Proceeding of the 2015 IEEE

International Conference on Information and Automation

Lijiang, China, August 2015

Sliding Mode Speed Control for Brushless DC Motor

Based on Sliding Mode Torque Observer
SHAO Yunbin, YANG Rongrong, GUO Jianwen, FU Yongling
School of Automation Science and Electrical Engineering School of Mechanical Engineering & Automation
Beihang University Beihang University
Beijing, China 100191 Beijing, China 100191
[email protected] [email protected]

Abstract - Brushless DC (BLDC) motor control system is a structure control method and so on, have been conducted.
multi-parameter nonlinear systems, and the ultra-low inductance However, the simple control process of the PI control method
of it makes more stringent requirements on the performance of cannot meet the demand of multi-variable nonlinear control
the control system, which make it difficult using the traditional system for high performance, moreover, the performance of
PI control method and the intelligent control methods, such as
the PI control method in the adaptability to load variations is
fuzzy control method and neural network control method, to
effectively control a low-inductance BLDC motor with high very poor. Turn to the intelligent control methods, such as the
performance demands. Considering this, we propose a new dual- neural network method and fuzzy control method, they need
loop control strategy in which the current hysteresis control is continuous learning, optimization and adjustment in the
adopted for the current loop, and the speed loop control is control process, and consequently, high-performance
realized based on the exponential approach law sliding mode underlying controller, mostly, beyond reality, which make it
speed controller and sliding mode observer to construct the hard for the control system based on them to realize both in
control system for a low-inductance BLDC motor. The hardware and software. In a word, it is hard to use either of
simulation and experimental verification show that compared to these two kinds of control strategies to construct the control
the traditional PI control strategy, the proposed control strategy
system with high performance for a BLDC motor. On the
shows obvious advantages in overshoot control, response speed
and adaptability to load variations. other hand, the sliding mode control method is of great
robustness, easy to realize, and especially applicable to the
Keywords - BLDC motor, Variable Structure Control, Sliding motor control process with strong nonlinearity and frequently-
Mode Control, Sliding Mode Observer varying load, and it has been used in the position loop control,
I. INTRODUCTION speed loop control, current loop control in the BLDC motor
control system [3] [4] [5].
Brushless DC motor (BLDC motor) is a kind of In this paper, the dual loop control structure is used to
permanent magnet synchronous motor with a trapezoidal back construct the control system for the BLDC motor with an
electromotive force （EMF） wave. Due to its large power ultra-small inductor. In this control system, the current
density, high output torque, good dynamic characteristics, hysteresis control is adopted for current loop, and the speed
simple structure and control process and high reliability, loop control is realized based on the exponential approach law
BLDC motor has been widely used in machine tools, electric sliding mode speed controller and sliding mode observer. The
automotive, aerospace and other fields, and has been gradually simulative and experimental results show that compared with
replacing traditional induction motor and brush motor in many that using the traditional PI control method, the control system
application fields. of BLDC motor using the control method proposed in this
Usually, dual-loop control strategy consisting the current paper has better speed performance and adaptability to load
loop control and speed loop control or direct torque strategy is variations. Meanwhile, because of its simplicity to realize, this
used to construct the control system of BLDC motor, and the control system is easy to use and promote, and hence with
former is much more frequently used. In the dual-loop control great engineering value.
strategy, the current hysteresis control method (CHCM),
voltage pulse width modulation (PWM) method, and current II. MATHEMATICAL MODEL OF A BLDC MOTOR
predictive control method are three commonly used methods For the sake of simplicity of the design and simulation,
for the current loop control, while among them, owing to its assumptions about the physical structure and working state of
simple control process, rapid current response, and inherent the BLDC motor are made as follows,
over-current protection capability [1], CHCM has become the 1. Star winding with three completely symmetrical
preferred method for high-performance control system of phases and full pitch is used.
BLDC motor, especially that with ultra-small induction. As 2. The back EMF waveform is trapezoidal with a flat
regards the speed loop control, a large amount of researches top width of 120°.
on it using different control methods [2], such as PI control 3. The motor works in the two-phase-in-conduction
method, model reference adaptive control method, fuzzy state, and the conducted electrical angle is 120°.
control method, neural network control method, variable

