See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/324759114

Multi-Resolution Wavelet PID Speed and Current Controllers of BLDC Motor

Based on Invasive Weed Optimization Technique

Article  in  International Journal of Applied Engineering Research · April 2018

CITATION READS

1 341

2 authors:

Adel Obed Abbas K. Kadhim

Middle Technical University Middle Technical University
25 PUBLICATIONS   39 CITATIONS    5 PUBLICATIONS   5 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Solar tracking system View project

Speed Control of Brushless DC Motor Based on Wavelet Neural Network (Wavenet) View project

All content following this page was uploaded by Abbas K. Kadhim on 25 April 2018.

The user has requested enhancement of the downloaded file.

 International Journal of Applied Engineering Research ISSN 0973-4562 Volume 13, Number 8 (2018) pp. 6234-6243
© Research India Publications. http://www.ripublication.com

Multi-Resolution Wavelet PID Speed and Current Controllers of BLDC

Motor Based on Invasive Weed Optimization Technique

1
Adel A. Obed, 2Abbas K. Kadhim
1
Assistant Professor Dr., Department of Electrical Power Engineering Techniques,
Electrical Engineering Technical College, Middle Technical University, Baghdad, Iraq.
Email: adelrazin@gmail.com.
2
M.Sc. Researcher, Department of Electrical Power Engineering Techniques,
Electrical Engineering Technical College, Middle Technical University, Baghdad, Iraq.
Email: hmosa703@gmail.com.

Abstract both stainless steel and magnet. No damper windings are

The brushless DC (BLDC) motor is a non-linear and multi-
variable system so that the classical linear PID controlling
methods have difficult to obtain a favorable result for BLDC
motor. This paper proposes a multi-resolution PID (MRPID)
controller wavelet approach for BLDC motor speed and
current control. The discrete wavelet transform (DWT) used
in this controller, has to decompose the error signal into
different high, medium and low-frequency ingredients. The
underlying phenomena represented by DWT different scales
coefficients such as the noise of measurement, external
troubles effects, process dynamics are weighted by their own
gains then the control signal is generated from added them
together. This paper presents also the wavelet neural network
(wavenet) with infinite impulse response (IIR) filter as the
reference model to auto-tuning the MRPID controllers. The
controller gains are optimized using invasive weed
modeled because they will reduce the magnetic force [3]. The
stator current must be semi-square in order to synchronize
with b-emfs to make stable and higher torque at steady speed.
The operation of this motor is done when two phases are ON
state and the third is floating for every 60 electrical degrees
[4, 5].
BLDC motor modeling can appear in the parallel attitude as a
synchronous machine with three-phase windings. As any
model multi-phase motors, one style of them is fed by three-
phase voltage source as a BLDC motor is as shown in Figure
1. In order to operate this motor, the peak voltage must not to
be overtaken the ultimate voltage frontier of the motor and not
to be lower than the back-emf induced voltage [6, 7].
+ Inverter

optimization (IWO) algorithm. This drive system is D1 D3 D5 Stator

implemented by Matlab/Simulink program and the tested Ra ea
T1 T3 T5 La
system gives strong robustness, high control accuracy, and ia
Q1 Q3 Q5 Rotor
good dynamic characteristics. vas Rb ib Lb eb
vbs
Keywords: Brushless DC Motor, Infinite Impulse Response D2 D4 vcs D
6 Rc ic Lc ec
Filter, Invasive Weed Optimization Algorithm, Wavelet
T2 T4 T6 sensor
Decomposition, Wavelet Neural Network. Q2 Q4 Q6

INTRODUCTION
The BLDC motor applications turn into most requests and the
rapid rising such as aerospace, automotive, office automation,
household appliances and different industries. It has important
advantages like long operation life, noiseless operation, high
efficiency, high dynamic response, wide speed range, low
temperature and can withstand vibrations and shock; these
will get better stability of the drive system [1, 2]. BLDC
motors are really a type of permanent magnet synchronous
motors in spite of its name. The commutation of the currents
is done by electronically switches inverter which is driven by
a DC power supply. The commutations are resolved by rotor
position; this is detected either by sensorless mechanisms or
position sensor. This motor type is consisting of permanent
magnets in the rotor and three phase publicized windings in
the stator that are wound as trapezoidal b-emfs. The induced
currents in the rotor can be lapsed due to the high resistivity of
-
T1 ---- T6
Trigger generator and
controller
Figure 1: Three-phase BLDC motor equivalent circuit
Different simulation models are done by many searchers to
anatomize the behavior and operation of the arrangement to
give the accurate torque rate that is identifying with current
and b-emfs corresponding to suitable set point speed, they
work on tuning PID parameters as based on Ziegler-Nicholas,
genetic algorithm tuning methods and other work on adjusting
PID parameters by fuzzy optimized algorithm or using
fractional order PID controller and PID controller with two
degrees of freedom [8-12].
Speed control of BLDC motor takes a necessary role in the
modern control. Some researchers have expanded the

