Professional Documents
Culture Documents
Abstract—A control system by means of the back electromotive Because of a series of advantages such as simple structure,
force method is designed for the position-sensorless electric vehi- reliable performance, high efficiency, large starting torque, etc.,
cle (EV) after analyzing the principle of position-sensorless con- the brushless dc motor (BLDCM) has been widely applied
trol for brushless dc motor. Aiming at the main problems of EV:
variation of model and short driving range, a fuzzy sliding-mode in the field of EVs [7], [8]. However, BLDCM still suffers from
controller (FSMC) with regeneration is designed for the EV. The the extra mechanical position sensor for proper commutation.
controller uses fuzzy switching gain adjustment to avoid the whip- As a result, when a disturbance occurs on the position sensor,
pings. The experimental results show that the control system of BLDCM will run unsteady, and noise is produced. Additionally,
position-sensorless EV can successfully achieve position-sensorless the position sensor is easily damaged and poses difficulty in re-
commutation control and replace Hall sensors. FSMC possesses
better performance and higher efficiency than the proportional– pair. The cost of BLDCM also increases due to the presence of
integral–differential controller. Additionally, it can recover more the position sensor. Therefore, research on position-sensorless
energy and increase more driving range. control for BLDCM has become a focus in recent years [9], [10].
Index Terms—Electric machines, fuzzy control, fuzzy logic, mo- In this paper, a position-sensorless BLDCM is employed for the
tion control, motor drives, position control, sliding block codes. EV. In order to eliminate the position sensor, many position-
sensorless control methods of BLDCM with trapezoidal back
I. INTRODUCTION electromotive force (BEMF) have been proposed in the litera-
ture over the last two decades, such as the BEMF method, the
UE to the dual pressure from environmental pollution
D and energy crisis, the development of electric vehicles
(EVs), called “green vehicles” [1], [2] has become a general
phase current method, the magnetic linkage method, the model
reference position estimation method, etc. [11]–[16]. Though
the BEMF method is popular and widely used because of its
trend. EVs have improved their performance, and have also been easy implementation, yet it has the drawback of measurement
made suitable for commercial and domestic use during the last error at low speed. To eliminate the influence of measurement
decades. At present, EVs commonly comprise five parts: control error at low speed, the BEMF method is improved in this paper
system, motor, batteries, bodywork, and charger. Control system from a hardware and software viewpoint, by suitably discon-
as EV’s brain is a very important part of the whole system, and, necting the reference point of detection circuit from the battery
to a great extent, determines the entire vehicle performance, cathode, and by phase-shifting compensation of BEMF.
as it is the key to improve motor efficiency, lengthen the life Nowadays, the proportional–integral–differential (PID) con-
of batteries, increase driving range, and optimize control. In troller has been widely used in the industrial field. It has a simple
the driving and regenerative-braking process, the road state, the structure, and can be easily realized. Yet, the PID controller has
driving mode, and the voltage of batteries are variational. To its shortcomings, especially for a target that is nonlinear and un-
improve the stability and reliability of EV, the control strategy certain. EV’s working condition is complicated and dynamic, so
for driving and braking should be earnestly researched [3]. The the advanced control algorithm should be researched to improve
problem that EVs still have not achieved ranges as good as gas- the performance of EV and the energy-regenerative efficiency.
powered conventional vehicles restricts EV’s fast development. Using rotational speed and acceleration as the state variables,
It is very significant for the increase in range that the technology the sliding-mode controller (SMC) for EV could improve the
of energy-regenerative braking is applied to EVs [4]–[6]. Since robustness and real-time performance of the system [17]. Yet,
a permanent-magnet motor can work as a dynamo in the braking it is unable to resist disturbance while S-function is on the
process, we designed an intelligent control system for EV, which way of approaching the sliding plane. A fuzzy controller has
can recover electrical energy to batteries from the vehicle’s good self-adaptability and the merit of not needing a precise
kinetic energy while braking, and has the following advantages: model [18]–[20]. Yet, it does not have the strong robustness
saving energy, increasing driving range, having no mechanical and real-time performance of SMC. This paper combines fuzzy
braking noise, and so on. controller and SMC; the fuzzy controller adaptively outputs the
switching gain adjustment of the SMC online and ensures the
Manuscript received June 2, 2008; revised November 9, 2008 and February robustness while S-function is approaching the sliding plane.