978-1-4673-9104-7/15/$31.00 ©2015 IEEE 2466

 4. The motor core is not saturated, and the eddy current
loss, hysteresis loss and cogging are all ignored.
5. System power inverters, diodes all work in perfect
condition.
The governing voltage equations for a BLDC motor is as
follows,
ua   r 0 0 ia  ea   l 0 0 ia 
u   0 r 0 i   e   0 l 0 d i 
 b   b   b    dt  b  , (1)
uc  0 0 r  ic  ec  0 0 l  ic 
ia  ib  ic  0 , (2)
where u, i, e are the phase voltage, phase current, and winding
back EMF respectively, and the subscripts a, b and c
correspond to the three phases respectively, r is the stator
resistance, l is the equivalent inductance considering both of Fig. 1 Structure of sliding mode controller based on exponential
approach law
the self and mutual inductances of the stator.
Assume the input of current loop in the controller is u.
Regardless of the dynamic process of reversing, (1) and
Since frequency response of the current loop is much faster
(2) can be written as
than the speed loop, it is reasonable to render that the current
di
u  Ri  L  k e , (3) loop input is equal to the actual one within the limited range
dt of the current loop, i.e u  i . The motion equation of motor is
where u is the line voltage, R is the line resistance, L is the described by
line inductance, ke is the back EMF coefficient, and ω is the
motor speed. J  kt u  B  Tl . (7)
Motor torque equations are given by
d The racking error is
Te  J  B  Tl , (4) e  d   . (8)
dt
Te  kt i , (5) The sliding surfaced is designed as
s ( )  e . (9)
where Te is the electromagnetic torque, Tl is the load torque, J
is the moment of inertia, B is the damping coefficient, and kt is Let s( )  0 , then
the torque coefficient. s( )  e
  d  
III. EQUIVALENT CONTROL BASED ON EXPONENTIAL . (10)
kt B 1
APPROACH LAW   d  u    Tl  0
J J J
The sliding motion is composed of the reaching motion The equivalent control ueq is
and sliding mode motion, according to the principle of sliding 1
mode control, while the reaching condition can only guarantee ueq  ( J d  B  Tl ) . (11)
kt
that system state can hit the sliding surface from anywhere of
the phase plane, however, the reaching trajectory is not The switching control u sw is
described explicitly [5]. In order to improve the reaching 1
u sw   sgn( s)  ks  . (12)
trajectory and reduce the chattering, the exponential approach kt
law is selected for system designing. The actual control is
Exponential approach law is given by 1
s   sgn( s )  ks   0, k  0 , (6) u  ueq  u sw  ( J d  B  Tl   sgn( s )  ks ) . (13)
kt
where s  ks is the exponential approach term, and its Substituting Eq. (13) into Eq. (10), the following equation can
solution is s  s (0)e  kt .The function of the exponential term is be obtained
guaranteeing that system state can approach the sliding 1
s( )  (  sgn( s )  ks ) . (14)
surface rapidly when the value of s is larger, and the J
exponential term will decrease gradually and to zero Obviously, the stability condition is satisfied
eventually with the system state approaching the sliding 1
s( ) s ( )  (  sgn( s )  ks) s  0 . (15)
surface. In order to reach the sliding surface within finite time, J
the constant reaching law s   sgn(s ) is added. In addition, However, considering that the load torque Tl is unknown, Tl
by adjusting parameters  and k , a faster approach speed and can be seen as disturbance and eliminated from the control
alleviated chattering can be achieved. law, then the reaching condition can be represented by
1
s( ) s ( )  (  sgn( s )  ks  Tl ) s  0 . (16)
J