6234
 International Journal of Applied Engineering Research ISSN 0973-4562 Volume 13, Number 8 (2018) pp. 6234-6243
© Research India Publications. http://www.ripublication.com

structure of Artificial intelligence to control the nonlinear where ̅̅̅̅̅̅ ̅̅̅̅̅̅ is

system such as wavelet and wavelet neural network (wavenet)
approaches [13, 14]. In this paper, the decomposition
approach of the wavelet is used as MRPID controller to
control the current and speed of the BLDC motor. This
controller has the capability to supply smooth control signal
and best rejection to the disturbances. Later, the wavenet with
IIR filter is proposed to identify the BLDC motor and to on-
line auto-tune the MRPID controller gains where this method
gives more stable control system. The gains of these
controllers will optimize by the IWO algorithm, Because of its
characteristics as locative dispersal, competitive exclusion and
the way of reproduction.
WAVELET DECOMPOSITION ANALYSIS
A wavelet is a small wave which grows and decays in a
limited period. It is known as a very short period oscillatory
wave and fulfills the provision of admissibility [15], given by
∞ conjugate filters corresponding to scaling and wavelet filters
Ψ(0) = ∫−∞ ψ(t)dt = 0 (1)
where ψ is the wavelet function for the Fourier transform Ψ.
Wavelet function is called the "mother wavelet", different
"daughter wavelets" are generated from the contraction or
expansion, and translation, that has the mathematical
personification given by
φ(t) is conjugate of scaling function φ(t) and ψ(t)
conjugate of wavelet function ψ(t). The (DWT) is the time
scale exemplification of a digital signal using techniques of
digital filtering.
PROPOSED MRPID FOR BLDC MOTOR SPEED
CONTROL
The classic PID control law is a linear error signal
decomposition which is fixed with always only has three
terms (Proportional, Integral, and Derivative) [8-12]. The
main variance between the conventional PID and the MRPID
controllers is the later decomposes the error signal to low,
intermediate and high frequencies. The decompositions
number can be more than three (depending on the level of
decomposition) with different scales of time-frequency. The
error signal components are calculated using equation (3)
through a sketch of sub-band coding for three levels, as shown
in Figure 2. The decomposition filters ̅̅̅̅̅̅
h[k] and ̅̅̅̅̅
g[k] are the
respectively. By using inverse discrete wavelet transform
(IDWT) algorithm reconstruct the high, medium and low-
level error signals as shown in the decomposition synthesis
Figure 3.
a1 a2
↓2 ↓2 ↓2 a3

1 t−b
ψa,b (t) = ψ( ) , a ≠ 0, a, b ∈ Ʀ (2) ↓2 d3
√a a
e[k]
where Ʀ is vector space of the real numbers, a is the dilation
↓2 d2
parameter that allows for the contractions and expansions of
the mother wavelet and the translation parameter b allows
↓2 d1
translate in time.
Multi-resolution analysis (MRA) by using wavelet transforms
technique where different resolution analyzed by different Figure 2: Error signal decomposition
frequencies. The signal is decomposed into groups of
frequency or scales bands. The multi-resolution a2 a1
decomposition is the basic function in the wavelet transforms; a3 ↑2 h[k] ↑2 h[k] ↑2 h[k] eh
that decomposing the signal into two parts are high and low- a2 a1
frequency components. The detail information represents the d3 ↑2 g[k] ↑2 h[k] ↑2 h[k] em2
high-frequency components like sharp edges while the a1
identity of the signal represents the low-frequency component d2 ↑2 g[k] ↑2 h[k] em1
which is the original signal smoothed version. This
decomposition operation can be repeated, in each time the d2 ↑2 g[k] el
low-frequency components are decomposed in turn to obtain
low- and high-frequency components. The detail information Figure 3: Error signals decomposition synthesis
is generated by a high-pass filter, while the low-pass filter will
generate the information of approximation. The coefficients of
low- and high-pass filters discrete wavelet transform (DWT) The selected of a mother wavelet which has better fits the
are specified by the function of mother wavelet [16]. The signal counts on the signal itself and the application. In the
original signal can be reconstructed in the same way using BLDC motor drive, the goal is to perform the error signal
inverse (DWT). The N-level of DWT series exemplification wavelet analysis. In order to determine the perfect mother
of a signal f(t) can be known as [16]: wavelet, the criterion of Mean Square Error (MSE) is used.
The determined (MSE) index for the original and the
f(t) = ∑k cN,k φN,k (t) + ∑N
m=1 ∑k dm,k ψm,k (t) (3) reconstructed signal as follows [17]:
̅̅̅̅̅̅̅̅̅
cm,k = ∑k f(t)φ N,k (t) (4)
MSE = ||e − ȇ|| = ∑Ł−1
n=0 √|e − ȇ|
2 (6)
dm,k ̅̅̅̅̅̅̅̅̅̅
= ∑k f(t)ψ m,k (t) (5) where e is the original error signal, ȇ is the reconstructed error
signal and Ł is the signal frame length 64 sample. As a result