14, 2009. Current version published September 11, 2009. Recommended for Based on establishing the mathematical model of the control
publication by Associate Editor A. A. Consoli.
J.-B. Cao is with the Transportation College, Zhejiang Normal University, system, the fuzzy SMC (FSMC) for the position-sensorless EV
Jinhua 321004, China (e-mail: caojianbo2008@163.com). was developed and tested successfully by different experiments.
B.-G. Cao is with the Research and Development Center of Electric Vehicle, Additionally, a traditional PID controller was also provided in
Xi’an Jiaotong University, Xi’an 710049, China.
Digital Object Identifier 10.1109/TPEL.2009.2020429 this paper, and the experimental comparison between FSMC
TABLE I
RELATIONSHIP BETWEEN BEMFS AND MOSFETS
ua + ub
uo = (3)
2
ua + ub
ec = uc − uo = uc − . (4)
2
ub + uc
ea = ua − (5) In this case, take phase A as an example. The transfer function
2
ua + uc of ua1 –ua can be deduced from Fig. 4 as follows:
eb = ub − . (6)
2
ua1 R2
= . (7)
From (4)–(6), BEMFs ea , eb , and ec do not include neutral ua R1 + R2 + R1R2C1s
voltage uo , and can be obtained from ua , ub , and uc directly.
As shown in Fig. 3, if the zero-crossing points of ea , eb , and From (7), its phase-shifting angle can be calculated as
ec are obtained, then by delaying the electric angle by 30◦ , the
commutation points can be obtained. The relationship between 2πR1R2C1f
α = arctan (8)
BEMFs and MOSFETs is shown in Table I. R1 + R2
This study involved the design of the detection circuit of
BEMFs, as shown in Fig. 4. To eliminate the influence of the where f is the frequency of BEMF.
motor neutral point, the reference point of detection circuit is As shown in Fig. 6(a), when 0o ≤ α ≤ 30o , if the influence
disconnected from the battery cathode, and the virtual neutral of phase-shifting angle α is not considered, the actual commu-
point N is taken as the reference point, uN ≈ uo . In addition, tation point will offset from point n3 to point n 3. Therefore, the
point N is connected to the ground of the controller so that uref adjusted angle of phase shifting should be: β = 30o − α, i.e.,
would not change along with the speed of BLDCM, because when the zero-crossing point is gained, the controller will carry
the ground of detection circuit is different from that of the main out commutation after delaying the electric angle β. In this case,
circuit, and uref is no longer subject to the influence of the the adjusted angle of phase shifting may also be β = 90o − α; at
main circuit. After sampling voltages ua1 , ub1 , and uc1 are this time, the actual commutation point will offset to point n4. As
compared with uref , zero-crossing signals ua2 , ub2 , and uc2 can shown in Fig. 6(b), when 30o < α ≤ 60o , the adjusted angle of
be obtained, as shown in Fig. 5. The controller will carry out phase shifting is β = 90o − α. Because α is too big to lose com-
commutation control according to these signals. mutation point n3, so the actual commutation point must offset
Because BEMF detection circuit employs an RC filter, the to point n4. So, whenever 0o ≤ α ≤ 30o or 30o < α ≤ 60o , the
phase shifting of zero-crossing point is inevitable, which makes adjusted angle of phase shifting can entirely be β = 90o − α,
the position detection inaccurate. The phase-shifting compen- which can avoid the complexity of program and lead to better
sation must be done during application [25]. reliability.
CAO AND CAO: FUZZY-LOGIC-BASED SLIDING-MODE CONTROLLER DESIGN FOR POSITION-SENSORLESS ELECTRIC VEHICLE 2371
T1_OFF (d · T < t ≤ T )
dim
(La + Lb ) = −ea − eb − im · (ra + rb + rt + rd ) (10)
dt
where d is the duty cycle of T1_PWM, T is the operation period
of T1_PWM, vb is the voltage of batteries, rbat is the equiva-
lent series resistance (ESR) of the batteries, rt is the ESR of a
Fig. 7. Equivalent circuit topology as buck converter. MOSFET conducting, rd is the ESR of the diode, and im is the
phase winding current of stator.
