2021 18th International Multi-Conference on Systems, Signals & Devices (SSD'21)

A Comparative Study of BLDC Motor Speed

Control Using PI and ANN Regulator
Djamel Eddine Beladjine Djamel Boudana Abdelhafidh Moualdia
LREA Laboratory, Departement of LCP Laboratory, Departement of LREA Laboratory, Departement of
Electrical Engineering Electrical Engineering Electrical Engineering
University of Medea National Polytechnic School of Algiers University of Medea
Medea, Algeria Algiers, Algeria Medea, Algeria
[email protected] [email protected] [email protected]
2021 18th International Multi-Conference on Systems, Signals & Devices (SSD) | 978-1-6654-1493-7/21/$31.00 ©2021 IEEE | DOI: 10.1109/SSD52085.2021.9429474

Mohamed Hallouz Patrice Wira

LREA Laboratory, Departement of IRIMAS Laboratory, Departement of
Electrical Engineering Electrical Engineering
University of Medea University of Mulhouse
Medea, Algeria Mulhouse, France
[email protected] [email protected]

Abstract— This article discusses a comparative study known for its simplicity [7], according to the type of back-
between classical PI control and Artificial Neural Network EMF waveforms.
(ANN) based control of a brushless DC motor. BLDC is
First of all, we use a classic PI regulator due to its
recommended for drives that require a large load torque, these
simplicity of application and its reduced cost for speed
types of motors are non-linear systems which require robust
control, although the application of this type of regulator in
control. Due to the saturation characteristic presented by the
unstable conditions due the parametric variation of the
conventional PI regulator which causes the instability of the
machine causing instability of the system to regulate [8], to
system, we have used an artificial neuron-based controller to
avoid this issue, we opted for a more robust control strategy
overcome these aspects and to provide a better dynamic and
than the PI regulator and replaced it with a regulator based on
quick response, in order to control the speed of a BLDC motor
Artificial Neural Networks (ANN) and evaluated their
as well as the reduction in torque ripples. Simulation results
performance.
obtained using MATLAB / SIMULINK show that the artificial
neuron network controller performance evaluation are better This paper is divided into three sections, the modeling of
than those obtained by the standard PI controller. the BLDC motor gives in the first section, secondly, the
synthesis of the PI and ANN regulators has been developed,
Keywords—BLDC, Artificial Neural Network, Proportional - then, we discuss the results of the simulations, in the last
Integral, Speed Control. section.

I. INTRODUCTION II. BLDC MODELING

Until the 19th century the DC motor was the most common The overall structure of the BLDC motor controlled by a
used in industrial drive systems because of their simplicity of three-phase inverter with two voltage levels is shown in Fig.1.
control, however due to the major disadvantage of sparks
created by the brushes of a DC motor, so that it can not be used
in applications which require a long life service, to overcome
this aspect, the BLDC or permanent magnetic DC
synchronous motor, is become one of the most used machines
in the autonomous power systems due to its improved torque,
important speed, high precision control and easier wired
connections [1], [2]. As it can develope better dynamic
performance [3], also they don’t need regulary periodic
maintenance [4] which allows it to be qualified as one of the
best choices to make in electric vehicles, aerospace, robotics
and biomedical devices [5], [6].
Several types of research and studies have been used to
Fig. 1. Structure of a three-phase voltage inverter controling BLDC motor.
develop the various BLDC control techniques, the back
electromotive force (EMF) detection technique of which is

978-0-7381-4392-7/21/$31.00 ©2021 IEEE 1291

Authorized licensed use limited to: Motilal Nehru National Institute of Technology. Downloaded on June 13,2024 at 12:17:48 UTC from IEEE Xplore. Restrictions apply.
 A. Mathematical model  di a R 2 1
The BLDC motor has a balanced three-phase arrangement d t = L i a + 3L (V ab − Eab ) + 3L (V bc − Ebc )
of symmetrical, rear trapezoidal-EMF coil windings, and the  (5)
dib = − R i + 1 (V − E ) + 1 (V − E )
rotor of a BLDC motor is a permanent magnet. The rotor d t L
b
3L
ab ab
3L
bc bc

current is neglected due to its high resistance value of the

permanent magnet [9]. The interaction resulted between the Then the state space model is represented by
PM and the coils will train the BLDC motor [10], [11].
 R  2 1 
− 0 0   0
i 'a   L  ia  3L 3L  V − E 
ab ab
i '  = 0 - R 0  i  + − 1 1
0  Vbc − Ebc  (6)
 b   L   b   3L 3L
ω 'm    ωm    T −T 
0 0 β 0 1   em L 
-  0
 j   j 