6235
 International Journal of Applied Engineering Research ISSN 0973-4562 Volume 13, Number 8 (2018) pp. 6234-6243
© Research India Publications. http://www.ripublication.com
of the test, Symlet with order 5 ʻsym5’ is selected as a mother
wavelet because of smallest sum of MSE indices.
The number of gains of the BLDC motor drive controller
depends on the decomposition level number. In this work, the
entropy criterion as a base is used to find the optimal level of
the signal of error decomposition. The signal e[n] entropy
H(e) by length L is known as [18]:
H(e) = − ∑L−1 2
n=0 |e(n)| log |e(n)|
2
the entropy is estimated through each level. If there is a next
level j such as
H(e)j ≥ H(e)j−1
then remove the DWT tree level j. By using Matlab wavelet
toolbox with the selected above mother wavelet (sym5), it was
found that the 2nd level of decomposition is suitable for the
efficient resolution of the speed and current error signal in the
proposed MRPID controller for the BLDC motor speed
control.
The coefficients of decomposition represent many implicit
phenomena as measurement noise, external disturbances
Wavelet Decomposition
Analysis and Synthesis
eh
kh
es
Reference
em
+ km
speed - Inverter
el
kl
MRPID Speed controller
Rotor
speed
Figure 4: The schematic diagram of a BLDC motor drive based on MRPID speed controller
Generally in practice, trouble and commands are signals of a
low-frequency, and noises of a sensor are signals of high-
frequency. So, the low-frequency gain (kl) of the error signal
is increased to improve the disturbance rejection of the
system. The medium frequency gain (km) of the error signal
adds damping to the system and is increased to improve the
transient responses. However, the gain (kh) for high-frequency
components of the error signal is set to zero in order to reduce
the effect of noise on the system [15].
effects, and process dynamics. These ingredients are scaled by
their particular gains by using one of the methods of
optimization algorithms, and there are added jointly to
produce the control signal:
u = k l el + k m1 em1 + ⋯ + k m(N−1) em(N−1) + k h eh (9)
where gains kh, km1, . . . , km(N-1) are used to tune the error
signal (eh , em1, . . . , em(N-1)) high and medium frequency
(7) components. Gain kl is used for the error signal (el) low-
frequency components tuning, and N is the decomposition
levels number. The MRPID controller will have three gain
parameters. In terms of frequency information, the error signal
(8) low-frequency information captured by the proportional and
integral terms, and the high-frequency information captured
by derivative term [16], a conventional PID control has three
gains to be tuned kP, kI and kD, while the MRPID control has
two or more gains and the number of gains depends on the
decomposition level number is applied to the error signal. The
schematic diagram of a BLDC motor drive using a MRPID
control is shown in Figure 4. The inner loop is to control the
current and the outer loop is to control the speed of the BLDC
motor to the desired value.
Wavelet Decomposition
Analysis and Synthesis
eh
kh
u(t)
ei 3-Φ BLDC
∑ ∑
em
+- km PWM
motor
ϴ
el
kl
Rotor
MRPID Current controller
position
Idc DC-link
ia,b,c
current
RECURRENT WAVELET NEURAL NETWORK
BASED IIR FILTER
An important privacy of neural networks (NN) is their
capability to generalize and learn. A wavelet neural network
mostly consists of a feed-forward and may feed-backward
neural network, with one hidden layer, consists of neurons,
that their activation functions are drawn from wavelet family.
The output neuron amplitude is controlled by an activation
function. These wavelet neurons are usually referred to as
wavelons [19]. The configurations of recurrent wavelet neural
network (RWNN) are the input of WNN consists of delayed
samples of the system input x and the system output y. The
activation function of the hidden layer is the wavelet basis as
in equation (2). The input of this layer is denoted as [20]:
6236
 International Journal of Applied Engineering Research ISSN 0973-4562 Volume 13, Number 8 (2018) pp. 6234-6243
© Research India Publications. http://www.ripublication.com

