See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/242571707

Speed Sensorless Control of PMSM Using An Improved Sliding Mode Observer

With Sigmoid Function

Article · January 2007

CITATIONS READS

21 523

2 authors, including:

Mongkol Konghirun
King Mongkut's University of Technology Thonburi
65 PUBLICATIONS 636 CITATIONS

SEE PROFILE

All content following this page was uploaded by Mongkol Konghirun on 02 December 2016.

The user has requested enhancement of the downloaded file.

Speed Sensorless Control of PMSM Using An Improved Sliding Mode Observer With Sigmoid Function 51

Speed Sensorless Control of PMSM Using An

Improved Sliding Mode Observer With
Sigmoid Function
1
Kittithuch Paponpen and Mongkol Konghirun2 , Non-members

ABSTRACT effects of integrator drift is evident. The sensorless

An improved sliding mode observer is presented to speed control using extended Kalman filter is pre-
estimate the rotor position of the permanent mag- sented to avoid this problem [2] - [3]. However, the
net synchronous motors (PMSMs). In this observer, system model used in Kalman filter greatly depends
the discontinuous control is replaced by using the sig- on the variations of motor parameters and accuracy
moid function in order to reduce the chattering prob- of complex calculations.
lem commonly found in the conventional sliding mode In this paper, the sliding mode observer is thus
observer. When the chattering is reduced, the back proposed to overcome these disadvantages. The slid-
emf can be obtained directly from the switching sig- ing mode observer is robust to variation of motor pa-
nal. Thus, the low pass filters may not be necessary rameters. Since the switching signals of the sliding
to extract the low-frequency back emfs. As a result, mode observer contain the back emfs of the motor, it
the delay time due to low pass filters would not be is possible to obtain the speed and position of the mo-
appeared in this proposed observer. Then, the esti- tor directly from the switching signals [4]-[5]. How-
mated rotor position would not necessarily be com- ever, the chattering problem due to the discontinuous
pensated as well. As a result, the proposed sliding controls in sliding mode observer is the major factor
mode observer simplifies sliding mode observer with to make the high system oscillation. Generally, the
minimal parameter adjustments. The overall system low pass filer is used to reduce such problem but it
will be implemented on the TMS320F2812 based digi- produces the delay time. So, the calculated rotor po-
tal controller. The experimental results show that the sition is normally added with offset position.
proposed observer can reduce the chattering problem The improved sliding mode observer is replaced
and simplify the conventional sliding mode observer. the discontinuous control by using the sigmoid func-
tion [6] in order to reduced chattering problem. The
ripples of the state variables are reduced. So the low
Keywords: permanent magnet synchronous motor, pass filter and the back emf calculations are not es-
sliding mode observer, sigmoid function sentially used to reject them.

1. INTRODUCTION 2. PMSM MODEL

The PMSMs have found in several applications The PMSM model in the stationary reference
such as machine tools and robotics, due to their high frame (αβ − axis) is
ratio of power to weight and ease of control. Speed
control of PMSM usually requires the mechanical sen-
∗ R 1 1
iα = − iα − eα + uα
sor for sensing rotor position to achieve drive system. L L L
However, the position sensors such as resolver, ab- ∗ R 1 1
solute position encoder, and QEP sensor increase the iβ = − iβ − eβ + uβ (1)
overall cost and reduce reliability of the system. From L L L
disadvantages of the mechanical sensors, the sensor- eα = −λ0 ωe sin θe ; eβ = λ0 ωe sin θe
less operation of PMSM has been receiving wide at-
where R is the stator resistance (ohm), L is sta-
tention.
tor self inductance (H),iα , iβ , uα , uβ and eα , eβ are
The sensorless speed control of PMSM is presented
the phase currents (amp), phase voltages (volt) and
in [1]. In this paper, the rotor speed is estimated from
back emf (volt) in the stationary reference frame,
calculating the flux linkage that is obtained by inte-
respectively. The ωe is electrical angular velocity
grator processing. However, the suffering from the
(rad/sec), λ0 is the flux linkage of permanent magnet
Manuscript received on July 31, 2006 ; revised on October 3,
(volt.sec/rad) and θe is the electrical rotor position
2006. (rad).
1,2 The authors are with department of Electrical Engi-
Assuming that the rotor speed changes slowly, i.e.,
neering, King Mongkut’s University of Technology Thon- ∗
buri, Bangkok 10140, E-mail: paponpen@yahoo.com and ωe ≈ 0 , then the back emf can be simply rewritten as
Mongkol.Kon@kmutt.ac.th.
52 ECTI TRANSACTIONS ON ELECTRICAL ENG., ELECTRONICS, AND COMMUNICATIONS VOL.5, NO.1 February 2007