2467
 In order to meet stability condition, the values of  , k
should be large enough. Nevertheless, in practice, too-large
values of them may bring about serious chattering
phenomenon owing to some factors such as the backlash, and
the control period, which will ruin the control performance. In
order to solve this problem, an observed value of Tl from
observer is compensated into u. The sliding control schematic
is shown by Fig. 1.
IV. LOAD TORQUE OBSERVER DESIGN BASED ON SLIDING
MODE CONTROL
From the motion equation of motor, the following
Fig. 2 Structure of the load torque observer designed based on sliding mode
equation is obtained control
kt B T To validate the proposed algorithm, a system model is
  u  l . (17)
J J J built using the Matlab/Simulink software. Parameters of the
The extended load torque observer is designed by BLDC motor model are as follows: the rated voltage u is 48V,

 kt B  T the rated power P is 1kW, the rated speed n is 500rpm, the
  u    l   sgn(s ) . (18)
J J J number of pole pairs p is 23, the stator phase resistance R is
 0.1743Ω, the stator phase inductance L = 0.139mH, the
Tl  g sgn( s ) . (19)
 maximum allowable stator current I is 50A, the motor inertia J
where  is the estimation value of  , g is the sliding mode
is 1.36kg·m2. Powergui is used to build inverter power
coefficient,  is the switching coefficient, s is the sliding
 module parts of the model, and the discrete system interval is
surface, and s  e     . Ts  5e  6 s .
The observed error dynamic equation can be achieved by As a comparison, another control system using the

 B  T T traditional PI controller is also built. Back-calculation is
     (   )  l l   sgn( s ) . (20)
J J applied as Anti-windup algorithm. The maximum output of PI
The stability condition of the sliding mode is controller is set to 50A, namely, the maximum allowed current
 
ss  (   )(   ) of the motor. The proportional coefficient and integral
 coefficients are 5 and 1 respectively. System model is shown
 B  Tl  Tl . (21)
 (   )( (   )    sgn( s ))  0 in Fig. 3. A 300rpm step input is given at T = 1, and the step
J J
From the above stability condition,  should be selected by response is shown in Fig. 4. A 30N·M disturbance load is
 given at T=10, and the disturbance response is represented in
B  T T
 (   )  l l . (22) Fig. 5.
J J From the simulation results, we can see that when using
When the system reaches the sliding surface, we get the PI control method, as a result of the maximum current
 
        0 . (23) limitation, a uniform acceleration process occurs at the initial
Substituting Eq. (23) into Eq. (20), we can obtain stage of the step response. When added an integrator to, the
e  J sgn( s ) , (24) system has no static error, a clear overshoot of 125rpm arises.
When anti-windup algorithm is applied, the overshoot is
e  g sgn( s ) . (25)
reduced to about 21rpm, but cannot be eliminated either. After
The observed error of load torque is the solution of a disturbance load is conducted, the speed fluctuates
g
e  e  0 . (26) significantly, and the maximum reduction is about 33rpm,
J while the rate of speed change is over 11%, and the recovery
Obviously, g<0 can guarantee the error converges to zero. In time is over 3.0s. Moreover, if increasing the integral
addition, the rate of convergence of observer should much coefficient, the output would fluctuate, and it is hard to get a
faster than that of the controller. good control performance.
V. SIMULATION AND EXPERIMENT

2468
 Fig. 3 Matlab model based on sliding mode control
400 60

300
40

current(A)
speed(rpm)

200
20
100

0
0

-100 -20
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6
time(s) time(s)

50
Fig. 6 Step response and temporal evolution of the current of the sliding
40
mode controller
30
Current(A)

300
20
280
speed(rpm)

10
260
0

-10 240
0 1 2 3 4 5 6 7
Time(s) 220

Fig. 4 Step response and temporal evolution of the current of the PI 200
7 8 9 10 11 12 13 14 15
controller time(s)
60
300

280 40
speed(rpm)

current(A)

260
20
240

220 0

200
8 9 10 11 12 13 14 15 16 -20
7 8 9 10 11 12 13 14 15
time(s)
time(s)
50
Fig. 7 Disturbance response and temporal evolution of the current of the
40
sliding mode controller
30
40
current(A)

20
30
10
torque(NM)

20
0

-10 10
8 9 10 11 12 13 14 15 16
time(s) 0
Fig. 5 Disturbance response and temporal evolution of the current of the -10
7 8 9 10 11 12 13 14 15
PI controller
time(s)
400
Fig. 8 Output of the sliding mode torque observer with filter
300
Response of the controller using sliding mode torque
speed(rpm)