uN (n) = xN (n − i) + ψN (n − 1) ∗ rsN + y(n − 1) ∗ rN (10) h(z) =

d0 +d1 z−1 +⋯+dN z−N
(12)
1+c1 z−1 +⋯+cM z−M
where rsN is the self-feedback gain, rN is the output feedback
∑N
k=0 dk z
−k
gain, N is the number of wavelon and i is the number of input = (13)
1+∑M
k=0 ck z
−k
delay. Now the output of the network is:
y = ∑N where ck is the feed-forward filter coefficient, d k is the feed
N=1 wN ∗ ψN (11)
backward filter coefficient, M and N are the maximum filter
where 𝑤𝑁 is the output weight of the wavelon. The RWNN order. In order to perform this filter the following equation of
has been blended with IIR filter. the output approximation ̂(t)
y is required:
A wavenet is a topical network, where the output function is ŷ(t) = ∑M N
i=0 ci y(t − i) + ∑j=1 dj y
̂ (t − j) v(t) (14)
well localized in both frequency and time domains. In
addition, another topical network can be accomplished to this Figure 5 shows the wavenet in cascading with IIR filter,
wavenet architecture, as cascaded with a local IIR. The IIR where y(t) is the output of the wavenet and v(t) is the co-input
filter function is to filter the wavelons that have “little” is kept small for the purposes of stability feedback.
contribution to the identification process [21]. IIR filters are
helpful for a high-speed design because they typically need a
less number of multiplies compared to other filters. The
general equation of an IIR filter can be written as [22]:
Input layer Hidden layer Output layer

rs1
r1
Ψ1

w1 c0
rs2 c1
cM-1
w2 -1 -1 cM
x(t) Ψ2 ∑
y(t) z z z-1 ∑ (t)
d1
dN-1 X v(t)
r2 dN
rsN
wN
z-1 z-1 z-1
rN
ΨN

Figure 5: Wavenet based IIR filter configuration

PROPOSED BLDC MOTOR SPEED CONTROL BASED

ON MRPID WITH WAVENET AND IIR FILTER
The BLDC motor is a non-linear system and multi-variable,
nref(t) - e(t) MRPID u(t) BLDC motor n(t)
so it is troublesome to get a favorable result using MRPID +
controller system
controller alone. Here will present the combination between
ê(t) +
that’s controller and the wavenet. The wavenet is built as the K E
-
on-line reference model of the BLDC motor and the MRPID
controller get on-line tuning of its parameters by an auto- Auto-tuning Wavenet with n(t)
algorithm Γ(t) IIR filter
tuning algorithm.
v(t)
Figure 6 shows the schematic diagram of the BLDC motor
speed control based on MRPID with wavenet identification. Figure 6: Schematic diagram for BLDC motor speed control
The overall system has three stages to manipulate the speed of based on MRPID controller with wavenet
the BLDC motor in the desired way. The first stage is the
completed wavenet with IIR filter, this stage required the
input u(t) as the controlling signal to the BLDC motor and the The parameters of the wavenet are minimized by a cost
wavenet output is the estimation reference speed n̂(t); the function using invasive weed optimization based on the least
estimation error ê(t) is defined as: mean squares (LMS) as:
1
ê(t) = n(t) − n̂(t) (16) Cost function = ê(t)2 (17)
2
where n(t) is the output speed of the BLDC motor. The second In order to tune the PID parameters use the following
stage is the MRPID controller which has the speed error equations [21, 23]:
signal e(t) as the input and the control signal u(t) as the output
to manipulate the BLDC motor speed. The third stage is the k p (t) = k p (t − 1) + μp ∗ ê(t) ∗ Γ(t) ∗ (e(t) − e(t − 1)) (18)
auto-tuning of the MRPID controller gains.

