IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 67, NO.

7, JULY 2020 6147

Sensorless Optimal Commutation Steady Speed

Control Method for a Nonideal Back-EMF BLDC
Motor Drive System Including Buck Converter
Xi Chen and Gang Liu , Member, IEEE

Abstract—For the low inductance brushless dc motor

with nonstandard trapezoidal waveform and asymmetric
three-phase back-EMF, this article first analyses the influ-
ence of nonideal back-EMF on the position of the detected
commutation point when sensorless method is applied.
Then, the fluctuations of electromagnetic torque driven by
square wave current are compared when commutating ac-
cording to several different commutation points, and the
optimal commutation control method is proposed. In addi-
tion, for the low inductor motor with nonideal back-EMF and
buck front-end drive circuit structure, an integrated nonlin-
ear model of motor and buck drive circuit including precise
back-EMF parameters is established. Based on the nonlin-
ear model, the feasibility of model linearization is analyzed.
The linearization method based on input–output feedback
Fig. 1. Back-EMF of BLDCM. (a) Ideal trapezoidal back-EMF.
method is proposed, and a steady-state controller based (b) Nonideal actual back-EMF.
on linearization model is designed. Finally, experimental re-
sults verify the effectiveness of the proposed optimal com-
mutation control and steady speed controller.
the attitude actuator [3]. In particular, especially in recent years,
Index Terms—Buck converter, brushless dc (BLDC) mo-
tor, linearization method, nonideal back-electromotive force with the rapid development of commercial aerospace on a global
(EMF), nonlinear BLDC model, optimal commutation, sen- scale, the demand for high-performance spacecraft attitude ac-
sorless control, steady speed control. tuators has become more urgent, which also puts higher require-
ments on the performance of BLDCMs. In order to reduce eddy
I. INTRODUCTION currents and hysteresis losses in space application, coreless and
RUSHLESS dc motors (BLDCM) are increasingly ap- slotless BLDCMs are usually adopted in MSCMG [4].
B plied in many fields such as home appliances, industrial,
aerospace, etc., due to their high energy density, high efficiency,
Traditional rotor position signals used to drive BLDCM are
usually provided by Hall sensors mounted on the stator. How-
simple structure, and easy control [1], [2]. In view of its se- ever, when the position sensor signal is disturbed, the BLDCM
ries of advantages, BLDCMs are very suitable for magnetically will not be commutated normally so that vibration and noise
suspended control moment gyroscope (MSCMG), which is con- are consequently generated. In addition, the position sensors
sidered to be the preferred attitude actuator of new generation are easily damaged and difficult to be repaired. Position sensors
spacecraft. In MSCMG, BLDCM as an inertia wheel drive pro- with increasing cost and reducing reliability limit the application
viding high precise and large angular momentum, its control per- of BLDCM in complex environments (such as in harsh space
formance has a significant impact on the overall performance of environments). Therefore, the position sensorless technology of
BLDCM control has become a research hotspot [5]–[7].
Manuscript received March 8, 2019; revised June 24, 2019 and In the past decades, many methods have been proposed and
September 10, 2019; accepted September 26, 2019. Date of publica- developed around sensorless control. Most of them are based on
tion October 8, 2019; date of current version March 4, 2020. This work the assumption that the back-electromotive force (EMF) is trape-
was supported by the National Natural Science Foundation of China
under Grant 61773041 and Grant 61822302. (Corresponding author: Xi zoidal and symmetrical waveform. However, due to the design
Chen.) and manufacturing defection, it is difficult to realize standard
The authors are with the School of Instrumentation Science and Op- trapezoidal and symmetrical back-EMF waveform. Most of the
toelectronics Engineering, Beihang University (BUAA), Beijing 100191,
China, and also with the Ningbo Institute of Technology, Beihang Uni- actual back-EMF waveforms lie between trapezoidal and sinu-
versity (BUAA), Ningbo 315800, China (e-mail:, [email protected]; soidal waveform, as shown in Fig. 1. If the back-EMF is no
[email protected]). longer standard trapezoidal, but the drive system still supplies
Color versions of one or more of the figures in this article are available
online at http://ieeexplore.ieee.org. the square currents in the traditional way, which will undoubt-
Digital Object Identifier 10.1109/TIE.2019.2945282 edly cause current regulation difficulty and significant principle
0278-0046 © 2019 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: Chang Gung Univ.. Downloaded on September 09,2024 at 13:32:35 UTC from IEEE Xplore. Restrictions apply.
6148 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 67, NO. 7, JULY 2020

