1576 IEEE TRANSACTIONS ON MAGNETICS, VOL. 41, NO.

5, MAY 2005

A Phase Variable Model of Brushless dc Motors

Based on Finite Element Analysis and Its Coupling
With External Circuits
Osama A. Mohammed, Fellow, IEEE, S. Liu, Senior Member, IEEE, and Z. Liu
Department of Electrical and Computer Engineering, Energy Systems Laboratory, Florida International University,
Miami, FL 33174 USA
This paper presents a fast and accurate brushless dc motor (BLDC) phase variable model for drive system simulations. The developed
model was built based on nonlinear transient finite-element analysis to obtain the inductances, back electromotive force as well as the
cogging torque. The model was implemented in a Simulink environment through the creation of an adjustable inductance component
to account for the dependence of inductances on rotor position. Since no model for BLDC actually exists, the significance of this
work is that it provides an accurate equivalent circuit model of BLDC motors for utilization in simulation environments. Using the
developed model, the sensorless control and the torque ripple control issues were investigated and the simulation results show its practical
effectiveness.
Index Terms—Brushless dc motor (BLDC), finite element analysis (FEA), motor control, phase variable model.

I. INTRODUCTION This gives accurate results but is time consuming. Using the de-
veloped phase variable model, the drive system simulation
A CCURATE and efficient simulation of brushless dc motor
(BLDC) machines, driven by power electronic switching
devices, is important for drive system design and optimization
behaves much faster with the same level of accuracy.
The equation-based Simulink phase variable model is intro-
[1]–[5]. Two key issues related to this topic are the machine duced by the authors in the context of PM synchronous machine
modeling and the coupling between the machine model and ex- implementation [6]. In order to connect the equation-based model
ternal circuits. and external circuits, line voltage must be measured. For BLDC
A fast and accurate machine model is always desirable. Com- machines, each commutation sequence has one winding that is
pared with an equivalent electric circuit model, the finite ele- energized to positive power (current enters into the winding), one
ment (FE) description is more accurate but can be time con- winding is deenergized (current exits the winding) and the third
suming. winding is in a nonenergized condition. This means that there al-
Two types of circuit models are available for machines: the ways exists one phase which is open circuited. The input voltage
-model and the phase variable model. BLDC has a trape- to this phase is immeasurable. This shows that the equation based
zoidal back electromotive force (EMF) and requires rectangular model cannot be applied to BLDC. An alternative model, com-
stator currents to produce constant torque. The variation of the posed of circuit components, is built to implement the phase
self and mutual inductances of the stator windings is nonsinu- variable model of BLDC presented here. An adjustable induc-
soidal. No particular advantage exists in transforming the tance component is developed to represent the inductance depen-
equations to the frame. The commonly used model as- dence on the rotor position.
sumes that the self and mutual inductances are constant [1]. Due
to the physical rotation of the rotor and the nonlinear magneti- II. PHYSICAL PHASE VARIABLE MODEL
zation property of stator iron, the inductance varies with rotor
The phase variable model of BLDC machines is given as
position and winding current. Since the magnetic field of perma-
nent magnet (PM) machines is mainly established by permanent
magnets, the effects of current are usually ignored. The rotor (1)
position dependence of inductances can be accurately evalu- (2)
ated through nonlinear transient FE analysis. Similarly, the rotor
position dependence of the back EMF and the cogging torque (3)
can be calculated from nonlinear transient FE analysis as well.
and (4)
Using these rotor position dependent parameters, the physical
phase variable model of BLDC is developed.
where is the back EMF, is the cogging torque,
For dynamic performance studies, the time-stepping FE pro-
is the flux linkage contributed by the stator winding, and
cedure strongly couples the circuit equation with the FE motor
is the matrix of apparent inductance. The rest variables are used
equations then solve the coupled system simultaneously [3]–[5].
as their conventional meanings.
The , , and profiles are obtained from the non-
linear transient FE solutions, in which the rotor position depen-
Digital Object Identifier 10.1109/TMAG.2005.845042 dence as well as the saturation effect are considered.
0018-9464/$20.00 © 2005 IEEE

Authorized licensed use limited to: RMIT University. Downloaded on February 24,2010 at 09:00:07 EST from IEEE Xplore. Restrictions apply.
MOHAMMED et al.: A PHASE VARIABLE MODEL OF BDCM 1577

Fig. 1. Circuit diagram of BLDC stator phase “a” winding.