Because the motor’s three phases are similar, in permitted
IV. DRIVING AND REGENERATIVE-BRAKING PROCESS AND scope, have: La = Lb = Lm , ea = eb = em , and ra = rb =
THEIR MATHEMATICAL MODELS [26]–[28] rm . Then, (9) and (10) can be simplified as
A. Driving Process and Its Mathematical Model dim
2Lm = vb − 2em − im · (rbat + 2rm + 2rt ) (11)
The driving circuit topology of the control system is shown in dt
Fig. 2, which includes four batteries, a three-phase full-bridge dim
circuit, and a BLDCM. In this paper, the three-phase BLDCM 2Lm = −2em − im · (2rm + rt + rd ). (12)
dt
works according to six states, and in any state, the two-phase
work principle is the same. The BLDCM could be seen as a According to the deflecting couple equilibrium
dc motor under the condition of two phases conducting. In this
case, take phases A and B as an example to establish the driv- dω
J = Kt im − TL (13)
ing equivalent circuit and mathematical model. During driving dt
process, the current flow direction is shown in Fig. 2 (phases A em = Ke ω (14)
and B); at this time, T1 is PWM, T4 is ON, and other MOSFETs
are OFF. The batteries drive the BLDCM by means of the buck where J is the moment of inertia, ω is the rotational speed of
converter. The equivalent circuit topology for driving is shown motor, Kt is the torque coefficient, TL is the load torque, and
in Fig. 7. Ke is the BEMF coefficient of armature winding.
The mathematical model can be established as follows (see Suppose state variables are x = [ im ω ]T , output is y = im .
Fig. 7, for example). Hence, after average processing, we have (15), as shown at the
While T1 is in conducting state and cutoff state, the state bottom of this page.
equations are, respectively, described as Suppose the intermediate variable is u = d. After handling
by the perturbation method, and separation of steady-state vari-
T1_ON (0 ≤ t ≤ dT )
ables and instant variables, the linear small-signal mathematical
dim model of driving can be obtained as follows (16), as shown at
(La + Lb ) = vb − ea − eb − im · (rbat + ra + rb + 2rt ) the bottom of the next page, where D and X are, respectively,
dt
(9) the circuit steady values of d and x.
2r +r +(r +r )d + r (1 − d)
m t bat t d Ke v d
b
− −
ẋ = 2L m Lm 2Lm
x +
Kt TL . (15)
0 −
J J
y = [1 0]x
2372 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 24, NO. 10, OCTOBER 2009
B. Regenerative-Braking Process and Its Mathematical Model tem is controlled by making S-function tend to zero, and the
tracking error will also tend to zero. Additionally, FSMC does
While braking, EV works in the regenerative state, and
not require the precise mathematical model, and could soften
BLDCM works as a dynamo. The main circuit topology is the
the control signal to reduce or avoid the whippings that common
same as that of driving. In this case, we still take phases A and B
SMC has, and hence, it has good robustness [29]–[33].
as an example to establish the regenerative-braking equivalent
Fig. 9 shows the FSMC designed in this paper, which uses
circuit and mathematical model. While regenerative braking,
the fuzzy switching gain adjustment to eliminate the distur-
T2 is PWM, other MOSFETs are OFF, and the equivalent circuit
bance according to the condition of sliding-mode motion; there-
topology and the current flow direction are shown in Fig. 8. At
fore, it could avoid the whippings, ensure the robustness while
this time, batteries are charged by the boost converter. In the
S-function is approaching the sliding plane, and improve the sta-
whole regenerative-braking process, T2, T4, and T6 are PWM,
bility and reliability of the system. During the driving process,
in turn, according to the defined sequence, and other MOSFETs
the input of the controller is r = im ref , the output is y = im .
are OFF. As a result, the vehicle’s kinetic energy is converted into
During the regenerative-braking process, the input is r = ibatref ,
electrical energy, which can avoid the waste of energy, lengthen
the output is y = ibat . The uncertainties and disturbances in-
the life of batteries, increase the driving range, and have no
clude the parameter perturbation, the unmodeled dynamics, the
mechanical braking noise.
variation of the battery voltage, the road state, and the driving
Similar to driving, a regenerative-braking mathematical
mode of the vehicle.
model can be obtained as follows.
Suppose state variables are x = [ im ω ]T , output is y =
B. FSMC Design for Position-Sensorless EV
ibat , and intermediate variable is u = d. After average pro-
cessing, handling by perturbation method, and separation of Consider the uncertainty of the control system; its state equa-
steady-state variables and instant variables, the linear small- tion could be expressed as
signal mathematical model of regenerative braking can be ob- ẋ = (A + ∆A)x + (B + ∆B)u + (P + ∆P )f (18)
tained as follows (17), as shown at the bottom of this page.
where x ∈ Rn , u ∈ R, f ∈ R, ∆A, and ∆B are the variable
uncertain parameters, and P and ∆P are the disturbances.