torque ripples. In addition to the nonstandard trapezoidal back- viously insufficient to adopt only the PID method to achieve
EMF, the obvious three-phase back-EMF asymmetry also exists steady and accurate current and speed control. Some non-PID
in some motors, especially when the magnetic gap between the methods have also developed rapidly. In [17], a double loop
rotor and stator fluctuates. The asymmetry is embodied by the controller based on the ADRCT is designed for the bidirec-
fact that the magnitude and offset of the three-phase back-EMF tional Buck-Boost converter control. Piecewise linear (PWL)
are not exactly the same, as shown in Fig. 1(b). In this article, state-feedback control law and model predictive control are
the nonstandard trapezoidal and asymmetric waveform of the used for dc–dc converter control in [18]. Furthermore, with
back-EMF are collectively called as nonideal back-EMF. the development of nonlinear theory, nonlinear control methods
By dividing each pulsewidth modulation (PWM) cycle into have been applied to some dc–dc converters and motor con-
three segments in the commutation time, the commutation trol fields. Tan et al. [19] presents a general design method of
torque ripple can be eliminated and the average current con- sliding mode variable structure (SMVS) controller, which is ap-
version rate of the on-phase can be kept equal to the average plied to three different types of dc–dc converters and achieves
switching phase rate in each PWM cycle by selecting the appro- good performance. Other nonlinear control methods, such as
priate proportion of the three segments in [8]. In the previous robust control [20] and adaptive control [21], are also applica-
researches, there are some methods to deal with the torque ripple ble to power electronic drive control. In short, compared with
caused by nonstandard trapezoidal back-EMF, such as the op- PID controllers, these controllers improve system performance
timal current injection method adopted in [9], which generates in different aspects. Recently, feedback linearization methods
current commands including harmonics to ensure that the prod- have been widely used in power electronics [22], magnetic bear-
uct of phase current and back-EMF remains constant. In [10], ing control [23], and motor control [24]. This linearization has
the average torque control method in one cycle is used to re- higher accuracy due to the use of a more accurate model and
duce the torque ripple of nontrapezoidal back-EMF. Moreover, without the use of a linear approximation that ignores higher or-
some methods use direct torque control (DTC) method [11], der nonlinear terms [26]. In [25], feedback linearization is used
[12], in which instantaneous torque is selected as the control to solve the motion control problem of wheeled mobile robots
target variable. Several kinds of instantaneous torque estima- in an obstacle-free environment. In [27], feedback lineariza-
tion have been studied. In [13], a method of calculating torque tion method is applied to speed control of wind turbine with
using a precalibrated back-EMF waveform is proposed. The Boost–Chopper converter, and excellent control performance
method of estimating the back-EMF and rotor position to calcu- is obtained. Based on the previous research results, this arti-
late the instantaneous torque using sliding mode observer and cle applies the feedback linearization method to the sensorless
Kalman filter is proposed in [14]. In [15], the back-EMF is speed control of nonideal back-EMF BLDC motor with buck
estimated by using neural network fitting, and the rotor posi- converter.
tion is obtained from the three-phase Hall sensors. All of the In this article, the influence of nonideal back-EMF on the
above methods need accurate rotor position information. How- detection of sensorless commutation point is first analyzed.
ever, due to the existence of asymmetric back-EMF, the rotor Then, the electromagnetic torque fluctuation and efficiency of
position information obtained by the most commonly used sen- the motor commutated at different commutation points driven
sorless method may be disturbed. As far as I am concerned, by square current are compared, and the optimal commutation
no relevant analysis has been reported. Therefore, the nonideal control method is proposed. In addition, for the low inductor
back-EMF position sensorless control method considering both motor with nonideal back-EMF and buck converter front-end
asymmetric and nonstandard trapezoidal back-EMF is worth drive structure, an integrated nonlinear model of motor and
studying. drive circuit including precise back-EMF parameters is estab-
In addition, the full bridge PWM method is mostly adopted lished. Based on the nonlinear model, the feasibility of model
in BLDCM drive. However, when applied to low inductance linearization is analyzed. The linearization method based on
motors, it will bring serious high-frequency current modulation input–output feedback method is proposed, and a steady speed
pulsation. For BLDCM with very low inductance such as the controller based on linearization model is designed. Simulation
coreless one in MSCMG, [16] uses PWM modulated Buck con- and experimental results are given to verify the effectiveness of
verter front-end drive structure to improve current and torque the proposed method.
performance and reduce copper loss and iron loss caused by
high frequency.
Proportional-integral-differential (PID) controller has been II. INFLUENCE OF NONIDEAL BACK-EMF ON COMMUTATION
widely used in the industrial field because of its simple de- POINT DETECTION AND OPTIMAL COMMUTATION CONTROL
sign and good applicability. However, when this target error-
based method used in nonlinear systems, the parameters of its A. Deviated Commutation Points Detected by Virtual
controller are always difficult to tune, which must be deter- Neutral Points Method due to Nonideal Back-EMF
mined according to trial error methods and experiments. For In the virtual neutral point-based position sensorless zero
systems with time-varying control characteristics, even mul- crossing point (ZCP) detection method, when the neutral point
tiple sets of parameters need to be prepared, which compli- voltage is unmeasurable, three Y-type connected resistor net-
cates the actual application. Especially in the case of nonideal work are used to provide a virtual neutral voltage signal as
back-EMF motor and buck converter drive system, it is ob- shown in Fig. 2. Where, g denotes ground, n denotes the neu-

Authorized licensed use limited to: Chang Gung Univ.. Downloaded on September 09,2024 at 13:32:35 UTC from IEEE Xplore. Restrictions apply.
CHEN AND LIU: SENSORLESS OPTIMAL COMMUTATION STEADY SPEED CONTROL METHOD 6149