III. SIMULINK IMPLEMENTATION

A. Voltage Equation
Fig. 2. Adjustable self inductance block.
Expanding the derivative term of (1), one has

(5)
Substituting (2) into (5) and considering that the current in-
dependence of the winding inductances, one can obtain

(6)

Fig. 3. Equivalent circuit of phase “a” inductive voltage drop.

where are incremental inductances.
Substituting (6) into (1), the voltage (1) becomes
The mutual inductance voltage drop in the second term of (7),
( , ), are represented by CVSs.
Fig. 3 is the equivalent circuit of phase “a” inductive voltage
drop. The current passing through the weighted line is . The
inductances , , and are retrieved from the incre-
(7) mental inductance table by picking the values corresponding to
a specific rotor position.
Based on (7), the circuit diagram is constructed. As an ex-
ample, Fig. 1 shows the diagram of phase “a” winding. The con- B. Torque Calculation
trolled voltage source (CVS) component is adopted to describe As the rotor speed is used as a denominator in the torque
the voltage drop due to the flux cutting by the moving rotor. calculation of (3), it causes problems at the initial simulation
It represents the summation of the third and forth terms of (7). step due to the zero rotor speed. In order to solve this problem,
The derivative of apparent inductances with respect to the ro- a very small number is assigned to when starting the simula-
tation angle ( , ), are calculated in advance tion. In addition, an initial value of the electromagnetic torque
and stored in look-up tables. , which is larger than the load torque is assigned to .
An adjustable inductance component is developed to describe Otherwise, according to (4), one knows that the motor will not
the self-inductance voltage drop of the BLDC ( , start moving. Consequently, the back EMF, , , and , equal
), seen in the second term of (7). As the inductance zero and there will not be output toque .
current is a state variable, the adjustable inductance component The cogging toque are stored in tables and retrieved ac-
is built according to the integral description of inductance cording to the rotor position.

(8)
C. Phase Variable Model
Using as an example, the circuit diagram of the devel- The developed physical phase variable model is shown in
oped adjustable inductance is illustrated in Fig. 2. The initial Fig. 4. Subsystem 1 is the implementation of voltage (1), as in-
value of the integrator is set to zero. The voltage measurement troduced in section A. This subsystem is built using
block (VM) and the controlled current source (CCS) are used to and in form of tables. Subsystem 2 performs
realize the connection of the adjustable inductance component the torque calculation according to (2). A value of 3.5 Nm, max-
and the external circuits. imum torque of this BLDC motor, is used as the initial torque to

Authorized licensed use limited to: RMIT University. Downloaded on February 24,2010 at 09:00:07 EST from IEEE Xplore. Restrictions apply.
1578 IEEE TRANSACTIONS ON MAGNETICS, VOL. 41, NO. 5, MAY 2005

Fig. 4. Developed physical phase variable model for BLDC.

Fig. 6. (a)–(b) From full FE model. (c)–(d) From phase variable model.
Fig. 5. (a) Self inductance L . (b) Per-unit speed back EMF. (c) Cogging
torque. (d) dL =d .

start the rotor’s motion. It lasts 0.1ms, which is controlled by a

step function block. The motion (4) is performed at the right por-
tion of Fig. 4. The two tables in Fig. 4 are the cogging torque and
back EMF. The unit speed back EMF is stored
in the back EMF table to perform speed control simulation.
The table data are obtained from nonlinear transient FE anal-
ysis. In order to obtain high accuracy, attentions to the rotating
air gap mesh and the time step in transient FE analysis are given.
Nodes on the lateral dimension of the rotating air gap must be
evenly distributed. The time step of transient analysis should be
kept as the time required for moving the radial angle between
two contingent nodes.

IV. MODEL VERIFICATION

As an example, the phase variable model of a 4-pole 24-slot
24-V BLDC motor is built. The incremental inductance ,
per-unit speed back EMF, cogging torque, and the derivative of Fig. 7. Zero crossing detection of back EMF.

Authorized licensed use limited to: RMIT University. Downloaded on February 24,2010 at 09:00:07 EST from IEEE Xplore. Restrictions apply.
MOHAMMED et al.: A PHASE VARIABLE MODEL OF BDCM 1579

alternative way for the verification of control strategies. Fur-

thermore, with the developed model, the accuracy of the control
strategy can be investigated. In addition, the inverter parameter
design can be studied. Fig. 7(b) demonstrates the consequence
of inappropriate inverter parameters to the sensorless control
strategy of [7]. In this case, the zero crossing point cannot be
identified properly.