H is the sigmoid function [6] which will replace the

∗
eα = −ωe eβ conventional discontinuous control. This function can
∗ be formulated as the following:
eβ = ωe eα (2)
³ ´ 
· ¸ 2
−1
3. AN IMPOROVED SLIDING MODE OB- H(īα ) 1+exp(−aīα )
= ³ ´  (7)
SERVER H(īβ ) 2
−1
1+exp(−aīβ )
Fig.1 shows the system configuration of the speed
sensorless control of PMSM using the sliding mode
observer. Two phase currents are measured and
transformed to the αβ-axis. Then, these currents and
two computed phase voltages in αβ-axis are fed to the
sliding mode observer.
The conventional sliding mode observer is shown
in Fig.2. Discontinuous control or bang-bang control
is used to be an action law. Generally, equivalent
controls of conventional sliding mode observer can be
obtained in (3) [7]:

(k1 sign(i¯α ))eq = eα

(k1 sign(i¯β ))eq = eβ (3) Fig.1: Overall functional blocks
To extract eα and eβ from the corresponding equiv-
alent controls in (3), low pass filters where zα , zβ are
outputs are used.

z α = eα + ∆α
zβ = eβ + ∆β (4)
Where ∆α , ∆β are the errors determined by the
distortions of both slow and fast components of the
discontinuous filter inputs.
For high performance applications, the bang-bang
control is replaced by saturation function. Moreover, Fig.2: Conventional sliding mode observer
back emf observer is used to design for better filter-
ing and estimating the rotor speed simultaneously as
shown in Fig.3. This observer is formed as follows [7]:
∗
êα = ω̂e êβ − k2 (êα + zα )
∗
êβ = ω̂e êα − k2 (êβ + zβ ) (5)
∗
ω̂e = (êβ − zβ )êα − (êα − zα )êβ
Where k2 is positive gain. Its proper value is de-
rived by using Lyapunov stability.
Fig.3: Sliding mode observer with saturation func-
The proposed sliding mode observer can be de-
tion
picted in Fig.4. It only consists of the three parts,
i.e., current observer, continuous control function or where a is a positive constant that can be adjusted
sigmoid function and rotor position and speed calcu- the slope of the sigmoid function. The i¯α = îα − iα
lator, which explains as follows: and i¯β = îβ − iβ are the error between estimated
Current observer model: phase current and actual one in the α− and β−axis,
∗ respectively.
R 1 1
îα = − îα + uα − k1 H(îα − iα ) The sliding mode surface is defined as
L L L
T
∗
R 1 1 sn = [sα sβ ] (8)
îβ = − îβ + uβ − k1 H(îβ − iβ ) (6)
L L L where sα = i¯α and sβ = i¯β . When the estimation
where superscript “ ∧ ” represents the estimated error trajectories reach the sliding surface, i.e.,sn = 0.
quantities, k1 is constant current observer gain, and It is obvious that the observed currents will converge
Speed Sensorless Control of PMSM Using An Improved Sliding Mode Observer With Sigmoid Function 53