The following constraint equations can be obtained according

to the Y-type connection:
1
Usg = (Uag + Ubg + Ucg ) (4)
3
1
Un g = (Uag + Ubg + Ucg − Ea − Eb − Ec ) . (5)
3
The sensorless ZCP detection method based on the virtual
Fig. 2. Schematic diagram of zero-crossing detection method based
on virtual neutral point. neutral point compares the terminal voltage with the virtual
neutral point voltage to obtain ZCPs of the nonconducting phase
back-EMF. Taking the phase A as an example, when phase A
tral point of the three-phase windings, and s denotes the virtual is the nonconductive, the detected voltage difference in virtual
neutral point. neutral point-based method can be obtained as (6)
The terminal voltage mathematic model of the BLDC motor 1
is written as (1) Uas = Uag − Usg = (2Ea − Eb − Ec ) . (6)
3
dix Similarly, phase B and C are as (7), respectively:
Uxg = Lx + Rx ix + Ex + Un g (1)
dt 1
where Uxy represents the voltage difference between x and y, Ubs = Ubg − Usg = (2Eb − Ec − Ea )
3
Lx , Rx , ix , Ex represent the inductance, resistance, phase cur-
1
rent, and back-EMF of phase x, respectively. Here, the three- Ucs = Ucg − Usg = (2Ec − Ea − Eb ) . (7)
phase resistance and inductance are treated as constant and 3
equal. By taking (2) into (6), (7) and ignoring the third and above
The harmonic form expression of the three-phase nonideal harmonic components that have less influence on the ZCPs de-
back-EMF of the BLDCM is as (2) tection, detailed actual detected voltage can be expressed as the
⎧ equations at the bottom of this page. It is well known that odd
⎪
⎨ Ea = ωe ke ea (θe ) harmonics are the main components in trapezoidal wave sig-
Eb = ωe ke eb (θe ) (2) nals, while second or other even harmonics are very few. So,
⎪
⎩
Ec = ωe ke ec (θe ) after neglecting three or more harmonics, only the fundamental
and dc components are left when calculating the influence of
where
⎧ ∞ actual measured voltage on ZCP. Thus, the actual detected ZCPs
⎪
⎨ ea (θe ) =  m =1 Aam sin(m (θe )) + Da can be obtained by making them equal to zero. Moreover, the
eb (θe ) = ∞ =1 Abm sin m θe − 3
2π
+ Db theoretically ideal commutation point should be the intersection
⎪
⎩ m
∞   2π
 of two adjacent back-EMFs, that is, the points where Ea = Eb ,
ec (θe ) = m =1 Acm sin m θe + 3 + Dc
Eb = Ec , Ec = Ea . By comparing the phase relationship be-
ωe represents the electromagnetic speed, ke represents the back- tween the detected ZCPs and the ideal commutation points, the
EMF coefficient, θe represents the electrical angle position, Axm phase angle of the detected ZCPs deviating from the ideal ZCPs
represents the mth harmonic coefficient of the x phase back- ϕz 1−6 can be obtained.
EMF, Dx represents the dc offset of the x phase back-EMF,
and ex represents the x phase back-EMF waveform shape func- B. Comparison of Different Commutation Points on
tion. The corresponding coefficients can be obtained by offline Instantaneous Torque
measurement.
Considering the extremely low inductance value, the inconsis-
According to Kirchhoff’s law, the terminal voltage can also
tent current change rate during the commutation process caused
be expressed as (3)
by the inductance can be ignored. Thus, under the square current
Uxg = Uxs + Usg (3) drive mode, only two phases are turned on at any time. Then,

√
ke ωe 3/2(Ab1 −Ac1 )
Uas ≈ (2Aa1 +0.5Ab1 +0.5Ac1 )2 +3/4(Ab1 −Ac1 )2 sin θe +arctan +2Da −Db −Dc
3 2Aa1 +0.5Ab1 +0.5Ac1
 √ √
ke ωe √ √ − 3Ab1 − 3/2Ac1
Ubs ≈ (−Aa1 −Ab1 +0.5Ac1 )2 +( 3Ab1 + 3/2Ac1 )2 sin θe +arctan −Da +2Db −Dc
3 −Aa1 −Ab1 +0.5Ac1
 √ √
ke ωe √ √ 3/2Ab1 + 3Ac1
Ucs ≈ (−Aa1 +0.5Ab1 −Ac1 )2 +( 3/2Ab1 + 3Ac1 )2 sin θe +arctan −Da −Db +2Dc
3 −Aa1 +0.5Ab1 −Ac1

Authorized licensed use limited to: Chang Gung Univ.. Downloaded on September 09,2024 at 13:32:35 UTC from IEEE Xplore. Restrictions apply.
6150 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 67, NO. 7, JULY 2020

TABLE I fluctuations for large inertia motors, they will also affect the
BACK-EMF DIFFERENCE ex y (θe ) AT DIFFERENT INTERVALS
current waveform and influence steady speed accuracy.

C. Optimal Commutation Control

From the above analysis, it can be concluded that among
the three commutation points, the motor commutated according
to the ideal commutation point has the lowest torque fluctua-
tion and the highest average torque and efficiency. In addition,
besides the influence of nonideal back EMF on the phase devi-
ation of ZCP detection method, the phase delay introduced by
low pass filter (LPF) in various signal processing circuits is the
main part of commutation deviation. LPF cutoff frequency used
to eliminate noise interference to avoid false detection is very
low, so a large phase delay is introduced. Based on the analysis
of commutation deviation, this article presents a new optimal
commutation method. The optimal commutation method is an
improved commutation compensation method, which considers
low-pass filter delay and nonideal back-EMF factors. The pro-
posed optimal commutation method consists of the following
steps:
1) Estimate LPF delay phase: The LPF is generally ap-
proximated to a first-order inertia unit as LP F (s) =
ωc
s+ω c , ωc is the cutoff frequency speed, then LPF de-
lay phase ϕL P F can be estimated according to the phase
characteristics ϕL P F = − arctan(ωe /ωc ).
2) Calculate deviation phase ϕz 1−6 : First, obtain the actual
back-EMF waveforms by offline measurement at con-
stant speed and normalize back-EMF waveform function.
Fig. 3. Currents and torque waveform of BLDCM driven by square Then, a 1-2-9-1 structure four layers feed-forward back-
wave current. (a) Ideal back-EMF. (b) Nonideal back-EMF. (c) Torque propagation neural network is used for fitting the accurate
waveforms commutated by different commutation points.
back-EMF waveform function. The back-EMF harmonic
coefficients can be obtained through performing Fourier
the instantaneous torque Te of the motor can be expressed as (8) decomposition for the fitted back-EMF function. Finally,
Ex ix + Ey iy Ex ix + Ey iy the ZCP of the actual detected voltage can be calcu-
Te = =p lated by combining the harmonic coefficients obtained
ωm ωe
with equation on the bottom of page 3. The difference
= pke (ex ix + ey iy ) = pke exy (θe ) im . (8) between the six ZCPs and the ideal ZCPs is exactly six
phase deviations ϕz 1−6 caused by the nonideal back EMF
where ωm is the mechanical speed, p is the pole pairs, exy (θe )
in sensorless method [15].
is the back-EMF difference shape function of phase x and y at
3) Real-time compensation: ϕL P F and ϕz 1−6 are super-
position θe , and im is the dc-link current.
imposed to compensate the ZCPs online in real-time to
Ignoring the commutation process, AB interval is defined as
achieve optimal commutation.
the phase A and B conduction interval, where current flowing
The flowchart of implementing the proposed optimal com-
from phase A to phase B. Similarly, each cycle can be divided
mutation algorithm is shown in Fig. 4 .
into six intervals, namely AB, AC, BC, BA, CA, and CB.
The back-EMF difference shape function exy (θe ) at different
III. NONIDEAL BACK-EMF BLDCM MODEL WITH
intervals are listed in Table I.
BUCK CONVERTER
For current closed-loop control, it can be seen that the back-
EMF models in each interval are different and nonlinear in A. BLDC Model
Fig. 3, which makes it difficult to control the current waveform Ignoring the commutation process, the equivalent voltage
in the traditional square wave current drive mode. balance equation can be expressed as (9) in every conduction
Fig. 3(c) shows the instantaneous torque waveforms of the interval
motor commutated according to commutation points by dif-
ferent methods. Obviously, standard and detected commuta- udc −Ex + Ey = (Rx + Ry ) im . (9)
tion points will increase the fluctuation of instantaneous torque
Equation (10) can be obtained by combining (1) and (2)
near commutation points under nonideal back-EMF BLDCM.
Although these fluctuations are not enough to cause huge speed udc − ωe ke exy (θe ) = 2Rim . (10)