200
observer and sliding mode controller in the same condition as
100
that using in the PI controller is shown in Fig. 6 and 7. The
0
estimated value of disturbance load is shown in Fig. 8. The
-100
0 1 2 3 4 5 6 overshoot is below 3rpm, much reduced compared to that in
time(s)
the condition using PI controller. The speed change after a
disturbance is applied is also greatly reduced, and the
maximum reduction is 8rpm, with a rate of speed change
within 2.7%. It can be seen that the performance is much
better than that using PI controller. Though, a bad influence
that the current chatters slightly chatter is brought about. This
results from some inherit reasons such as the digital control

2469
interval, delay in the speed calculation with hall sensor,
which lead to the chattering of the system in a tight
neighbourhood of the sliding surface. This chatter amplitude
is positive correlated with the sliding coefficient  . On the
premise of satisfying the system reaching condition, the
chattering can be reduced by reducing  .
In order to verify the effectiveness of the proposed
control strategy, an experimental platform for BLDC motor
control is built using DSP (TI TMS320F28335) and CPLD
(ALTERA EPM1270), both the proposed control strategy and
the PI control strategy are applied. In this platform, digital
current hysteresis controller is implemented by CPLD and
high-speed ADC, while phase current collection, hysteresis
Fig. 10 Experimental step response of the PI controller
judgment and power module control signal output are
completed by CPLD. Output frequency of the current loop is
20 kHz. The speed loop is implemented on DSP, and the
output frequency is 2 kHz. System structure is shown in Fig.
9.
PI control experiment and sliding mode experiment are
both carried out on this platform, and results are shown in
Fig. 10 and 11 respectively. When the PI controller is used,
there is a clear overshoot of 36rpm and if increasing the
integral coefficient, the output will fluctuate. It is can be seen
that the control performance becomes better after using the
sliding mode controller.
VI. CONCLUSIONS
Fig. 11 Experimental step response of the sliding mode controller
Taking into account the non-linear characteristics of
BLDC motor control system and the demand of low- REFERENCES
inductance motor control system for high-performance [1] BIMAL K. BOSE, “An Adaptive Hysteresis-Band Current Control
current control, this paper proposes a control strategy Technique of a Voltage-Fed PWM Inverter for Machine Drive System,”
IEEE Transactions on Industrial Electronics, vol. 31, no. 5, pp. 402-408,
combining the current hysteresis control and the speed loop October 1990.
control based on the exponential approach law sliding mode [2] XIA Changliang, Fang Hongwei, “Permanent-Magnet Brushless DC
speed controller and sliding mode torque observer to Motor and Its Control,” Transactions of China Electrotechnical Society,
construct the high-performance control system for an ultra- vol 27, no. 3, pp.25-34, March 2012.
[3] Se-Kyo Chung, Jung-Hoon Lee, Jong-Sun KO, and Myung-Joong Youn,
low-inductance BLDC motor. The simulation and “A robust speed control of brushless direct drive motor using integral
experimental verification show that compared to the variable structure control with sliding mode observer,” Industry
traditional PI control strategy, the proposed control strategy Applications Society Annual Meeting, vol. 1, pp. 393-400, October 1994.
represents better performance in overshoot control, response [4] ZHANG Xiaoguang, SUN Li, ZHAO Ke, “Sliding Mode Control of
PMSM Based on a Novel Load Torque Sliding Mode Observer,”
speed and adaptability to load variations. In addition, due to Proceedings of the CSEE, Vol 32, No. 3, pp. 111-116, Jan 2012.
its simplicity, the proposed control strategy has obvious [5] XIA Chang-liang, LIU Jun-hua, YUWei, LI Zhi-qiang, “Variable
project value. Structure Control of BLDCM Based on Extended State Observer”,
Proceedings of the CSEE, Vol 26, No. 20, pp. 139-143, Oct. 2006.
[6] Liu Jinkun, “Sliding Mode Control Design And Matlab Simulation,”
Current
ADC Tsinghua University Press, Second Edition, pp 50-54, Oct. 2012
Sensor

PC DSP CPLD IGBT BLDC

Hall
Sensor
Fig. 9 Structure of the experimental platform

2470