4. EXPERIMENTAL RESULTS
The hardware configuration of a DSP-based exper-
imental system is shown in Fig.5. The process of the
system is a fixed-point TMS320F2812 DSP with a
clock frequency of 150 MHz. The PMSM is supplied
by a three phase voltage source PWM inverter with
a switching frequency of 20 kHz. A Quadratic En-
coder Pulse (QEP) sensor is used to detect the rotor
position and merely compare with the estimated ro-
Fig.4: Improved sliding mode observer tor position. Two motor phase currents are sensed,
rescaled, and converted to digital values by on-chip
to the actual ones, i.e., îα = iα and îβ = iβ . It is ADC with 12-bit resolution. The PWM gate firing
important that on the sliding surface, the observer signals from the desired phase voltage commands are
will not be affected by any system parameters or any generated by means of the space vector modulation
disturbances. technique. The sampling period of the control system
∗
The sliding mode exists when sn sn < 0, that is is set at 50 µs.
sn → 0 as t → ∞ . Let’s design the Lyapunov func- Fig.6 shows the speed response of PI controller sys-
tion to find such condition of sliding mode existence. tem when using the improved sliding mode observer
at no load condition for a step change of speed from
1 T 1 0.4 PU (360 rpm) to 0.7 PU (630 rpm) at time of 0.18
V = sn sn = (s2α + s2β ) (9)
2 2 s. It is seen that the motor speed can be successfully
The error equation is obtained by subtracting (6) converged to the speed command.
from (1) as Next, the steady-state estimated rotor position re-
∗
sponses and actual rotor position (from QEP) at a
∗ ∗ ∗ R 1 1
s¯α = i¯α = iˆα − iα = − i¯α + eα − k1 H(i¯α ) controlled speed of 0.5 PU (450 rpm) for three differ-
L L L ent structures of sliding mode observers are compared
∗ ∗ ∗ ∗ R 1 1 and shown in Fig.7 through 9. In conventional one
s¯β = i¯β = iˆβ − iβ = − i¯β + eβ − k1 H(i¯β ) (10) and one with saturation function, the rotor position
L L L
∗
errors are significant in the same magnitude, as seen
∗
The sliding mode exists V = sTn sn < 0 [3], i.e., in Figs 7(b) and 8(b). This is due to the effect of de-
lay time of low pass filter. Thus, the estimated rotor
∗ R ³ ¯ 2 ¯ 2´ 1 position should be further compensated by adding an
sTn sn = − iα + iβ + (eα i¯α − k1 i¯α H(i¯α ))
L L offset according to operating speeds. When using the
improved sliding mode observer, the slow component
1 could be extracted directly from the sigmoid function
+ (eβ i¯β − k1 i¯β H(i¯β )) < 0 (11)
L without low pass filter. Thus, it could represent the
As a result, back emf. Then the resulting rotor position can be
k1 ≥ max (|eα |, |eβ |) (12) computed without angle compensation due to delay
time of filter. The rotor position error is greatly re-
Once the sliding mode occurs when selecting a duced as shown in Fig 9(b).
large enough k1 , then the sliding surface becomes as Fig.10 shows the steady-state slow components in
h∗ ∗i the stationary reference frame (α-axis) obtained by
T
sα sβ = [sα sβ ] = [0 0] (13) these sliding mode observers at the same controlled
speed. In conventional one and one with saturation
and (13) can be rewritten as
function, the slow components are obtained from the
low pass filters. It is obvious that the magnitude
(k1 H (i¯α ))eq = eα
of slow component is significantly reduced when the
(k1 H (i¯β ))eq = eβ (14) continuous control (i.e., saturation and sigmoid func-
The chattering problem could be rejected by the tions) is being used rather than discontinuous con-
sigmoid function which is the continuous control. trol (i.e. bang-bang control). This implies that the
Back emf from equation (14) can be used to calcu- high oscillation, causing chattering problem, on the
late the rotor speed and position directly as follows. observed back emf is reduced.
µ ¶
−1 eα
θ̂e = − tan (15)
eβ

dθ̂e
ω̂e = (16)
dt
54 ECTI TRANSACTIONS ON ELECTRICAL ENG., ELECTRONICS, AND COMMUNICATIONS VOL.5, NO.1 February 2007

Fig.5: Overall hardware implementation

Fig.8: Rotor position response when using the slid-

ing mode observer with saturation function at speed
0.5 PU (a) the actual rotor position from QEP and
the estimated rotor position (b) error between them

Fig.6: Step speed response when using the improve

sliding mode observer

Fig.9: : Rotor position response when using the

improved sliding mode observer at speed 0.5 PU (a)
the actual rotor position from QEP and the estimated
rotor position (b) error between them