i a   1 0 0  i a 
     
i b  =  0 1 0  i b  (7)
i c   -1 -1 0  i c 
     
Fig. 2. Equivalent diagram of a BLDC motor. ωm   0 0 1  ωm 

The winding stator voltage is giving by the following

The back-EMF phases is represented by
equations :


 dI A  dI B dI C  e = k ω F (θ )
V A = R .I A + L dt + m  dt + dt  + e A  A
     2π 
 dI B  dI dI  e B = k ω F  θ −  (8)
V B = R .I B + L + m  A + C  + eB (1)   3 
 dt  dt dt    2π 
 dI  dI dI  eC = k ω F  θ + 
V C = R .I C + L C + m  B + A  + eC   3 
 dt  dt dt 
Where ω is the angular speed, k represents the constant of
As we have a balanced system BLDC motor, and F(ϴ) is the function of variation of the
position.
I A + I B + IC = 0 (2)
The torque expression of BLDC motor is given as follow

And we obtain the following matrix [12] : e A I A + e B I B + eC I C

T = (9)
ω
VA R 0 0IA L−m 0 0 IA eA
V =0 R 0I + d  0 L−m 0 I +e  ( 3) Where eA, eB, and eC are respectivily the three phases
 B   B dt   B  B
VC 0 0 RIC  0 0 L−mIC eC back-EMFs.
With:

For reasons of cost minimization, industries ignore the

neutral wire, for this we can present the voltages between 
T =k i F (θ )
phases by the following equations  a t a
  2π 
 d T b =k t ib F  θ −  (10)
  3 
Vab =R(ia −ib ) +( L−m) d t ( ia −ib ) +Eab
 ( 4)   2π 
V =R(i −i ) +( L−m) d ( i −i ) +E T c =k t ic F  θ + 
  3 
 bc b c
dt
b c bc

T em =Ta +Tb +Tc (11)

We have a balanced system and by neglecting mutual
inductance
The mechanical equation giving as

dΩ
J = Tem −T L − F Ω (12)
dt

1292
Authorized licensed use limited to: Motilal Nehru National Institute of Technology. Downloaded on June 13,2024 at 12:17:48 UTC from IEEE Xplore. Restrictions apply.
  p and cause large settling time [13]. To overcome these
θe = 2 θ problems an ANN controller is used to provide a better
 (13) dynamic and faster response in order to control the speed of a
ω = d θ BLDC motor as well as the reduction of torque ripple [14].
 dt

with the parameters are respectively

V: Phase voltage.
i: Phase current.
R: Resistance of one phase.
L: Phase inductance.
M: Mutual inductance.
e: Back-EMF.
T: Phase torque.
Tem: Electromagnetic torque.
TL: Mechanical torque. Fig. 4. Proposed PI regulator for BLDC drive.
K: Back-EMF constant.
kt: Constant of torque. TABLE I. REGULATOR PARAMETERS
ω: Angular rotor speed.
Parameter Value
ϴ: Mechanical rotor angle.
ϴe: Electrical rotor angle.
Kp 0.013
III. PI REGULATOR CONFIGURATION
PI controllers are mostly used because of their simplicity Ki 16.61
of adjustment and regulation. The BLDC motor is a system
For the proposed synoptic diagram, we calculated the
that has several variables and therefore the possibility of
parameters of the PI regulator by exploiting the poles
saturation presence. In most cases, control techniques based
imposition method, and by using the electrical and mechanical
on PI regulators are used for current and speed control.
parameters of the BLDC motor.
The PI controller (Fig. 3) processes the difference between
The reference current is generated by our PI regulator and
the set point and the speed or current feedback signal, and the
by extracting the measured DC supply current we obtain the
controller output signal is obtained from overlapping
PWM signals through a logical comparator.
proportional and integral operations,
IV. ARTIFICIAL NEURAL NETWORK (ANN) CONTROLLER
The ANN is an interpretation of the human nervous
system, it is represented by a mathematical model of an
interconnected neuron established via thoughtful relationships
in order to accomplish a well defined task. The system consists
of three essential parts, the entry layer, the hidden layer, and
the exit layer.
Fig. 3. PI controller block diagram. The control is carried out by passing the external
information from the input layer to the output layer through
The PI controller equation is giving as : activations functions which allows the ANN to have nonlinear
1
behavior generally so that the system behavior is similar to the
Kp desired one [15].
r (t ) = K p e (t ) +
Ti  e(t ) dt
0
(14)
In this paper, the ANN is used to replace a PI regulator
Where r (t) is the function of the output signal and e (t) is
which aims to improve the dynamic performance of a BLDC
the function of the input signal to be regulated as illustrated in
motor [8], [16], we use an artificial neural network composed
Fig.4, and the adjustable parameters are: Kp witch is the
of 2 neurons and 2 layer which the input represent the error
proportional coefficient, Ti represents the integral time
between the reference and the response of mechanical speed
constant, and τ is the time constant, these parameters generally
and the output is the reference of the Idc current. The artifial
expressed by coefficients to obtain this transfer function.
network has been trained from the PI controller mentioned
Ki previously.
G (s ) = K p + (15)
s The following figure shows a global diagram of a neural
network.
If we do a large change on the setpoint we could have a
saturation of controller which leads to overshoot the system