Authorized licensed use limited to: Chang Gung Univ.. Downloaded on September 09,2024 at 13:32:35 UTC from IEEE Xplore. Restrictions apply.
CHEN AND LIU: SENSORLESS OPTIMAL COMMUTATION STEADY SPEED CONTROL METHOD 6151

C. Overall Nonlinear Model

Considering the motor model and buck converter model to-
gether, (14) can be derived.
⎧ di udc uin
⎪
⎪
L
=− + D
⎪
⎪ dt
⎪
⎪
L L
⎪
⎪ dudc iL
⎪
⎪
udc ke ωe exy (θe )
⎪ dt = C − 2RC +
⎪
⎪ 2RC
⎪
⎨
dωe p2 ke p2 ke 2
⎪ = udc exy (θe )− ωe exy 2 (θe ) . (14)
⎪
⎪ dt 2RJ 2RJ
⎪
⎪
⎪
⎪ B p
⎪
⎪ − ωe − TL
⎪
⎪ J J
⎪
⎪
⎪
⎩ dθe = ω
e
dt
Select iL , udc , ωe , and θe as the state variables, x1 = iL , x2 =
Fig. 4. Flowchart of the proposed optimal commutation algorithm. udc , x3 = ωe , and x4 = θe , D as the input variable, u = D,
and θe as the output variable, y = x4 = θe . A fourth-order
single-input single-output (SISO) affine nonlinear system can be
obtained

ẋ = f (x) + g(x)u
(15)
y = h (x)
where
⎡ ⎤
β1 x2
⎢ ⎥
⎢ β2 x1 + β3 x2 + β4 x3 exy (x4 ) ⎥
⎢ ⎥
Fig. 5. BLDCM equivalent model diagram based on buck converter. f (x) = ⎢ ⎥
⎢ β5 x2 exy (x4 ) + β6 x3 exy 2 (x4 ) + β7 x3 + β8 ⎥
⎣ ⎦
The motion equations of BLDCM are expressed as x3
⎧  T
⎪ dω B
⎨ Jp e = pke exy (θe ) im − ωe − TL g(x) = β9 0 0 0 , h (x) = x4
dt p (11)
⎪
⎩ dθ e 1 1 1 ke p2 ke
= ωe β1 = − , β2 = , β3 = − , β4 = , β5 =
dt L C 2CR 2CR 2RJ
where, B is damping coefficient, TL is load torque.
p2 ke 2 B p uin
β6 = − , β7 = − , β8 = − TL , β9 = .
B. Buck Converter Model 2RJ J J L
The buck converter can operate in continuous current mode Then, the following (16) can be derived. All partial derivatives
during the normal operating range without inductive current of f1 (x) exist, and they are continuous and bounded. There-
interruption by selecting the appropriate value of the inductor. fore, f1 (x) is a smooth local Lipschitz. By proving that f2 (x),
As shown in Fig. 5, S(t) is a switching signal of the power f3 (x), f4 (x), g1 (x), g2 (x), g3 (x), and g4 (x) are smooth local
device, and the state of S(t) is defined as (12), where 1 represents Lipschitz, the same conclusions of f (x) and g(x) can be
switch ON, 0 represents switch OFF, and two operating modes derived
in one cycle are as shown in Fig. 5(a), (b). Since the converter ⎧ n
⎪ ∂ f1
⎪
⎪ = 0, n ≥ 1
operates in a high frequency state, the average state space model ⎪
⎪ ∂x n
⎪
⎪
1
of the step-down chopper circuit can be equivalent to Fig. 5(c), ⎪
⎪ ∂f1 ∂ n f1
and the mathematical model is (13). ⎪
⎨ = β1 ; = 0, n ≥ 2
∂x2 ∂x2 n
 . (16)
1, t ∈ (nT, (n + D)T ) ⎪
⎪ ∂ n f1
⎪
⎪ = 0, n ≥ 1
S(t) = (12) ⎪
⎪ ∂x3 n
0, t ∈ ((n + D)T, (n + 1)T ) ⎪
⎪
⎪
⎪ ∂ n f1
⎧ ⎩ = 0, n ≥ 1
di ∂x4 n
⎪
⎨ L L + udc = Duin
dt . (13) The variable x4 and exy (x4 ) in the nonlinear system periodi-
⎪ du
⎩ C dc + im = iL cally changes with conduction intervals, and exy (x4 ) values are
dt as shown in Table I.