6237
 International Journal of Applied Engineering Research ISSN 0973-4562 Volume 13, Number 8 (2018) pp. 6234-6243
© Research India Publications. http://www.ripublication.com

k i (t) = k i (t − 1) + μi ∗ ê(t) ∗ Γ(t) ∗ e(t) (19)
k d (t) = k d (t − 1) + μd ∗ ê(t) ∗ Γ(t) ∗ (e(t) −
2e(t − 1) + e(t − 2)) (20)
where μp,i,d are the learning rate of that gains which will be
optimized using IWO algorithm, e(t) is the tracking speed
error which is the difference between the reference speed
nref(t) and the actual speed n(t), and Γ(t) is parameter
estimated by the wavenet as[21]:
Γ(t) = ∑M
i=0 ci ∗ y(t − i) (21)
where M is the order number of the IIR filter, c is the feed-
forward coefficient of the IIR filter and y(t) is the output of
the wavenet as described previously.
INVASIVE WEED OPTIMIZATION TECHNIQUE
The algorithm of Invasive Weed Optimization (IWO) was
presented firstly 2006 by Mehrabian and Lucas [24]. It is easy
and simple to program and understands and also has
robustness and rapid global searching ability. So it is a
convenient emulator for other evolutionary algorithms to be
applied in many engineering fields [25, 26]. Over the
specified area a finite number of seeds are being spread, any
seed produce a new weed and then the weed will produce
other weeds from its seeds depending on its fitness. The
computational experiments show that the IWO is a highly
competitive property where it is able to find the best and
optimal solutions for the instances studied [27].
The Algorithm for Invasive Weed Optimization. The
unwanted growing plants are called weeds, which have
adaptive and robust nature. That will turn them into
unfavorable plants in agriculture. A common agronomy belief
is that “The Weeds Always Win”. The motivated new
algorithm is colonization of invasive weeds by a common
agriculture phenomenon. Figure 7 shows the algorithm flow
chart. The specifics of the IWO are classified as follows:
the produced seeds are randomly spread out around the parent
weed to generate new controller gains. However, standard
deviation σ of the random function will be reduced from a
previously defined initial value initial to a final value final in
every generation As a function of iteration (iter.) as:
(itermax −iter)n
cur = final + (initial − final ) (23)
(itermax )n
where cur is the current standard deviation, itermax is the
maximum number of iteration and n is the nonlinear
modulation index. So that,  determine weeds exploration
ability. Therefore, exploration ability of all weeds is gradually
reduced as iteration increases. At the end of the optimization
process, the exploration ability has diminished so much that
every weed can only find its position with respect to the
smallest BLDC motor current and speed error.
E. Competitive Alienation. Through increasing of the
iterations, the number of weeds in the colony will reach its
maximum, and adopted the removal mechanism. The seeds
and their parents are assorted together and only those with the
best fitness of the BLDC motor output can stay live and
become genital. Others are being removed.
Start
Commence
Estimation
Propagation
Spatial
dispersion

A. Commence. Weeds are generated randomly over the

search space as the inhabitance of initial solution. Containment of No Is weeds > Yes Competitive
B. Estimation. Calculate the fitness of each weed in the all seeds size max? alienation
inhabitance of the controller parameters.
C. Propagation. The best weed’s fitness in the
population will produce more seeds while the worst weed’s
fitness will produce fewer seeds depending on the response of
the BLDC motor variables. The increase of seeds number No Is iter. >
produced is linear based on weeds’ fitness depending on the iter. max
following formula:
Yes
f−fworst
weednum = Smin + (Smax − Smin ) (22)
fbest −fworst End
where weednum is a number of weeds, Smax , Smin are
maximum and minimum seeds respectively, f is the current Figure 7: IWO flow chart
fitness of the weed and fbest , fworst are the best and the worst
fitness of the current inhabitance. SIMULINK APPLICATION AND RESULTS
The plenary BLDC motor drive Simulink model has been
D. Spatial Dispersal. Here weed will generate seeds performed using MATLAB/SIMULINK software. Each part
around it leading to a local search for each plant. According to of the BLDC motor drive model is finalized by a set of
a normal distribution with mean equal to zero and variance 2,

6238
 International Journal of Applied Engineering Research ISSN 0973-4562 Volume 13, Number 8 (2018) pp. 6234-6243
© Research India Publications. http://www.ripublication.com

