considering
these facts, we propose an adaptive control law design method for a PMSM. We first design an adaptive speed
regulator which does not require any information on the PMSM parameter and load torque values. We also prove that
the speed error of the closed-loop system converges to zero and the parameter adaptation error signal is bounded.

This research was supported by Basic Science Research Program through the National Research Foundation of Korea(NRF) funded by
the Ministry of Education, Science and Technology (No. 2010-0015428). This was also supported by Dongguk Research Fund.
The authors are with the Division of Electronics and Electrical Engineering, Dongguk University - Seoul, 26, 3-ga, Pil-dong, Chung-gu,
Seoul 100-715, SEOUL KOREA (e-mail: hhchoi@dongguk.edu, vunga tdh2k45@yahoo.com, jinwjung@dongguk.edu)

Copyright (c) 2010 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

Finally, simulation and experimental results are shown to verify that the proposed method can precisely control the
speed of a PMSM under model parameter and load torque variations.

II. M ODEL D ESCRIPTION

By taking the rotor coordinates of the motor as reference coordinates, a surface mounted PMSM can be represented
by the following nonlinear equation:


 ω̇ = k1 iqs − k2 ω − k3 TL


i̇qs = −k4 iqs − k5 ω − ωids + k6 Vqs + dq (1)




i̇ds = −k4 ids + ωiqs + k6 Vds + dd
where TL represents the load torque, ω is the electrical rotor angular speed, iqs is the q axis current, Vqs is the
q axis voltage, ids is the d axis current, Vds is the d axis voltage, dq and dd represent disturbance inputs, and
ki > 0, i = 1, · · · , 6 are the parameter values depending on the number of poles, the stator resistance, the stator
inductance, the rotor inertia, the viscous friction coefficient, the magnetic flux. The load torque disturbance term TL
and uncertainties on the parameters ki can severely deteriorate the control performance if they are not appropriately
accounted for.
We will use the following assumptions:

A1: ω , iqs , ids are available.

A2: The desired speed ωd is constant and ω̇d = ω̈d = 0.
A3: TL is unknown, ṪL can be set as ṪL = 0.
A4: ki are not known accurately, k̇i can be set as k̇i = 0.
A5: dq and dd are unknown, d˙q and d˙d can be set as d˙q = d˙d = 0.

Denote the electrical rotor angular acceleration ω̇ by β = k1 iqs − k2 ω − k3 TL = ω̇ . Then, by introducing the
speed error ωe = ω − ωd and by denoting Vqf and Vdf as



1
Vqf = (k4 iqs + k5 ω + ωids − dq )



 k6
1
+ [k2 − γq ]β (2)

 k1 k6

 1

 Vdf = (k4 ids − ωiqs − dd )
k6
we can obtain the following error dynamics


 ω̇e = β


β̇ = −γq β + k1 k6 (Vqs − Vqf ) (3)




i̇ds = k6 (Vds − Vdf )
where γq > 0 is a constant design parameter.

Copyright (c) 2010 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

After all, our design problem can be formulated as designing an adaptive control law [Vqs , Vds ]T for the above
error dynamics (3).
Before proceeding further, we give a background result which will be used to derive our main results.
Lemma 1: There exist constant parameter vectors ξq∗ = [ξq1
∗ , ξ ∗ , ξ ∗ , ξ ∗ ]T and ξ ∗ = [ξ ∗ , ξ ∗ , ξ ∗ ]T such that
q2 q3 q4 d d1 d2 d3

∑
4 ∑
3
hTq ξq∗ = ∗
hqi ξqi = Vqf , hTd ξd∗ = ∗
hdi ξdi = Vdf (4)
i=1 i=1

where hq = [hq1 , hq2 , hq3 , hq4 ]T = [ω, iqs , ωids , 1]T and hd = [hd1 , hd2 , hd3 ]T = [ids , ωiqs , 1]T .
Proof : The assumptions A3-5 imply that (4) holds with ξd∗ = [k4 , −1, −dd ]T /k6 and
 
 (k2 γq − k2 + k1 k5 )
2

 
 k (k − γ + k ) 
1  1 2 q 4 
ξq∗ =   (5)
k1 k6  
 k1 
 
(γq k3 TL − k2 k3 TL − k1 dq )