Authorized licensed use limited to: Chang Gung Univ.. Downloaded on September 09,2024 at 13:32:35 UTC from IEEE Xplore. Restrictions apply.
6152 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 67, NO. 7, JULY 2020

IV. DESIGN OF NONLINEAR STEADY SPEED CONTROLLER relationships ż4 = v, y = z1 = x4 .

⎧
A. Input and Output Feedback Linearization of ⎪
⎪ z1 = h = x4
⎪
⎪
Nonlinear Models ⎪
⎪ z2 = Lf h = x3
⎪
⎪
⎨
According to differential geometry theory and input–output z3 = L2 f h = f3
 
feedback linearization method, if affine nonlinear system ⎪ z4 = L3 f h = β5 exy (x4 )f2 + β6 exy 2 (x4 ) + β7 f3
⎪
⎪
⎪  
(15) can be linearized by input–output feedback method, the ⎪
⎪ dexy 2 (x4 )
⎪
⎪ dexy (x4 )
following two necessary and sufficient conditions must be ⎩ + β5 x2 + β6 x3 f4
satisfied. dx4 dx4
1) The rank of matrix [g, adf g,..., adn f g] is n. (20)
2) The set g, adf g,..., adn f g is involutive. ⎧ ⎡ ⎤ ⎡ ⎤
⎪
⎪
0 1 0 0 0
By the calculation of the Lie brackets, a fourth-order square ⎪
⎪ ⎢0 0 1 ⎥ ⎢ ⎥
⎪
⎨ ⎢ 0⎥ ⎢0⎥
matrix can be obtained as (17) ż = ⎢ ⎥z + ⎢ ⎥v
⎣0 0 0 1⎦ ⎣0⎦ . (21)
  ⎪
⎪
⎪
⎪
g adf g ad2 f g ad3 f g ⎪
⎩  0 0 0 0 1
⎡ ⎤ y= 1 0 0 0 z
β9 0 β1 β2 β9 −β1 β2 β3 β9
⎢ 0 β β ⎥ (17) The relationship of input variables between two systems is as
⎢ 2 9 β2 β3 β9 β10 ⎥ following:
=⎢ ⎥
⎣ 0 0 β2 β5 β9 exy (x4 ) β11 ⎦
m(x) v
0 0 0 −β2 β5 β9 exy (x4 ) u=− + (22)
n(x) n(x)
where
where 
m(x) = L4 f h
  . (23)
β10 = −β2 β9 β1 β2 + β3 2 + β4 β5 exy 2 (x4 ) n(x) = Lg L3 f h = β2 β5 β9 exy (x4 )
de x y (x 4 )
x3 − B. Steady Speed Controller Design
β11 = β2 β5 β9  dx 4
2
 .
exy (x4 ) β3 + β6 exy (x4 ) + β7
The BLDCM of MSCMG normally operates at a steady
speed. Therefore, the motor speed is chosen as the control target,
As shown in Table I, exy (x4 ) is never equal to zero. Therefore, and the reference speed is set to the rated speed of MSCMG.
the fourth-order square matrix has no diagonal element of zero Assuming that e is the rotational speed difference, and it is a
and its rank is 4, which satisfies the first condition. constant value under steady speed operation mode. Thus, there
Then, the Lie bracket of g, adf g, and ad2 f g are derived are (24) in detail
as (18). ⎧ 
⎪
⎪ yr ef − y = θer ef − θe = edt
⎪
⎪
⎪  
g = adad f g g = adad 2 f g g = 0 (18) ⎨ y r ef − y = ωer ef − ωe = e
⎪
y r ef − y  = −y 

. (24)
⎪
⎪
⎪ y r ef − y = −y
⎪ (3) (3) (3)
According to the differential geometry theory, it can be con- ⎪
⎪
cluded that the set g, adf g,..., adn f g is involutive, which satis- ⎩ (4)
y r ef − y (4) = −y (4)
fies the second condition.
The closed loop controller can be designed as (25) according
Equation (19) can be obtained by recursive calculation of
to the system order
higher-order Lie derivatives
 (4)   
y r ef − y (4) + c1 y (3) r ef − y (3)
⎧     = 0.
⎪ L h=0
⎪ g +c2 y  r ef − y  + c3 y  r ef − y  + c4 (yr ef − y)
⎪
⎨L L h = 0
g f (25)
. (19)
⎪ Lg L2 f h = 0
⎪
The control rule of linear canonical system is designed as
⎪
⎩ 
Lg L3 f h = β2 β5 β9 exy (x4 ) = 0 (4) (3) 
v = ż4 = y = c1 y − c2 y + c3 e + c4 edt (26)

It can be seen that the relative degree of the system is where, c1 − c4 are the coefficients to be determined.
4, which means that the system can achieve whole feed- Then, by substituting (26) into (22), the input variable u of the
back linearization. The original nonlinear system can be trans- actual buck converter switching device can be calculated. The
formed into a canonical form equation system as shown in parameter c1 − c4 can be adjusted by the existing mature control
(21) by coordinate transformation according to (20). In the theory and methods according to the linearized system. Note that
new system, z is the state variable, v is the input variable, unlike the time-varying parameters in traditional PID method
y is still the output variable, and there are the following for dealing with time-varying model, once the parameters are

Authorized licensed use limited to: Chang Gung Univ.. Downloaded on September 09,2024 at 13:32:35 UTC from IEEE Xplore. Restrictions apply.
CHEN AND LIU: SENSORLESS OPTIMAL COMMUTATION STEADY SPEED CONTROL METHOD 6153

Fig. 6. Schematic diagram of BLDCM control system based on buck Fig. 7. MSCMG motor experimental system.
converter.

tuned, the parameters will not need to change with six different
conduction intervals.

C. Overall Controller Structure

The overall schematic diagram of BLDCM control system
based on buck converter is shown in Fig. 6. The continuous Fig. 8. LPF circuit structure.
angular position and speed information for steady-speed
control is estimated from the ZCPs detected by the sensorless TABLE II
virtual neutral point method according to proposed optimal MSCMG MOTOR PARAMETERS
commutation. The voltage and current signals on the buck
converter side are directly measured and collected by the
analog to digital (AD) converter. In addition, the back-EMF,
resistance and other parameters of the motor are given by offline
measurement. The inductance and capacitance parameters of
the buck converter are selected according to the conditions for
ensuring the continuous current mode. Then, state variables
and parameters of the nonlinear system are calculated, and
the nonlinear system are linearized based on (20) and (21).
Finally, the input variable v of the linear system calculated
based on (22) is converted to the actual input duty cycle D of
the nonlinear system to control the power device.