mathematical model. The motor parameters used here are advanced control strategy is PWM control is selected. The
shown in Table 1. IWO parameters are given in Table 2. These parameters are
selected to obtain the optimal MRPID gains are shown in
Table 1: Data of BLDC motor [28] Table 3 to control the speed of the BLDC motor.
Parameter Table 2: IWO parameters
Parameter type Unit
value
Parameter type value
Rated power 1200 W
Number of initial plants 4
Rated voltage 76 V
Maximum number of 100
Rated current 16 A plants
Rated torque 2.9 N.m Maximum iteration number 50
No load current 0.66 A Maximum seeds number 5
No load speed 3500 rpm Minimum seeds number 1
Torque constant 0.207 Nm/A Initial standard deviation 10
Voltage constant 0.207 V.s/rad Final standard deviation 0.0000001
Terminal resistance 0.110 Ω Modulation index 3
Inductance 0.6 mH Dimension 4
−4
Viscous damping factor 1.3x10 N.m s/rad
Rotor inertia 1.7x110−3 kg.𝑚2 Table 3: MRPID optimized parameters
Number of poles 8 - Parameters Values
Speed controller
A. MRPID Controller High level (kh) 7.28
The multi-resolution analysis methods are used here to control
Medium level (km) 0.4786
speed and stator currents of the BLDC motor. From the
schematic diagram in Figure 4, the Simulink model of the Low level (kl) 0
proposed MRPID speed controller is shown in Figure 8,
where the speed error signal is the input to the S-Function Current controller
block which has the MRA program. The decomposition High level (kh) 15.48
program use Symlet with order 5 as the mother wavelet
(sym5) as selected in the previous section and the Medium level (km) 84.71
decomposition level is second as it chose. The outputs of the Low level (kl) 0
S-Function block are the decomposition synthesis signals of
the speed error signal which are multiplied by their
corresponding gain before added them to generate the control Figure 9 shows the speed response at the 2000 rpm reference
signal. The values of the gains are optimized on-line using speed starting with no load and the full load 2.9N.m is applied
IWO algorithm. The MRPID current controller is building at 0.2 second. The electromagnetic torque, current phase ia
same as the speed controller as explained above. and line voltage vab at no load and full load are shown in
Low level error Figures10 - 12 respectively.
Kl
+
Medium level error
1 MRA Km + 1
Error Control signal
S-Function + Limiter
High level error
Kh Sum

Figure 8: Simulink model of the proposed MRPID

speed/current controller
The cost function used by IWO to optimize the parameters of
the MRPID controllers as:
F = min(∫ e2s dt + ∫ e2c dt + Mp ) (24)
where F is the fitness function, es is the speed error, ec is the
current error and Mp is the maximum overshoot. The most Figure 9: Speed response based on MRPID controllers

6239
 International Journal of Applied Engineering Research ISSN 0973-4562 Volume 13, Number 8 (2018) pp. 6234-6243
© Research India Publications. http://www.ripublication.com
Figure 10: Electromagnetic torque based on MRPID
controllers
Figure 11: Current phase ia based on MRPID controllers
Figure 12: Line voltage vab based on MRPID controllers
Figure 13: Simulink model of the proposed BLDC motor speed control based on MRPID with Wavenet and IIR Filter
6240
The graphs obtained are given in Figures 9-12, shows the
specific features of the BLDC motor by MRPID controller,
i.e. smooth controlling that restraining harass and stress the
engine. Also, the plant has better performance with low
operating frequency. As the classical PID generates high-
frequency may lead to rise the temperature and damage the
actuator. The motor speed has flowed the reference speed with
delay time about 0.06 second and the speed is reduced to a
small amount through applied a full load, this will improve
later.
B. MRPID Controller with Wavenet and IIR Filter
To overcome the decrease in speed in the previous section,
Figure 13 below shows the Simulink model of the proposed
BLDC motor speed control based on MRPID with Wavenet
and IIR Filter. The reference model of the BLDC motor is
built by the wavenet with IIR filter. The wavenet Simulink
model is constructed from equations (2, 10 and 11). The
neurons are built from the daughter wavelet that is generated
from dilation a'th and translation d'th with the mother wavelet
and self-feedback r'th and output feedback rr, each neuron is
weighted by w'th gain. The wavelet mother function used here
as the Mexican hat function. After trying and error the
Mexican hat mother wavelet is more suitable to represent this
model.
 International Journal of Applied Engineering Research ISSN 0973-4562 Volume 13, Number 8 (2018) pp. 6234-6243
© Research India Publications. http://www.ripublication.com