▽▽▽

III. A DAPTIVE C ONTROLLER D ESIGN AND S TABILITY A NALYSIS

Theorem 1: Let the control input variables Vqs and Vds be given by the following adaptive control law
∑
4 ∑
3
Vqs = −δq σq + ξqi hqi , Vds = −δd ids + ξdi hdi (6)
i=1 i=1
∫ t ∫ t
1 1
ξqi = − hqi σq dτ, ξdi = − hdi ids dτ (7)
ϕ qi 0 ϕ di 0
∗ and ξ ∗ , σ = γ ω + β , ϕ > 0, ϕ > 0, δ > 0, δ > 0. Then, ω converges
where ξqi and ξdi are estimates of ξqi di q q e qi di q d e

to zero, ξqi and ξdi are bounded.

Proof : Along the similar line of [12], we will show the stability of the adaptive system. Let us define the Lyapunov
∑ ∑3 ∗ ∗
functional as V (t) = σq2 + i2ds + 4i=1 k1 k6 ϕqi ξ˜qi
2 + ˜2 ˜ ˜
i=1 k6 ϕdi ξdi where ξqi = ξqi − ξqi , ξdi = ξdi − ξdi . Its time

derivative along the error dynamics (3) is given by

∑
4 ∑
3
2V̇ = σq σ̇q + ids i̇ds − k1 k6 ϕqi ξ˜qi ξ˙qi − k6 ϕdi ξ˜di ξ˙di (8)
i=1 i=1
On the other hand, (3) implies that

σ̇q = −k1 k6 (Vqf − Vqs ), i̇ds = −k6 (Vdf − Vds ) (9)

Lemma 1, (6), and (7) imply that 


 ∑
4

 V = −δ σ + V − ξ˜qi hqi
 qs q q qf
i=1 (10)

 ∑
3

 ξ˜di hdi
 Vds = −δd ids + Vdf −
i=1

Copyright (c) 2010 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

Fig. 1. Block diagram of the proposed control algorithm.

1 1
ξ˙qi = − hqi σq , ξ˙di = − hdi ids (11)
ϕ qi ϕ di
therefore, (8) can be reduced to
2V̇ ≤ −k1 k6 δq σq2 − δd k6 i2ds ≤ 0 (12)

where ki > 0, δq > 0, and δd > 0 are used. The inequality (12) implies that
∫ ∞ ∫ ∞ ∫ ∞
2 V̇ (τ )dτ ≤ −k1 k6 δq σq dτ − δd k6
2
i2ds dτ (13)
0 0 0

Multiplying (13) by -1 and integrating the left-hand side of (13) gives

∫ ∞ ∫ ∞
k1 k6 δq 2
σq dτ + δd k6 i2ds dτ ≤ 2V (0) (14)
0 0

where V (t) ≥ 0 is used. Because k1 k6 δq > 0 and δd k6 > 0, we can obtain

∫ ∞
σq2 dτ < ∞ (15)
0

which implies that σq ∈ L2 . Since V̇ ≤ 0 as shown in (12), we can see that V (t) is nonincreasing and is upper
bounded as V (t) ≤ V (0). This implies that σq ∈ L∞ , ids ∈ L∞ , ξq ∈ L∞ , ξd ∈ L∞ . Because the transfer function
from σq to ωe , Hωe σq (s), is strictly positive real and σq ∈ L2 , we can use the result of [12] to state that ωe converges
to zero.
▽▽▽

IV. D ISCRETIZED C ONTROLLER

We can rearrange Vqs , Vds as

Vqs = uqs + uqd , Vds = uds + udd , (16)

where uqs , uds are the static terms given by

uqs (t) = −δq γq ωe , uds (t) = −δd ids (17)

Copyright (c) 2010 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

Fig. 2. Block diagram of the proposed PMSM control system.