V. EXPERIMENT RESULTS
A. Setup
for MSCMG are listed in Table II. fc is the LPF cutoff fre-
Fig. 7 illustrates the configuration of MSCMG system for quency. The phase-frequency response characteristic are shown
experiments. The BLDCM is built inside the gyro room and as Fig. 9.
its control system is integrated in controller cabinet. The mi-
croprocessor of the experimental control system adopts TI’s B. Actual Measured Back-EMF Experiment
28335 digital signal processor (DSP), which is mainly respon-
The actual measured back-EMF waveform for the motor used
sible for the calculation of the proposed method. In addition,
in MSCMG are shown in Fig. 10.
the control system is also equipped with a Microsemi’s A3P250
It can be seen that the three-phase back-EMF is nonstan-
field-programmable gate array (FPGA), which is responsible
dard trapezoidal and asymmetrical. By offline measurement,
for logic operations such as commutation control. AD conver-
the back-EMF parameters as in (2) can be obtained, and the
sion is realized by a 12-bit high-precision AD converter built
detailed results are listed in Table III.
in the DSP. As the terminal voltage ZCPs are sensitive to the
interference, before the signal enters AD, a special low-pass
filter as Fig. 8 is designed to filter the signal to avoid ZCP false C. Verification of Zero-Crossing Point of
Nonideal Back-EMF
detection. Where R1 and R2 are the divider resistances, C is a
filter capacitance, uxn and uxn  are the terminal voltage before The validation of the influence of nonideal back-EMF on the
and after LPF, respectively. Detailed parameters of the BLDCM ZCPs detection by different methods is carried out directly on

Authorized licensed use limited to: Chang Gung Univ.. Downloaded on September 09,2024 at 13:32:35 UTC from IEEE Xplore. Restrictions apply.
6154 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 67, NO. 7, JULY 2020

TABLE IV
COMPARISON OF COMMUTATION POINTS

Fig. 9. LPF phase-frequency response characteristic.

Fig. 10. Actual measured three-phase back-EMF.

TABLE III
DETAILED MEASURED BACK-EMF PARAMETERS

Fig. 12. Comparison experiments of different methods. (a) Nonphase

compensation PI speed control method. (b) Nonphase compensation
nonlinear speed control method. (c) Traditional phase compensation PI
speed control method. (d) Traditional phase compensation nonlinear
speed control method. (e) Optimal phase compensation PI speed con-
trol method. (f) Optimal phase compensation nonlinear speed control
method

the natural frequency of new system relatively large. The four

parameters c1 − c4 are set to about 12 200, 3500, and 10 000,
respectively. As shown in Fig. 12, the phase current waveform,
electromagnetic torque, and velocity waveform are shown in
each figure. Fig. 12(a) is nonphase compensation proportional
integral (PI) speed control method; Fig. 12(b) is nonphase com-
pensation nonlinear speed control method; Fig. 12(c) is tradi-
tional phase compensation PI speed control method; Fig. 12(d) is
traditional phase compensation nonlinear speed control method;
Fig. 11. Comparison of the ZCPs of different methods.
Fig. 12(e) is optimal phase compensation PI speed control
method; Fig. 12(f) is optimal phase compensation nonlinear
speed control method. From the experimental results shown in
the BLDCM. First, the measured three-phase back-EMF ZCPs
Fig. 12, according to Fig. 12(a) and (b), it can be seen that,
waveforms are given in Fig. 11.
comparing with PI speed control method, the peak–peak value
Then, the calculated commutation points are as Table IV. It
of phase current in nonlinear speed control method increases
can be seen that the results are consistent with the theoretical
due to the need to adapt to the nonideal back-EMF resulting
analysis and calculation in Section II.
in increased torque ripple, but the torque became smoother and
the steady-state speed ripple decrease significantly. Fig. 12(c)
D. Comparison Experiments by Different Methods and (d) show that both under the traditional phase compen-
Comparison experiments are carried out by different meth- sation method, when the nonlinear speed control is used, the
ods according to phase control and speed control at rated speed peak–peak current increases from 1.45 to 1.49 A, the torque
(10 000 rpm). In order to improve the response speed, we set ripple decreases by 32% and the speed stabilization accuracy

Authorized licensed use limited to: Chang Gung Univ.. Downloaded on September 09,2024 at 13:32:35 UTC from IEEE Xplore. Restrictions apply.
CHEN AND LIU: SENSORLESS OPTIMAL COMMUTATION STEADY SPEED CONTROL METHOD 6155

Fig. 13. Phase current waveforms for unused and used proposed Fig. 15. Experimental diagram of applying the proposed method in the
method at 100 rpm. dynamic process of acceleration.

value, and the current waveform is more suitable for non-

ideal back-EMF waveforms to produce a smoother torque.
At very low speeds, the phase delay caused by the low-pass
filter is small, and the improvement of the current peak-to-
peak-peak value is not obvious by the applied method, but
there will be some improvement from the waveform point
of view.