The IIR filter Simulink model is constructed from equation electromagnetic torque, current phase ia and line voltage vab at
(14) where the filter order M and N are kept to be 2 for both. no load and full load are shown in Figures 15-17 respectively.
The input to this model is the wavenet output and the
estimation speed and Γ(t)(as equation (21)) are the output of
that model. The auto-tuning Simulink model is constructed by
using equations (18 - 20).
Firstly, the operation is done with MRPID controller without
the auto-tuning stage to learn the reference BLDC motor. The
parameters of the reference BLDC motor are optimized by
using IWO algorithm based on the LMS as the cost function
equation (17). The IWO parameters are same as shown in
Table 2 but here with 23 dimensions. The wavenet optimized
parameters are shown in Table 4 and the IIR filter optimized
parameters are shown in Table 5. Secondly, the auto-tuning
stage is performed and the learning rate parameters are Figure 14: The speed response by on-line tuning controller
optimized by the IWO algorithm as shown in Table 6. based on wavenet with IIR filter
Table 4: The wavenet optimized parameters
Wavenet Dilation Parameters
a1 a2 a3 a4
0.9684 1.654 6.377 0.05941
Wavenet Translation Parameters
b1 b2 b3 b4
5.317 -1.429 0.5897 3.549
Wavenet Weight Parameters
Figure 15: The electromagnetic torque by on-line tuning
w1 w2 w3 w4 controller based on wavenet with IIR filter
0.5566 -3.655 3.678 2.146
Wavenet Feedback Parameters
r1 r2 r3 r4 rr
1.244 0.8909 -1.102 0.4564 -0.04522

Table 5: The IIR filter optimized parameters

Feedback
Feedforward Parameters Co-input
Parameters
c0 c1 c2 d1 d2 v Figure 16: The current phase ia by on-line tuning controller
2.724 2.831 2.724 0.1618 2.525 -0.3642 based on wavenet with IIR filter

Table 6: The learning rate optimized parameters

Speed Controller Current Controller
μp μi μd μp μi
24.35 43.03 -12.71 -7.101 20.88

The simulation results are performed by applying all the

above parameters to the BLDC motor drive model with the
0.4 second simulation time. Figure 14 shows the speed
response at the 2000 rpm reference speed starting with no load Figure 17: The line voltage vab by on-line tuning controller
and the full load 2.9N.m is applied at 0.2 second. The based on wavenet with IIR filter

6241
 International Journal of Applied Engineering Research ISSN 0973-4562 Volume 13, Number 8 (2018) pp. 6234-6243
© Research India Publications. http://www.ripublication.com

From results above can be seen that the speed has good The results of different BLDC motor variables are plotted and
response with removing the decrease after loading and the discussed.
other machine variables are stable with that proposed
controller. Another merit is that the control signal
characteristics generate a soft control even in the presence of REFERENCES
noise.
[1] Chang-Liang Xia, "Permanent magnet brushless DC
motor drives and controls" John Wiley & Sons,
ISBN 978-1-118-18833-0 2012.
COMPARISON BETWEEN CONTROLLING
METHODS [2] Stefan B., "BLDC motor modeling and control – a
Matlab/Simulink implementation," Master Thesis in
The comparison is done with respect to the operation and
Electrical Power Engineering, Chalmers University
results from the output waveforms of the performing models
of Technology, May 2005, Gothenburg, Sweden.
as shown in Table 7 below.
[3] Shivraj S. Dudhe and Archana G. Thosar,
Table 7: Comparison between controlling methods
"Mathematical modeling and simulation of three
MRPID Wavenet with phase bldc motor using Matlab/
Feature
controller PID controller Simulink," International Journal of Advances in
Engineering & Technology, vol.7, no.5, pp. 1426-
Rise time 0.06 0.07 1433, 2014
Steady-state [4] He Hao, Jianxin Shen, Cheng Yuan and Qiang Qu
0.15% 0%
error at no load "Influences of machine structure on high speed PM
Maximum BLDC motor," Electrical Machines and Systems
0% 0.35% (ICEMS), 17th International Conference on IEEE,
Overshoot
2014.
Overcome
poor Very good [5] Alexander G. Mikerov, "Brushless DC torque motors
Load
quality level indexes for servo drive applications,"
ripple Smooth Jagged EUROCON'09, IEEE, 2009.
Speed decrease [6] S. Rambabu, "Modeling and control of a brushless
Yes No
with full load DC motor," Master Thesis in Power Control and
Drives Technology, National Institute of Technology
Rourkela, 2007.
While the classical PID has simple hardware structure as
[7] P. Pillay and R. Krishnan, "Modeling, simulation,
compared with proposed MRPID, it does not have enough
flexibility to control the nonlinear system and generate high and analysis of permanent-magnet motor drivesII,"
frequency may lead to rise the temperature and damage the IEEE Transactions on Industry Applications, vol.2,
pp.265-273, Apr. 1989.
actuator.
[8] Md Mustafa Kamal, Lini Mathew and S. Chatterji,
Actually, the proposed wavenet method has complex
"Speed control of brushless DC motor using
hardware structure but it gives a better load rejection.
intelligent controllers," Engineering and Systems
(SCES), Students Conference on IEEE, 2014.
CONCLUSION [9] Songmao Zhang, and Yunliang Wang, "The
simulation of BLDC motor speed control based-
In this paper, the wavelet decomposition approach is used to
optimized fuzzy PID algorithm," Mechatronics and
produce modified advanced controller called MRPID
Automation (ICMA), International Conference on
controller. This is used to control the BLDC motor speed and
IEEE, 2016.
current and can provide better performance of the motor
variables than the classical controller. The current controller [10] Mohammed Abdelbar Shamseldin and Adel A. EL-
was setting the current and overcome the high starting current. Samahy, "Speed control of BLDC motor by using
Also, this paper achieved on-line control and on-line PID control and self-tuning fuzzy PID
identification by using wavenet with IIR filter approach. The controller," Research and Education in Mechatronics
benefit of the proposed methods is to reach a better response (REM), 15th International Workshop on IEEE,
of the speed. A proposed wavenet method is a modified 2014.
method of a MRPID in terms of overcoming the load rejection
and good robustness system for future BLDC motor change [11] Ameer L. Saleh and Adel A. Obed, "Speed control of
brushless DC motor based on fractional order PID
parameters. The parameters of controllers are optimized using
controller," International Journal of Computer
IWO algorithm which given optimal value with stable
Applications, vol.95 no.4, 2014.
operation. Also, the BLDC motor mathematical
representations are performed by using Matlab simulation.