1293
Authorized licensed use limited to: Motilal Nehru National Institute of Technology. Downloaded on June 13,2024 at 12:17:48 UTC from IEEE Xplore. Restrictions apply.
 V. SIMULATION RESULTS
The results presented in the following figures show a
comparative study of two different regulators used for the
control of the speed of the BLDC motor, the results are
obtained under environment MATLAB/SIMULINK.
The first one represents the controlled variable
Fig. 5. Proposed ANN perceptron structure.
(mechanical speed) at Fig.7, the reference speed was a
The functioning of an artificial neural network is rectangular signal to allow us to study and analyze the
characterized by the following equations : behavior of the BLDC motor well, and in order to observe the
evolution of the system in the acceleration and deceleration.
One can always affirm that the measured value follows
n
 p 
y = w k2 g  w kj1 ϕ j + b k1  + b 2 (16) perfectly its reference steps. either for the classical PI or ANN,
k =1  j =1  except that by using the regulator ANN one can distinguish a
decrease of the peak in transitional regime level and speed
With response than the classic one, and that the response time is
Φj : Network input. slightly improved compared to the PI regulator, for the PI
regulator we can obtain a response time equal to 0.0523s, and
wkj1 et bk1: The weights and the bias of the hidden layer with ANN we obtain a reduced time and equal to 0.0447s.
Sigmoïde function.

2
g ( x) = −1 (17)
1 + e −2 x

Wk2 et b2: The weights and the bias of the output layer.
The functional diagram for controlling BLDC by the ANN
method is illustrated in Fig.6.

Fig. 7. Shape of the speed curve.

Fig. 6. Proposed ANN regulator for BLDC drive.

In our offline supervised learning algorithm, we have used

multilayer neural networks (back propagation), which consists
of comparing the desired results with the one obtained in order
to change the biases and weights at each iteration, of such spell
gets a minimal error.
According to the simulation tests with the different
structures, we have chosen a number of hidden layers and the
number of their neurons, the one that gives the best results
where the optimal structure of network is according to the
error between the measurement of the speed and that of its
setpoint, the desired control output, and the control output
calculated by the neural regulator.
We have opted for a maximum number of iterations equal
Fig. 8. Back EMF trapezoidal forms.
to 1.21*10e-4 for a network in order to optain a better
performance.

1294
Authorized licensed use limited to: Motilal Nehru National Institute of Technology. Downloaded on June 13,2024 at 12:17:48 UTC from IEEE Xplore. Restrictions apply.
 that require precision and quick response, the reason why
BLDC is the most used in applications that require constant
torque.
The results obtained show an improvement in dynamic
and static performances in speed response as well as the torque
improvement compared to the conventional PI controller.
VII. REFERENCES