V. FSMC DESIGN FOR POSITION-SENSORLESS EV
Suppose the system satisfies the match condition
A. FSMC Introduction
∆A = B Ã, ∆B = B B̃, P + ∆P = B P̃ .
FSMC combines fuzzy controller and SMC together, and
Because ∆A, ∆B, B, P , and ∆P have been defined in (18),
has the advantages of both. The control target changes from
Ã, B̃, and P̃ could be deduced from them. Then
tracking error to sliding-mode function, the input is not
[e, ė, ë. . ., e(n −1) ], yet is (s, ṡ), which is always 2-D. The sys- ẋ = (A + B Ã)x + (B + B B̃)u + (B P̃ )f. (19)
2r + r + (r + r )D + r (1 − D)
m t bat t d Ke
rd − rbat − rt
vb
− −
ẋ = 2L m Lm
0
Xu + 2Lm u
x + 2Lm
Kt 0 0 0 (16)
0
J
y = [1 0]x
2r + r + r D + (r + r )(1 − D)
m d t d bat Ke
rd + rbat − rt
vb
−
ẋ = 2Lm Lm
0
Xu + 2Lm u
x + 2Lm
Kt 0 0 0 . (17)
− 0
J
y = [ 1 − D 0 ] x + [ −1 0 ] Xu
CAO AND CAO: FUZZY-LOGIC-BASED SLIDING-MODE CONTROLLER DESIGN FOR POSITION-SENSORLESS ELECTRIC VEHICLE 2373
Then
V̇ = sṡ = s[ë + cė − Ḟ (t)] = s[ë + cẋ1d − cẋ1 − Ḟ (t)]
= s[ë + cẋ1d − cA11 x1 − cA12 x2
− cB1 u(t) − cB1 E(t) − Ḟ (t)]. (25) Because c > 0, ε > 0, we get V̇ ≤ 0, and the S-function is
When u(t) is substituted by (23), (25) can be simplified as stable.
In the control rule (23), K(t) is the source that causes whip-
V̇ = s[−cK(t)sgn(s) − cB1 E(t)] pings, and is used to counteract E(t) so that the sliding-mode
condition could be satisfied. If E(t) is time varying, K(t) should
= −c[K(t) |s| + B1 E(t)s] ≤ −cε |s| . (26)
be also time varying, and can make the sliding-mode motion
2374 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 24, NO. 10, OCTOBER 2009
Fig. 12. Waveforms of u a 1 and u a 2 based on different reference points. (a) Battery cathode as reference point. (b) Neutral point N as reference point.
Fig. 13. Phase-shifting compensation experiments. (a) Waveform without compensation. (b) Waveform with compensation.
Fig. 14. Commutation control experiments. (a) Waveforms of u a and u b . (b) Waveforms of u a and u a 2 .
approach the sliding-mode plane. The value of K(t) should be resents the positive middle, and PB represents the positive
great enough to eliminate the influence caused by the uncertainty big.
and disturbance. Define the control rule of K(t) as follows. The universes of input sṡ and output ∆K could be defined,
respectively, as
Rule 1: If sṡ < 0, K(t) should be reduced.
Rule 2: If sṡ > 0, K(t) should be increased.
Define sṡ as input and ∆K as output. Then, their fuzzy sets
sṡ = {−15,−10, 0, 10, 15} ∆K = {−1.5,−0.5, 0, 0.5, 1.5}.
could be defined, respectively, as
sṡ = {NB NM ZO PM PB} ∆K = {NB NM ZO PM PB}
where NB represents the negative big, NM represents the The membership functions of input sṡ and output ∆K are,
negative middle, and ZO represents the zero part, PM rep- respectively, shown in Figs. 10 and 11.
CAO AND CAO: FUZZY-LOGIC-BASED SLIDING-MODE CONTROLLER DESIGN FOR POSITION-SENSORLESS ELECTRIC VEHICLE 2375
Fig. 18. Driving experiment comparison. (a) PID controller. (b) FSMC.
Fig. 19. Regenerative-braking experiment comparison. (a) PID controller. (b) FSMC.