Fig.7: Rotor position response when using the con-

ventional sliding mode observer at speed 0.5 PU (a)
the actual rotor position from QEP and the estimated
rotor position (b) error between them
Speed Sensorless Control of PMSM Using An Improved Sliding Mode Observer With Sigmoid Function 55

View publication stats

[2] Q. Albert., B. Wu, and K. Hassan., “Sensor-

less control of permanent magnet synchronous
motor using extended Kalman filter,” Electri-
cal and Computer Engineering Conference 2004,
pp.1557 - 1562.
[3] G. Garcia Soto, E. Mendes, and A. Razek,
“Reduced-order observers for rotor flux, rotor
resistance and speed estimation for vector con-
trolled induction motor drives using the ex-
tended Kalman filter technique,” Electric Power
Applications, IEEE Proceedings-Volume 146,
1999, pp. 282-288.
[4] T. Furuhashi, S. Sangwongwanich, and S.
Okuma, “A position - and - velocity sensorless
control for brushless DC motors using an adap-
tive sliding mode observer,” IEEE Trans. Indus-
Fig.10: Slow components from different sliding trial Electronics, Vol.39, 1992, pp. 89 - 95.
mode observer at speed 0.5 PU (a) conventional slid- [5] Z.M.A. Peixoto, Sa F.M Freitas, P.F. Seixas, and
ing observer with bang-bang control (b) sliding mode B.R. Menezes, “Application of sliding mode ob-
observer with saturation function and (c) the im- server for induced e.m.f., position and speed es-
proved sliding observer timation of permanent magnet motors,” in Proc.
Power Electronics and Drive Systems Int. Conf.,
1995, pp.599 - 604.
5. CONCLUSIONS
[6] M. Ertugrul, O. Kaynak, A. Sabanovic, and K.
The proposed sliding mode observer has been pre- Ohnishi, “A generalized approach for Lyapunov
sented to estimate the rotor speed and position of the design of sliding mode controllers for motion con-
PMSM. This observer is very easy to implement and trol applications,” in Proc. AMC’96-MIE Conf.,
requires a few parameters to be adjusted. Accord- 1996, pp.407 - 412.
ing to the experimental results, the estimated rotor [7] M. Elbuluk, and Li Changsheng, “Sliding mode
position is nearly identical to the actual one. The observer for wide-speed sensorless control of
proposed sliding mode observer greatly improves the PMSM drives,” IAS Annual Meeting Conf, 2003,
estimations, comparing with the conventional sliding pp.480 - 485.
mode observer. The chattering problem is signifi-
cantly lessened when using this proposed observer. It
can simplify the sliding mode observer that uses con-
tinuous control function by eliminating the low pass Kittithuch Paponpen received a
filter and back emf calculations. bachelor’s degree in Electrical Engi-
neering from Suranaree University of
Technology, Nakornrachasima, Thai-
land. Nowadays he is pursuing a Mas-
6. APPENDIX ter degree in Department of Electrical
Engineering, King Mongkut’s Univer-
sity of Technology Thonburi, Thailand.
His ongoing research is the sensorless
Table 1: Parameters of the 24 V PMSM motor drive system of permanent magnet syn-
Stator resistance 0.79 Ω chronous motors.
Stator inductance 1.17 mV
Permanent magnet flux 17.666 m (volt.sec/rad)
Number of poles 8
Slope of sigmoid function a=4
Sliding Gain k1 =1.2 Mongkol Konghirun received a B.Eng.
Base voltage 13.86 V Electrical from King Mongkut’s Univer-
Base current 3A sity of Technology Thonburi, Thailand
Base electric frequency 60 Hz in 1995. And he received M.Sc. and
Base Speed 900 rpm Ph.D. degrees in electrical engineering
from the Ohio State University, USA in
1999 and 2003, respectively. Presently,
he is working in Electrical Engineering
department at King Mongkut’s Univer-
sity of Technology Thonburi, Thailand.
References His research interests include electrical
machine drives, and real-time control using digital signal pro-
[1] S. Tomonobu, S. Tsuyoshi, and U. Katsumi, cessors.
“Vector control of permanent magnet syn-
chronous motors without position and speed sen-
sors,” in PESC ’95 , pp.759 - 765.