1

Adaptive Controller for Improved Performance

of Brushless DC Motor
N. Leena, Member, IEEE, R. Shanmugasundaram, Member, IEEE

performance during such changes in operating conditions like

Abstract— This paper presents the development and variations in load, saturation, changes in parameters or noise
performance analysis of model reference adaptive controller propagations. This has resulted in an increased interest in
using Artificial Neural Network (ANN) for Brushless DC motor intelligent and adaptive controllers.
(BLDC) drives. The model reference adaptive systems (MRAS)
One of the major approaches to adaptive control is Model
have a parameter adjustment mechanism along with the normal
feedback loop and hence give better solutions when there are Reference Adaptive Control (MRAC).The objective is to
variations in process parameters. Neural networks (NNs) with design a controller with adjustable parameters so that the
their inherent parallelism, learning capabilities and fault behavior of the plant to be controlled follows a desired
tolerance have proven to be a promising solution in estimating behavior in spite of variations in plant parameters or other
and controlling nonlinear systems. This paper combines a MRAS uncertainties’.
with ANN to solve the problems of non-linearity, parameter
Artificial Neural Network “ANN” has been applied
variations and load excursions that occur in BLDC motor drive
systems. The performance of the traditional PID controller based successfully to a wide range of control system applications in
speed control method is compared with the model reference recent years. Artificial neural networks have high learning and
based speed control for BLDC motor drive system using nonlinear mapping essences and its parallel and distributed
MATLAB Simulink software. Simulation results are presented to structure can provide a nonlinear mapping between inputs and
prove that the MRAC based model is capable of speed tracking outputs of an electric drive system, without the knowledge of
as well as reduce the effect of parameter variations.
any predetermined model. This makes ANN a good choice to
Index Terms— Artificial neural network (ANN), Brushless DC be used in the adaptation mechanism of a MRAC system.
Motor, Model reference adaptive control (MRAC), PID In the proposed work, a speed control strategy for BLDC
controller. motor is proposed using a model reference adaptive controller
based on Artificial Neural Networks. The performances of the
proposed drive system and the conventional PID control are
I. INTRODUCTION evaluated at different operating conditions, such as sudden
load impact, parameter variations, etc.
T he industrial world of today is fast growing and so is the
demand for precision. There are several applications today
that demand high performance. Among the various motors,
The information gathered from the literature to carry out
this work is as follows. The modeling of brushless dc motor is
brushless dc motors are gaining widespread popularity in discussed in [2], [3].The effect of changes in motor parameters
HVAC industry, medical equipment’s, electric vehicles, and load disturbances on the performance of a brushless dc
aerospace, military equipment’s, hard disk drives, due to its motor drive are presented in [4],[5]. Several tuning methods
well-known advantages like high efficiency, high power factor for PID controllers are described in [7]-[9].The tuning method
and low maintenance. suggested in [9] is found to yield optimum performance and is
The conventional controllers used in high performance adapted in this work to determine PID controller gain
drives are proportional integral (PI) or proportional integral parameters. Robust and adaptive speed control of motor drives
derivative (PID). These are constant gain controllers and using ANN based speed controllers are reported in [14]-[17],
require accurate mathematical models or system response for [19].
their design. The BLDC motor drive is highly non-linear. It is
often very difficult to obtain an accurate mathematical model
for the motor using the conventional techniques. Furthermore, II. BLDC MOTOR DRIVE DYNAMICS
the properties of the motor are usually unknown and time- To design the artificial neural network based adaptive
varying. The conventional controllers fail to give optimal controller for BLDC motor drive, the modeling of BLDC
motor is essential. The mathematical model of BLDC motor is
N.Leena is working as Assistant Professor in theElectrical and Electronics represented in the form of mathematical equations. The
Engineering Department of Federal Institute of Science and BLDC motor drive system can be described by the following
Technology,Angamaly-683577, India (e-mail: [email protected]).
R.Shanmugasundaram is working as Assistant Professor in Electrical and
equations:
Electronics Engineering Department of Sri Ramakrishna Engineering College,
Coimbatore-641022, India (e-mail: [email protected]).