U represents the input voltage of the controller, I represents the three parameters of the PID controller are fixed, which makes
input current of the controller, P1 represents the input power of the controller inappropriate for complex nonlinear control
the controller, P2 represents the output power of the motor, M systems. When the working condition of the system is changed,
represents the output torque of the motor, n represents the rota- the PID controller cannot always maintain the best control
tional speed of the motor, efficiency (EFF) represents the system effect.
working efficiency. When EV runs normally, the current value The regenerative-braking experiments were, respectively,
of the motor is less than 15 A and the torque range of the motor done with the PID controller and the FSMC at an initial speed
is 3–15 N·m. By comparison in the range, as shown in Figs. 16 of 25 km/h, and the results are shown in Fig. 19. The refer-
and 17, we find that the FSMC possesses higher efficiency and ence value of the charge current is: ibatref = 5 A. During the
wider effective work sector than the PID controller. For exam- regenerative-braking process, the actual value of the charge cur-
ple, the maximum efficiency point in Fig. 17 is about 85.6%, rent with the PID controller is smaller than the reference one
and the one in Fig. 16 is about 84.8%. The performance curves and has a big undulation. Yet, the FSMC is superior to the PID
of the FSMC are also superior to that of the PID controller. controller at response speed, steady-state tracking error, and re-
Because FSMC has good self-adaptability and strong robust- sisting disturbance, because it has good robustness. As a result,
ness, it could adaptively output the switching gain adjustment it can improve the energy-regenerative efficiency and recover
of the SMC online to ensure the robustness according to the work more energy.
condition. Therefore, the FSMC could make BLDCM work bet-
ter, and, consequently, enables EV to run more smoothly and
efficiently. D. Driving Range Experiments
The main aim of researching the technology of energy-
C. Driving and Regenerative-Braking Control Experiments regenerative braking is to increase the driving range of EV.
The driving and regenerative-braking control experiments The experimental results with different control methods are
were done on the EV made at Boyuan EV Company. The compared and shown in Fig. 20. The experiments were per-
comparative experimental results of driving current, with the formed at the same EV, load, and road conditions. The outdoor
PID controller and the FSMC, are shown in Fig. 18. To facil- temperature was about 28 ◦ C. A smooth and nontransportation
itate analysis and comparison, the output voltage of the sen- road section with 3 km was selected. The EV was braked one
sor was converted to the current value. Approximate step re- time per 300 m, and the voltage of batteries was recorded one
sponse is used to test the startup performance of the controllers. time per 3 km. The initial voltage of the batteries was 54.2 V,
The reference value of the armature current is: im ref = 20 A. and the final voltage was 43.2 V. As shown in Fig. 20, the
In the startup process of EV, the FSMC has better perfor- FSMC with regeneration can recover more energy and increase
mance at response speed and steady-state tracking error than more driving range than the PID controller without regenera-
the PID controller, for the PID controller has its shortcomings, tion by about 20%, than the PID controller with regeneration by
especially for the target that is nonlinear and uncertain. The about 5%.
CAO AND CAO: FUZZY-LOGIC-BASED SLIDING-MODE CONTROLLER DESIGN FOR POSITION-SENSORLESS ELECTRIC VEHICLE 2377
[30] J. K. Liu, MATLAB Simulation for Sliding Mode Control. Beijing, Bing-Gang Cao received the B.S., M.S., and Ph.D.
China: Tsinghua Univ. Press, 2005. degrees in mechanical engineering from Xi’an
[31] Z. Z. Chen, C. H. Shan, and H. L. Zhu, “Adaptive fuzzy sliding mode Jiaotong University, Xi’an, China, in 1976, 1982, and
control algorithm for a non-affine nonlinear system,” IEEE Trans. Ind. 1992, respectively.
Inf., vol. 3, no. 4, pp. 302–311, Nov. 2007. He is currently a Professor at the School of
[32] B. Yoo and W. Ham, “Adaptive fuzzy sliding mode control of nonlinear Mechanical Engineering, Xi’an Jiaotong University,
system,” IEEE Trans. Fuzzy Syst., vol. 6, no. 2, pp. 315–321, May 1998. where he is also the Director of the Research and
[33] R. J. Wai and K. H. Su, “Adaptive enhanced fuzzy sliding-mode control Development Center of Electric Vehicle. His current
for electrical servo drive,” IEEE Trans. Ind. Electron., vol. 53, no. 2, research interests include robust control, intelligent
pp. 569–580, Apr. 2006. control of electric vehicles, noise and vibration con-
trol of liquid systems, and control technology of re-
newable energy.