and uqd (t), udd (t) are the dynamic terms given by

d ∑ 4 ∑ 3
uqd (t) = −δq ωe − hqi ξqi , udd (t) = − ξdi hdi (18)
dt
i=1 i=1

For digital implementation under a sufficiently small sampling time T , the static terms uqs and uds at the sampling
instant kT can be straightforwardly set as

uqs (k) = −δq γq ωe (k), uds (k) = −δd ids (k) (19)

whereas the dynamic terms uqd (k) and udd (k) cannot be straightforwardly obtained. By using the relation ω̇ = β
and the previous result [11], we can compute the derivative term β(k) by using the following recursive equation :
ρ 1
β(k) = β(k − 1) + [ω(k) − ω(k − 1)] (20)
T +ρ T +ρ
where ρ is a sufficiently small filter time constant to limit the susceptibility of the derivative term β = ω̇ to noise.
By using the relation (7) and the previous result [11], we can compute the integral terms ξqi (k) and ξdi (k) by using
the following recursive equation :
T
ξqi (k + 1) = ξqi (k) − hqi (k)[γq ωe (k) + β(k)] (21)
ϕqi
T
ξdi (k + 1) = ξdi (k) − hdi (k)ids (k) (22)
ϕdi

Copyright (c) 2010 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

Fig. 3. Simulation results under no parameter variation.

Fig. 4. Simulation results under 200% variations of some parameters (Rs , Ls , J, TL , and λm ).

And therefore the dynamic terms uqd (k) and udd (k) can be computed by the following equation:


 ∑
4

 u (k) = −δ β(k) + ξqi (k)hqi (k)
 qd q
i=1 (23)

 ∑
3


 udd (k) = ξdi (k)hdi (k)
i=1

where β(k), ξqi (k), ξdi (k) are updated by the recursive equations (20), (21), (22), respectively. Fig. 1 shows overall
block diagram of the proposed control algorithm.

V. S IMULATION AND E XPERIMENT

For simulation and experiment, we consider a PMSM (1) with p = 12, Rs = 0.99[Ω], Ls = 5.82[mH], λm =
0.0791[V · sec/rad], J = 0.00121[kg · m2 ], B = 0.0003[N · m · sec/rad]. With δq = 0.01, δd = 0.001, ϕqi = 2,

Copyright (c) 2010 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

Fig. 5. Simulation results under speed reverse case.

ϕdi = 2, γq = 100, ρ = 0, we can obtain the following control law



 Vqs (k) = −ωe (k) − 0.01 [ω(k) − ω(k − 1)]



 T

 ∑4

+ ξqi (k)hqi (k)
(24)

 i=1

 ∑3



 Vds (k) = −0.001ids (k) +
 ξdi (k)hdi (k)
i=1

where ξqi (k) and ξdi (k) are updated by the following recursive equations

 1

 ξqi (k + 1) = ξqi (k) − hqi (k)ωe (k)

 100
1
− hqi (k)[ω(k) − ω(k − 1)] (25)

 10000T


 ξdi (k + 1) = ξdi (k) − 1 hdi (k)ids (k)
10000
and hq (k) = [ω(k), iqs (k), ω(k)ids (k), 1]T , hd (k) = [ids (k), ω(k)iqs (k), 1]T . Fig. 2 shows the overall block diagram
of the proposed PMSM control system. In simulations and experiments, the switching frequency and the sampling
frequency (1/T ) are selected as 5[kHz], and a space vector PWM (SVPWM) technique is adopted. Figs. 3 and 4
show the simulation results using Matlab/Simulink about two cases : no parameter variation and 200% variations of
some parameters (Rs , Ls , J , TL , λm ). In both cases, the desired motor speed (ωd ) increases from 125.66 [rad/sec]
to 251.32 [rad/sec] and then decreases from 251.32 [rad/sec] to 125.66 [rad/sec]. Fig. 3 shows the simulation
results (ωd , ω , Vqs , Vds , iqs , ids , Van , ia ) under no parameter variation. In Figs. 3 and 4, we can observe that our
adaptive speed controller is very robust to model parameter and load torque variations. Fig. 4 shows the simulation
results under 200% variations of some parameters (Rs , Ls , J , TL , λm ). For further verification, we also consider
the case when the desired speed (ωd ) is reversed. In this case, the desired motor speed (ωd ) goes down from 125.66
[rad/sec] to -125.66 [rad/sec] and then goes up from -125.66 [rad/sec] to 125.66 [rad/sec]. Fig. 5 shows the
simulation results for that case. Fig. 6 shows the experimental results about motor speed, voltage and current under