978-1-4673-2149-5/12/$31.00 ©2012 IEEE 117

 2

di a y(t) where r(t) is the reference input and y(t) is the output..
V a = R a i a + La + k bω m         (1) The equivalent transfer function in the s-domain is given by
dt
& 1 #
di u (t ) = K $1 + + Td ( s )! E ( s ) (9)
Vb = Rb ib + Lb b + k bω m (2)
dt % Ti s "
The PID controller shows a smaller maximum overshoot
and has no steady state error due to the integral action.
dic
Vc = Rc ic + Lc + kbω m (3) B. Simulation of PID control scheme for BLDC motor
dt
Fig.1 shows the MATLAB/ Simulink model of BLDC
dωm
Te = kt i = J + Bωm +T l (4) motor with PID controller. The most popular design technique
dt for PID controllers is Ziegler-Nichols method, which relies
where Ra, Rb, Rc are the per phase resistance of phase a, b solely on parameters obtained from the system step response.
and c respectively, La, Lb, Lc are the per phase self-inductance In this paper, the PID controller is designed according to the
of phase a, b and c respectively, m is the rotor speed, Va, Vb, method in [9] which also relies solely on parameters obtained
Vc is the per phase voltage and i= ia = ib = ic are the phase from plant step response. The values of Kp, Ki and Kd
current of phase a, b and c respectively, Te and Tl are calculated are 0.03, 0.0029 and 0.0024 respectively for the
electromagnetic torque developed by the motor and load motor parameters given in Appendix.
torque, J and B are inertia and friction coefficients.
Assuming the load to be a fan or propeller load, the relation
between the load torque and speed can be described by the
following relation:
Tl = µ (ω m ) 2 (5)
where µ is a constant used for modeling the nonlinear
mechanical load.
Equations (1)-(3) are combined and continuous quantities
replaced by finite difference equations to get,
ω m ( k + 1) = αω m ( k ) + βω m ( k + 1) −
γω m 2 ( k ) + δω m 2 ( k − 1) + ς V ( k ) (6)

where !, ", #, $ and % are constants that can be expressed in

terms of the motor parameters. Equation (4) can be further
modified to obtain the inverse dynamic model of the drive
system as: Fig.1 Simulink model of BLDC motor with PID controller
The system response obtained under different operating
V (k ) = f (ωm (k + 1), ω m (k ), ωm (k − 1)) (7) conditions like change in reference speed, change in inertia
Thus it can be seen that the control voltage is a non-linear and change in phase resistance for the load in (5) are shown in
function of three consecutive samples of motor speed. In the Fig.2-Fig.4. Fig.2 shows the speed response for a step change
above equation, one sample of predicted speed m(k+1)is in reference speed when the resistance value is doubled. It can
replaced by one sample of reference speed. The ANN can be seen that the system takes 30ms to reach steady state with a
learn this non-linearity between input and output percentage overshoot of 7% and zero steady state error.

III. PID CONTROLLER

A. Conventional control schemes of BLDC motor

The conventional controllers used for robust control are
mainly constant gain controllers, such as proportional integral
(PI) or proportional integral derivative (PID). The idealized
equation of a proportional-integral-derivative (PID) controller
is
t
1 de(t )
u (t ) = K (e(t ) + e(t )dt + Td ) (8)
Ti 0 dt
where K is the proportional constant, Ti is the integral time,
Td is the derivative time, and e(t) is the error; i.e., e(t) =r(t) –

2012 International Conference on Data Science & Engineering (ICDSE) 118

 3

Fig. 2. Speed response of PID control system with step change in reference
speed and change in resistance(R to 2R)
Fig.4 Speed response of PID control system with step change in reference
Fig.3 shows the speed response for step change in reference speed and change in parameters (R to 2R and J to 2J)
speed with an increase in inertia. It can be seen that an
It can be seen that the PID controllers can produce
increase in inertia from J to 2J increases the settling time to
satisfactory results for fixed parameter system. However in
50ms with a percentage overshoot of 10% and zero steady
practical systems, system parameters change during working.
state error.
Hence there is a need to find alternative control strategies like
neural controllers or fuzzy controllers to achieve the desired
performance.

IV. ADAPTIVE CONTROL OF BLDC MOTOR

A. Model Reference Adaptive Control

Artificial neural network has self-learning and self-
regulatory capability and are widely used for identification
and control of non-linear systems. The most fundamental
neural network based controllers are probably those using the
“inverse” of the process as the controller called direct inverse
control. Another technique is known as model reference
adaptive control (MRAC) and is widely used for control of
non-linear systems. It has the advantage that the plant
response can be driven towards a reference model response
Fig.3 Speed response of PID control system with step change in reference which gives the desired transient behavior. The architecture of
speed and change in inertia (J to 2J) model reference adaptive controller is shown in Fig.5.
The speed response of the system with step change in
reference speed and an increase in resistance and inertia to
double the initial values are shown in Fig.4. It can be seen that
the system response is slightly sluggish with a settling time of
80ms and a maximum overshoot of 23% and a steady state
error of 1rad/s.