F. Dynamic Experiment Results

In order to demonstrate the performance of the proposed
method under the speed-up dynamics, we designed a dynamic
experiment during the speed-up process of the motor. The ex-
Fig. 14. Phase current waveforms for unused and used proposed perimental results in Fig. 15 show that in the process of motor
method at 12 000 rpm. speed-up, the proposed method is turned ON and OFF several
times, and the phase current of the motor changes consequently.
When the method is enabled, the pulsation of the phase current
increases significantly. According to Fig. 12(e) and (f), compar- is reduced, and the waveform adapts to the nonideal back EMF
ing with PI speed control method, the peak–peak value of phase waveform to obtain a smoother electromagnetic torque. It can be
current in nonlinear speed control method increases from 1.41 found that the proposed method is effective during acceleration.
to 1.45 A due to the need to adapt to the nonideal back-EMF, In addition, since the experimental motor is built in the gyro
but the torque ripple decreases by 10% and the steady speed ac- room and supported by the magnetic bearing, and the rated
curacy became higher. As for optimal phase control, according working environment is in a vacuum state, there is no contact
to Fig. 12(a), (c), (e), and, (b), (d), (f), the peak–peak value of with the outside world, and the traditional load torque test is
phase current decreases by about 16% and 3% when using opti- difficult to carry out. In order to demonstrate the performance
mal phase control method compared with the noncompensation of the proposed method, we designed a method to simulate the
and traditional compensation method, respectively. At the same load change by controlling the vacuum conditions in the gyro
time, the accuracy of electromagnetic torque and steady speed room. Fig. 16 shows the dynamic waveform of the designed
accuracy is also improved significantly. experiment. Because it is difficult to control the condition of
simulating load change through vacuum environment, the sim-
E. Maximum and Very low Speed Experiment Results ulated load will jitter when it increases.
The experimental results of other rotational speeds have been
G. Commutation Accuracy Experiment Results of
carried out, including the maximum speed (12 000 rpm for the
experimental motor) and the low speed (100 rpm in this ex- Proposed Method
periment, since the proposed method is not used in the start- Commutation accuracy of the proposed method are verified
ing process, the minimum speed in this article depends on in the experiments, and comparison is conducted on the uncom-
the starting method proposed in [15]). Figs. 13 and 14 show pensated sensorless, traditionally compensated sensorless, and
the phase current waveforms at low speed and high speed, optimal commutation signal. In the test, the commutation sig-
respectively. It can be seen that at high speed, the proposed nals generated by uncompensated sensorless method, tradition-
method can significantly reduce the current peak-to-peak-peak ally compensated sensorless and optimal commutation signal

Authorized licensed use limited to: Chang Gung Univ.. Downloaded on September 09,2024 at 13:32:35 UTC from IEEE Xplore. Restrictions apply.
6156 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 67, NO. 7, JULY 2020

sensorless method was applied, and then compared the

fluctuation of electromagnetic torque driven by square wave
current when commutating according to several different
commutation points. Then, optimal commutation control
method was proposed based on aforementioned work. In
addition, for the nonstandard trapezoidal back-EMF and the
buck converter front-end drive structure of small inductance
motor, an integrated nonlinear model of motor with precise
back-EMF parameters was established. Based on the nonlinear
model, the feasibility of model linearization was analyzed. A
linearization method using input–output feedback method is
proposed, and a steady speed controller based on the linearized
model was designed. Finally, experimental results validated the
Fig. 16. Experimental diagram of applying the proposed method in the effectiveness of the proposed optimal commutation control and
dynamic process of load torque change. steady speed controller based on linearization method.

REFERENCES
[1] I. Petrov and J. Pyrhonen, “Performance of low-cost permanent mag-
net material in PM synchronous machines,” IEEE Trans. Ind. Electron.,
vol. 60, no. 6, pp. 2131–2138, Jun. 2013.
[2] G. Buja, M. Bertoluzzo, and R. K. Keshri, “Torque ripple-free operation
of PM BLDC drives with petal-wave current supply,” IEEE Trans. Ind.
Electron., vol. 62, no. 7, pp. 4034–4043, Jul. 2015.
[3] Q. Hu and J. Zhang, “Attitude control and vibration suppression for flex-
ible spacecraft using control moment gyroscopes,” J. Aerosp. Eng., vol.
29, no. 1, Jan. 2016, Art. no. 04015027.
[4] W. Li, J. Fang, and H. Li, and J. Tang, “Position sensorless control without
phase shifter for high-speed BLDC motors with low inductance and non-
ideal back EMF,” IEEE Trans. Power Electron., vol. 31, no. 2, pp. 1354–
1366, Feb. 2016.
[5] P. P. Acarnley and J. F. Watson, “Review of position-sensorless operation
of brushless permanent-magnet machines,” IEEE Trans. Ind. Electron.,
vol. 53, no. 2, pp. 352–362, Apr. 2006.
Fig. 17. Accuracy comparison between different commutation method. [6] G. A. Capolino and A. Cavagnino, “New trends in electrical machines
technology-Part II,” IEEE Trans. Ind. Electron., vol. 61, no. 9, pp. 4931–
4936, Sep. 2014.
[7] S. Tsotoulidis and A. N. Safacas, “Deployment of an adaptable sensor-
are displayed successively, and the six three-phase full bridge less commutation technique on BLDC motor drives exploiting zero se-
commutation signals are synthesized together to form a syn- quence voltage,” IEEE Trans. Ind. Electron., vol. 62, no. 2, pp. 877–886,
thesized signal S (S = SaSb + SbSc + ScSa). In this way, Feb. 2015.
[8] J. Shi and T.-C. Li, “New method to eliminate commutation torque ripple
the six commutation signals generated by different sensors can of brushless DC motor with minimum xommutation time,” IEEE Trans.
be presented simultaneously, and the synthesized signal fre- Ind. Electron., vol. 60, no. 6, pp. 2139–2146, Jun. 2013.
quency is tripled compared with original signal as shown in [9] Park. S. J, Park. H. W and Lee. M. H, “A new approach for minimum-
torque-ripple maximum-efficiency control of BLDC motor,” IEEE Trans.
Fig. 17. Before compensation, the phase difference between Ind. Electron., vol. 47, no. 1, pp. 109–114, Feb. 2000.
uncompensated sensorless and optimal commutation signals is [10] T. Sheng, X. Wang, and J. Zhang, and Z. Deng, “Torque-Ripple miti-
about 20◦ at tested speed (synthesized signals phase difference gation for brushless DC machine drive system using one-cycle average
torque control,” IEEE Trans. Ind. Electron., vol. 62, no. 4, pp. 2114–2122,
is 60◦ = 20◦ × 3 as presented in the oscilloscope waveform). Apr. 2015.
Then, starting the optimal commutation, it can be found that [11] S. J. Kang and S. K. Sul, “Direct torque control of brushless DC motor with
the phase difference converges within one cycle time, and the nonideal trapezoidal back EMF,” IEEE Trans. Power Electron., vol. 10,
no. 6, pp. 796–802, Nov. 1995.
phase difference converges to near zero. It can be noted that the [12] H. Lu, L. Zhang, and W. Qu, “A new Torque control method for torque
proposed method achieves a high accuracy of the commutation ripple minimization of BLDC motors with un-ideal back EMF,” IEEE
signal adapted to nonideal back-EMF, while the conventionally Trans. Power Electron., vol. 23, no. 2, pp. 950–958, Mar. 2008.
[13] Y. Liu, Z. Q. Zhu, and D. Howe, “Direct torque control of brushless DC
compensated sensorless signal does not eliminate the effect of drives with reduced torque ripple,” IEEE Trans. Ind. Appl., vol. 41, no. 2,
the ideal back EMF on the zero crossing detection and will result pp. 599–608, Mar./Apr. 2005.
in small current ripple as shown in Fig. 17. [14] Y. Liu, Z. Q. Zhu, and D. Howe, “Instantaneous torque estimation in
sensorless direct-torque-controlled brushless DC motors,” IEEE Trans.
Ind. Appl.,vol. 42, no. 5, pp. 1275–1283, Sep./Oct. 2006.
VI. CONCLUSION [15] J. Fang, X. Zhou, and G. Liu, “Instantaneous torque control of small
inductance brushless DC motor,” IEEE Trans. Power Electron., vol. 27,
Aiming at the small inductance BLDCM including nonstan- no. 12, pp. 4952–4964, Dec. 2012.
[16] J. Feng, K. Liu, and Q. Wang, “Scheme based on buck-converter with
dard trapezoidal wave and three-phase asymmetric back-EMF, three-phase H-bridge combinations for high-speed BLDC motors in
this article first analyzed the influence of nonideal back-EMF aerospace applications,” IET Electric Power Appl., vol. 12, no. 3, pp. 405–
on the position of the detected commutation points when 414, Mar. 2018.