[1] J. C. Gamazo-Real, E. Vázquez-Sánchez, and J. Gómez-Gil, “Position

and speed control of brushless DC motors using sensorless techniques
and application trends,” sensors, vol. 10, no. 7, pp. 6901–6947, 2010.
[2] H. F. Prasetyo, A. S. Rohman, F. I. Hariadi, and H. Hindersah,
“Controls of BLDC motors in electric vehicle Testing Simulator,” in
2016 6th International Conference on System Engineering and
Technology (ICSET), 2016, pp. 173–178.
[3] V. K. S. Patel, “Modeling and simulation of brushless DC motor using
PWM control technique,” International Journal of Engineering
Research and Applications (IJERA), Vol. 3, Issue 3, pp.612-620, 2013.
Fig. 9. Stator current. (a) PI control approch. (b) ANN control approch.
[4] U. Neethu and V. R. Jisha, “Speed control of Brushless DC Motor: A
comparative study,” in 2012 IEEE International Conference on Power
The Fig.9 shows the stator current, we can notice the same Electronics, Drives and Energy Systems (PEDES), 2012, pp. 1–5.
thing about the transient regime, which by applying ANN [5] V. M. Hernández-Guzmán and J. Orrante-Sakanassi, “PID control of
controller, it allows a reduction of the peak almost up to 2 robot manipulators actuated by BLDC motors,” International Journal
of Control, pp. 1–10, 2019.
Amperes compared to the application of the traditional PI
[6] S. Uniyal and A. Sikander, “A Novel Design Technique for Brushless
regulator. DC Motor in Wireless Medical Applications,” Wireless Personal
Communications, vol. 102, no. 1, pp. 369–381, 2018.
[7] X. Zhou, X. Chen, C. Peng, and Y. Zhou, “High performance
nonsalient sensorless BLDC motor control strategy from standstill to
high speed,” IEEE Transactions on Industrial Informatics, vol. 14, no.
10, pp. 4365–4375, 2018.
[8] M. Hallouz, S. Moulahoum, N. Kabache, and S. Kouadria, “Neural
Network Based Field Oriented Control for Doubly-Fed Induction
Generator,” International Journal of Smart Grid-ijSmartGrid, vol. 2,
no. 3, pp. 183–187, 2018.
[9] A. Ouarda, B. El Badsi, and A. Masmoudi, “DTC of B4 inverter fed
two-phase IM drives,” in 2016 IEEE Vehicle Power and Propulsion
Conference (VPPC), 2016, pp. 1–6.
[10] A. L. Saleh and A. A. Obed, “Speed control of brushless DC motor
based on fractional order PID controller,” International Journal of
Computer Applications, vol. 95, no. 4, 2014.
[11] Chang-Liang Xia and Wei Chen, “Sensorless control of brushless DC
motors at low speed using neural networks,” in 2005 International
Conference on Machine Learning and Cybernetics, Aug. 2005, vol. 2,
pp. 1099-1103 Vol. 2, doi: 10.1109/ICMLC.2005.1527107.
[12] D. Kumar, R. A. Gupta, and N. Gupta, “Minimization of current ripple
and overshoot in four switch three-phase inverter fed BLDC motor
using tracking anti-windup PI controller,” in 2017 IEEE International
Fig. 10. Mecanical torrque waveforme Conference on Signal Processing, Informatics, Communication and
Energy Systems (SPICES), 2017, pp. 1–6.
The motor that was used in this study included the [13] F. Z. Belaimeche, A. Bentaallah, S. Massoum, and P. Wira, “A
comparative study between a simplified fuzzy PI and classic PI input-
following parameters: output linearizing controller for the wind-turbine doubly fed induction
generator,” Elektrotehniski Vestnik, vol. 85, no. 4, pp. 142–148, 2018.
Rs = 2.875 Ω, Ls= 8.5 mH, J= 0.8 g.m2, F=0.001 N.m.
[14] S. Kaliappan, B. Karunamoorthy, and J. Ramprabu, “Performance
analysis of sensorless BLDC motor using PI and anfis controller,”
VI. CONCLUSION International Journal of Pure and Applied Mathematics, vol. 116, no.
For this article, the principal aim is to illustrate a 11, pp. 201–209, 2017.
comparative control study of the mechanical speed of the [15] D. O. Abdeslam, P. Wira, J. Merckle, and Y.-A. Chapuis, “A neural
approach for the control of an active power filter,” in 5th International
BLDC motor by using two different types of regulators. First, Power Electronics Conference (IPEC’2005), 2005, pp. 923–929.
we use the classical PI controller. we replace the PI controller [16] A. Moualdia, D. Boudana, O. Bouchhida, and P. Wira, “Direct torque
with the ANN regulator, according to the study of the control based multi-level inverter and artificial neural networks of wind
energy conversion system,” in 2016 8th International Conference on
fundamental concepts of neural network, the supervised Modelling, Identification and Control (ICMIC), 2016, pp. 49–54.
learning strategy has been proposed to improve the response
time from 0.0523 seconds in proportional integrator controller
down to 0.0447 seconds in ANN controller, it might be a little
difference but it's needed especially in sensitive applications

1295
Authorized licensed use limited to: Motilal Nehru National Institute of Technology. Downloaded on June 13,2024 at 12:17:48 UTC from IEEE Xplore. Restrictions apply.