Fig.5 Model reference controller architecture

2012 International Conference on Data Science & Engineering (ICDSE) 119

 4

B. Simulation of Model reference adaptive controller

The simulation of MRAC controller is carried out using the
Neural Networks Toolbox of MATLAB. This neural
controller has two neural networks; one is used to model the
system and is called the plant identifier and the other acts as
the controller. The identifier is a multi-layer neural network
with a back-propagation learning scheme. The controller is a
recurrent learning multilayer neural network. Using the
measured values from the plant, the identifier is trained
offline. This block estimates the behavior of the plant and its
output is used to generate model error as shown in Fig.5. The
controller is then trained so that the system response follows
that of the reference model (input to the system).The reference
model chosen is that of BLDC motor with PID controller
which is fine tuned to give optimum response. The BLDC
model is shown in Fig.6.
Fig.7 Performance of the neural network model reference controller after
training.

Fig.6 BLDC motor model

The magnitude of the error for the plant identification was Fig.8 MATLAB /Simulink model of BLDC motor with MRAC controller
of the order of 10e-9. The plant identifier network has one
delayed input and two delayed outputs and ten hidden layers. The system response as obtained under different operating
Ten thousand samples are used for the one thousand epochs of conditions like change in reference speed, change in inertia
training. The Levenberg-Marquardt algorithm is used for the and change in phase resistance and are shown in Fig.9-Fig.11.
training of the plant model. The controller training is trained Fig.9 shows the speed response for a step change in reference
after the plant identification. The BFGS (Broyden, Fletcher, speed when the resistance value is doubled. It can be seen that
Goldfarb, and Shanno) Quasi–Newton training algorithm is the system takes 12ms to reach steady state with a no
adopted here and the performance (magnitude of error) is of overshoot and zero steady state error.
the order of 10e-6. A network with thirteen hidden layers and
one delayed input and two delayed outputs, is chosen as the
neural network controller.
After the controller training is completed the results
obtained to evaluate the learning capability of the controller is
shown in Fig.7. The plant response clearly shows that the
neural network output clearly follows the response of the
reference model (desired output).

2012 International Conference on Data Science & Engineering (ICDSE) 120

 5

Fig.11 Speed response of a MRAC system with step change in reference

speed and change in resistance and inertia (R to 2R and J to 2J)
Fig.9 Speed response of a MRAC system with step change in reference speed
and change resistance(R to 2R)

Fig.10 shows the speed response for a step change in V. RESULTS AND DISCUSSION
reference speed with an increase in inertia. It can be seen that
The simulation studies clearly show that a neural network
an increase in inertia from J-2J increases the settling time to
based model referenced control algorithm improves transient
24ms with a percentage overshoot of 8.3% and steady state
and steady state behavior of BLDC motor over classical PID
error of 2.5 rad/s.
control technique. The PID and Model reference adaptive
controllers are both able to track the changes in reference
speed. However the MRAC technique performs better than the
PID controller during system parameter changes and is able to
track reference speed more accurately.
TABLE 1
PERFORMANCE PARAMETERS OF PID AND MRAC CONTROLLER
Performance Measures Maximum Steady
Settling Percentage State Error
time ts Overshoot ( rad/s)
(sec) (%)
Changes in operating Controller
conditions
Change in reference
speed with change in MRAC 0.012 Zero Zero
resistance
(R to 2R)
PID 0.03 7 Zero

Change in reference
Fig.10 Speed response of a MRAC system with step change in reference speed with change in MRAC 0.024 8.3 2.5
speed and change in inertia (J to 2J) inertia
(J to 2J)
The speed response of the system to step change in
reference speed with an increase in resistance and inertia to PID 0.05 10 Zero
double the initial values are shown in Fig.11. It can be seen
that the system response is slightly sluggish with a settling Change in reference
time of 40ms with no overshoot or steady state error. speed with change in MRAC 0.04 Zero Zero
resistance and inertia
(R to 2R and J to 2J)