Authorized licensed use limited to: Chang Gung Univ.. Downloaded on September 09,2024 at 13:32:35 UTC from IEEE Xplore. Restrictions apply.
CHEN AND LIU: SENSORLESS OPTIMAL COMMUTATION STEADY SPEED CONTROL METHOD 6157

[17] X. Chang, Y. Li, and W. Zhang, Nan Wang, and Wei Xue, “Active dis- [27] C. L. Xia, Q. Geng, X. Gu, T. N. Shi, and Z. F. Song,
turbance rejection control for a flywheel energy storage system,” IEEE “Input-output feedback linearization and speed control of a surface
Trans. Ind. Electron., vol. 62, no. 2, pp. 991–1001, Feb. 2015. permanent-magnet synchronous wind generator with the Boost-Chopper
[18] C. Vlad, P. Rodriguez-Ayerbe, E. Godoy, and P. W. Lefranc, “Advanced converter,” IEEE Trans. Ind. Electron., vol. 59, no. 9, pp. 3489–3500,
control laws of DC-DC converters based on piecewise affine modelling. Sep. 2012.
Application to a step-down converter,” IET Power Electron., vol. 7, no. 6,
pp. 1482–1498, Jun. 2014.
[19] S. C. Tan, Y. M. Lai, and C. K. Tse, “A unified approach to the design of
PWM-based sliding-mode voltage controllers for basic DC-DC converters Xi Chen received the B.S. degree in automa-
in continuous conduction mode,” IEEE Trans. Circuits Syst. I Reg. Papers, tion from Shandong University, Jinan, China, in
vol. 53, no. 8, pp. 1816–1827, Aug. 2006. 2014. He is currently working toward the Ph.D.
[20] V. Mummadi, “Design of robust digital PID controller for H-bridge degree in precision instruments and machin-
softswitching boost converter,” IEEE Trans. Ind. Electron., vol. 58, no. 7, ery with the School of Instrumentation Science
pp. 2883–2897, Jul. 2011. and Optoelectronics Engineering, Beihang Uni-
[21] F. Luo and D. S. Ma, “Design of digital tri-mode adaptive-output buck- versity, Beijing, China. His research interests
boost power converter for power-efficient integrated systems,” IEEE include the spacecraft attitude actuator control
Trans. Ind. Electron., vol. 57, no. 6, pp. 2151–2160, Jun. 2010. and motor control.
[22] R. P. Burgos, E. P. Wiechmann, and J. Holtz, “Complex state-space mod-
eling and nonlinear control of active front-end converters,” IEEE Trans.
Ind. Electron., vol. 52, no. 2, pp. 363–377, Apr. 2005.
[23] C. Liu, G. Liu, and J. Fang, “Feedback linearization and extended state
observer based control for rotor-AMBs system with mismatched uncer- Gang Liu received the B.S. and M.S. degrees in
tainties,” IEEE Trans. Ind. Electron., vol. 64, no. 2, pp. 1313–1322, Feb. industrial automation from Shandong University,
2017. Jinan, China, in 1992 and 1998, respectively,
[24] H. Abootorabi Zarchi, J. Soltani, and G. Arab Markadeh, “Adaptive input- and the Ph.D. degree in water conservancy and
output feedback-linearization-based torque control of synchronous re- hydropower engineering from the Dalian Univer-
luctance motor without mechanical sensor,” IEEE Trans. Ind. Electron., sity of Technology, Dalian, China, in 2001.
vol. 57, no. 1, pp. 375–384, Jan. 2010. He is currently a Ph.D. Supervisor with
[25] G. Oriolo, A. De Luca, and M. Vendittelli, “WMR control via dynamic the School of Instrumentation Science and Op-
feedback linearization: design, implementation, and experimental valida- toelectronics Engineering, Beihang University,
tion,” IEEE Trans. Control Syst. Technol., vol. 10, no. 6, pp. 835–852, Beijing, China. His research interests include
Nov. 2002. spacecraft attitude control, high speed magnet-
[26] H. K. Khalil, Nonlinear Systems. Upper Saddle River, NJ, USA: Prentice- ically suspended motor, and ultra-high sensitivity atomic magnetic field
Hall, 2002. measurement.

Authorized licensed use limited to: Chang Gung Univ.. Downloaded on September 09,2024 at 13:32:35 UTC from IEEE Xplore. Restrictions apply.