Copyright (c) 2010 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

(a) (b)

(c) (d)

Fig. 6. Experimental results. (a) ωd and ω. (b) Vqs and Vds . (c) iqs and ids . (d) Van and ia .

the same condition as Fig. 3. Fig. 6 (a) shows the desired speed (ωd ), measured speed (ω ), and Fig. 6 (b) shows
the q -axis and d-axis voltages (Vqs , Vds ). Fig. 6 (c) shows the measured q -axis current (iqs ) and d-axis current
(ids ), and Fig. 6 (d) shows the line to neutral voltage (Van ) and phase a current (ia ). Fig. 7 shows the experimental
results about the speed reverse case.

VI. C ONCLUSION

We proposed an adaptive speed regulator design method for a PMSM. The proposed adaptive speed controller
is robust because it does not depend on the PMSM parameter and load torque values. We also proved that the
speed error of the closed-loop system converges to zero and the parameter adaptation error signal is bounded. Via
various simulation and experimental results, it was clearly proven that the proposed adaptive speed regulator gives
very remarkable speed control performance with no information on the PMSM parameter and load torque values.

Copyright (c) 2010 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

(a) (b)

(c) (d)

Fig. 7. Experimental results (speed reverse case). (a) ωd and ω. (b) Vqs and Vds . (c) iqs and ids . (d) Van and ia .

R EFERENCES

[1] T.-H. Liu, H.-T. Pu, and C.-K. Lin, “Implementation of an adaptive position control system of a permanent-
magnet synchronous motor and its application,” IET Electr. Power Appl., vol. 4, pp. 121-130, 2010
[2] W.-T. Su, and C.-M. Liaw, “Adaptive positioning control for a LPMSM drive based on adapted inverse model
and robust disturbance observer,” IEEE Trans. Power Electronics, vol. 21, pp. 505-517, 2007
[3] Y.A.-R.I. Mohamed, “A hybrid-type variable-structure instantaneous torque control with a robust adaptive torque
observer for a high-performance direct-drive PMSM,” IEEE Trans. Ind. Electron., vol. 54, pp. 2491-2499, 2007
[4] K.-H. Kim, “Model reference adaptive control-based adaptive current control scheme of a PM synchronous
motor with an improved servo performance ,” IET Electr. Power Appl., vol. 3, pp. 8-18, 2009
[5] C.-K. Lin, T.-H. Liu, and S.-H. Yang, “Nonlinear position controller design with input-output linearisation
technique for an interior permanent magnet synchronous motor control system,” IET Power Electron., vol. 1,
pp. 14-26, 2008

Copyright (c) 2010 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

10

[6] M.N. Uddin, and M.A. Rahman, “High-speed control of IPMSM drives using improved fuzzy logic algorithms,”
IEEE Trans. Ind. Electron., vol. 54, no. 1, pp. 190-199, 2007
[7] Y.-S. Kung, C.-C. Huang, and M.-H. Tsai, “FPGA realization of an adaptive fuzzy controller for PMLSM
drive,” IEEE Trans. Ind. Electron., vol. 56, no. 8, pp. 2923-2932, 2009
[8] K.-B. Lee., and F. Blaabjerg, “Robust and stable disturbance observer of servo system for low-speed operation,”
IEEE Trans. Industry Applications, vol. 43, pp. 627-635, 2007
[9] Y. Zhang, C.M. Akujuobi, W.H. Ali, C.L. Tolliver, and L.-S. Shieh, “Load disturbance resistance speed
controller design for PMSM,” IEEE Trans. Ind. Electron., vol. 53, pp. 1198-1208, 2006
[10] S. Li, and Z. Liu, “Adaptive speed control for permanent-magnet synchronous motor system with variations
of load inertia,” IEEE Trans. Ind. Electron., vol. 56, no. 8, pp. 3050-3059, 2009
[11] K.J. Astrom, and B. Witternmark, Computer-Controlled Systems - Theory and Design, Englewood Cliffs, NJ
: Prentice Hall Inc., 1990
[12] F.L. Lewis, C.T. Abdallah, and D.M. Dawson, Control of Robot Manipulators, New York, NY : Macmillan
Publishing Company, 1993

Copyright (c) 2010 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.