PID 0.08 23 1

From the table above, it can be inferred that BLDC drive

2012 International Conference on Data Science & Engineering (ICDSE) 121

 6

employing Model Reference Adaptive Controller is able to [9] J. C. Basilio and S. R. Matos, “Design of PI and PID Controllers with
Transient Performance Specification’, IEEE Transactions on
respond with smaller settling time and lesser overshoot and Education, vol. 45, no 4, pp 364-370.
steady state error than the one employing PID controller when [10] S.N. Balakrishnan and R.D. Weil, “Neurocontrol : A Literature Survey,”
there are changes in resistance or inertia or both along with Mathematical Modeling and Computing, vol. 23, pp. 101-117, 1996.
[12] R. Murray, D. Neumerkel and D. Sbarbaro, “Neural Networks for
changes in reference speed.
Modeling and Control of a Non-linear Dynamic System,” Proceedings
of the 1992 IEEE International Symposium on Intelligent Control, pp.
VI. CONCLUSION 404-409, 1992.
[13] Naga Sugatha. K, Vaisakh.K, Anand,G,”Artificial intelligence based
The tuning effort of an MRAC based system is much less speed control of brushless DC motor,’ 2010 IEEE Power and Energy
than that of a conventional PID system as it does not need the Society General Meeting, 25-29 July 2010 ,pp.1-6..
tuning of controller parameters for variations in plant [14] K. S. Narendra and K. Parthasarathy, “Identification and control of
dynamical systems using neural networks,” IEEE Trans. Neural
parameters. From the simulation results, it can be seen that Networks, vol. 1, pp. 1–27, Jan. 1990.
the MRAC based model is capable of speed tracking as well as [15] M. A. Hoque, M. R. Zaman, and M. A. Rahman, “Artificial neural
reduce the effect of parameter variations. This makes the network based permanent magnet dc motor drives,” in Proc. IEEE-IAS
Annu. Meeting, 1995, vol. 1, pp. 98–103.
motor suitable in applications such as Position Sensing and [16] M.A. Hoque, M.R. Zaman and M.A. Rahman, “Artificial neural network
Robotics. The proposed ANN-based control scheme is robust, based permanent magnet dc motor drives”, Conference Record,
efficient and easy to implement. The hardware implementation IEEE/IAS Annual Meeting, vol. l., 1995, pp. 99-103.
[17] M. Azizur Rahman, Fellow, IEEE, and M. Ashraful Hoque “On-Line
of the proposed system needs to be done to validate the Self-Tuning ANN-Based Speed Control of a PM DC Motor” in
capability of the MRAC controller to cope up with speed and IEEE/ASME Transactions on Mechatronics, vol. 2, no. 3, September
other parameter variations. 1997.
[18] N.Norgaard, O.Ravn, N.K.Poulsen and L.K.Hansen, “Neural networks
for modeling and control of dynamic systems”; 2nd edition, Springer—
Verlag London Ltd.
APPENDIX [19] S. Weerasooriya and M. El-Sharkawi, “Identification and control of a
dc motor using back-propagation neural networks,” IEEE Trans.
BLDC MOTOR SPECIFICATIONS Energy Conversion, vol. 6, pp. 663–669, Dec. 1991.
Rated speed 4000 rpm [20] Ting-Yu Chang; Ching-Tsai Pan; Fang, E.; “A Novel High Performance
Number of phases 3 Variable Speed PM BLDC Motor Drive System”; Power and Energy
Engineering Conference; pp. 1-6, 2010 Asia-Pacific.
Number of poles (P) 4 [21] Fredrick.M.Ham, Ivica Kostanic, “Principles of neurocomputing for
Rated current 5A Science and Engineering”, 1 st ed., Mc.Graw Hill,2001.
Rated voltage 36 V
Per phase Resistance 0.57 ohms
Per phase Inductance 1.5mH
Moment of inertia (J) 23e-6kg-m^2
Rated Torque (T) 0.42N.m
Torque constant 0.082N.m/A

ACKNOWLEDGMENT
We would like to thank the Management, Principal,
HOD(EEE), and other staff members of Sri Ramakrishna
Engineering College for providing the necessary facilities,
support and guidance throughout the duration of this work.

REFERENCES
[1] Astrom, K.J. and Wittenmark, B. (2000). Adaptive Control, 2nd ed.,
Englewood Cliff, NJ: Prentice Hall, pp.187-222.
[2] J.R.Hendershot and T.J.E.Miller, “Design of brushless Permanent –
magnet Motors”; Oxford,UK:Oxford Science, 1994.
[3] P.Pillay and R.Krishnan, “Modeling, Simulation and Analysis of
Permanent-magnet Motor Drives, part II: The brushless DC Motor
Drive,” IEEE Transactions on Industry applications,
vol.25,no.2,march/April 1989.
[4] A.K.Wallace and R.Spee, “The effects of motor parameters on the
performance of brushless DC drive,” IEEE Transactions on Power
Electronics, vol.5, no.1, pp.2-8, January 1990.
[5] Varatharaju,V.M., Mathur.B.L., Udhyakumar.K., “Speed control of
PMBLDC motor using MATLAB/Simulink and effects of load and
inertia changes,” 2010 2nd International Conference on Mechanical and
Electrical Technology(ICMET) 10-12 Sept.2010, pp.543-548.
[7] J.G.Ziegler and N.B.Nichols, “Optimum settings for automatic
controllers,” Transactions of American Society of mechanical
Engineers, Vol.64, 1942, pp.759- 768.
[8] A.O’Dwyer, Handbook of PI and PID Controller Tuning Rules, London,
U.K:Imperial College Press, 2003

2012 International Conference on Data Science & Engineering (ICDSE) 122