B. Torque Ripple Reduction

The ripple torque occurs as the results of fluctuations of the
field distribution and the armature MMF. Using the developed
physical phase variable model, the fluctuations of the field dis-
tribution and the armature MMF can be captured. In this way,
the torque ripple control can be studied under the real situa-
tions. As an example, the phase variable model is applied to
a ripple torque forward-fed compensation method. The results
are shown in Fig. 8, which demonstrates the effects of the torque
ripple control. The torque ripple is greatly reduced and the speed
becomes smooth.

VI. CONCLUSION
Fig. 8. (a), (c), and (e): without torque ripple control; (b), (d), and (f): with A physical phase variable model of BLDC machines is pro-
torque ripple control.
posed. Its parameters are obtained from transient FE analysis
of the machine. Simulink implementation details of the pro-
mutual inductance with respect to rotor position are posed model are presented. The proposed phase variable model
shown in Fig. 5. has the accuracy of the full FE model with much faster simula-
The performance of the developed physical phase variable tion speed. To maintain the accuracy of the proposed model, the
model is examined by comparing it with the full FE model in inductance was distinguished as apparent and incremental in-
an electric commutation circuit. The obtained torque and three ductances. The per-unit speed back EMF was used as the back
phase current profiles during the starting process are given in EMF table data. The rotating air gap meshing and the time step-
Fig. 6. It shows that the developed physical phase variable model ping FE analysis were related. The effectiveness of the proposed
is capable of providing the same dynamic simulation character- phase variable model was tested through implementation of sen-
istics as the full FE model. sorless control and torque ripple reduction examples.

V. APPLICATIONS REFERENCES
Sensor-less control and torque ripple minimization of ma- [1] P. Pillay and R. Krishnan, “Modeling, simulation, and analysis of perma-
nent-magnet motor drives, part 2: The brushless dc motor drive,” IEEE
chines are two topics of major interess by researchers. The con- Trans. Ind. Appl., vol. 25, no. 2, pp. 274–279, Mar./Apr. 1989.
trol strategy study is usually performed via experiments and/or [2] M. A. Inerfield, S. L. Garverick, W. S. Newman, and Y. Zhang, “A
computer simulation. The computer simulation is the only pos- SPICE model for a novel brushless adjustable-speed drive,” IEEE Trans.
Ind. Electron., vol. 47, no. 6, pp. 1307–1318, Dec. 2000.
sible way in the machine design stage. A fast and accurate model [3] S. L. Ho, W. N. Fu, H. L. Li, H. C. Wong, and H. Tan, “Performance
facilitates such a study. analysis of brushless dc motors including features of the control loop
in the finite element modeling,” IEEE Trans. Magn., vol. 27, no. 5, pp.
3370–3374, Sep. 2001.
A. Sensorless Control [4] M. A. Jabbar, H. N. Phyu, Z. Liu, and C. Bi, “Modeling and numerical
The zero-crossing instance of back EMF waveform can simulation of a brushless permanent-magnet dc motor in dynamic con-
ditions by time-stepping technique,” IEEE Trans. Ind. Appl., vol. 40, no.
be utilized to perform senseless motor control. As the back 3, pp. 763–770, May/Jun. 2004.
EMF waveform of the motor cannot be directly measured, [5] G. H. Jang, J. H. Chang, D. P. Hong, and K. S. Kim, “Finite-element anal-
various methods were proposed to identify the back EMF ysis of an electromechanical field of a BLDC motor considering speed
control and mechanical flexibility,” IEEE Trans. Magn., vol. 38, no. 2,
zero-crossing. [7] introduced a method to find the zero-crossing pp. 945–948, Mar. 2002.
instance through the phase to ground PWM waveform, which [6] O. A. Mohammed, S. Liu, and Z. Liu, “Physical modeling of PM syn-
was verified by experiments. In this paper, we provide the chronous motors for integrated coupling with machine drives,” IEEE
Trans. Magn., vol. 41, no. 5, pp. 1628–1631, May 2005.
simulation verification of their zero-crossing detection method [7] T. Shao, D. Nolan, M. Teissier, and D. Swanson, “A novel micro-
by using the developed physical phase variable model. The controller-based brushless dc(BLDC) motor drive for automotive
back EMF waveform obtained from simulation as well as the fuel pumps,” IEEE Trans. Ind. Appl., vol. 39, no. 6, pp. 1734–1740,
Nov./Dec. 2003.
switching off process of PWM is shown in Fig. 7(a), which
reproduces the experiment results given in [7]. Therefore, we
can say that the developed phase variable model provides an Manuscript received June 8, 2004.

Authorized licensed use limited to: RMIT University. Downloaded on February 24,2010 at 09:00:07 EST from IEEE Xplore. Restrictions apply.