6242
 International Journal of Applied Engineering Research ISSN 0973-4562 Volume 13, Number 8 (2018) pp. 6234-6243
© Research India Publications. http://www.ripublication.com

[12] Abbas K. Kadhim and Adel A. Obed, " Brushless DC Technology Thesis in Electrical Engineering,
motor speed control based on PID controller with National Institute of Technology, Durgapur, India,
2dof and anti-windup techniques," The 2nd (May 2013).
Engineering Conference for Graduate Research,
[23] Byeong-Ho Jeong, Nam-Hoon Kim and Kang-Yeon
Middle Technical University-Electrical Engineering
Lee, "Optimized digital proportional integral
Technical Collage, Baghdad, Iraq, 20-21 Sep. 2017.
derivative controller for heating and cooling injection
[13] Maohua Zhang, Changliang Xia, Yang Tian, Dan Liu molding system," Journal of Electrical Engineering
and Zhiqiang Li, "Speed control of brushless dc and Technology, vol.10, no.3, pp. 1383-1388, 2015.
motor based on single neuron pid and wavelet neural
[24] Ali Reza Mehrabian and Caro Lucas, "A novel
network," IEEE International Conference on Control
numerical optimization algorithm inspired from weed
and Automation Guangzhou, CHINA, pp. 617-620,
colonization," Ecological informatics, vol.1, no.4,
June 2007.
pp.355-366, 2006.
[14] Gaviphat Lekutai., "Adaptive self-tuning neuro
[25] Gourab G.R., Swagatam D., Prithwish C. and
wavelet network controllers", Doctor of Philosophy
Ponnuthurai N.S., "Design of non-uniform circular
Thesis in Electrical Engineering, Virginia
antenna arrays using a modified invasive weed
Polytechnic Institute and State University, March
optimization algorithm," IEEE Transactions on
1997.
antennas and propagation, vol.59, no.1, pp.110-118,
[15] José Alberto Cruz Tolentino, Alejandro Jarillo Silva, 2011.
Luis Enrique Ramos Velasco and Omar Arturo
[26] F.M. Monavar and N. Komjani, "Bandwidth
Domínguez Ramírez, "Wavelet PID and Wavenet
enhancement of microstrip patch antenna using
PID: Theory and Applications, In PID Controller
jerusalem cross-shaped frequency selective surfaces
Design Approaches-Theory," Tuning and
by invasive weed optimization approach," Progress
Application to Frontier Areas InTech, 2012.
In Electromagnetics Research, vol.121, pp. 103-120,
[16] Stephane Mallat, "A wavelet tour of signal 2011.
processing," Academic Press, 2nd Edition, ISBN- 13:
[27] Souad Mekni and Besma Chaar Fayech, "A modified
978-0-12-466606-1, 1998.
invasive weed optimization algorithm for
[17] Khalaf Salloum Gaeid and Hew Wooi Ping, multiobjective flexible job shop scheduling
"Wavelet fault diagnosis and tolerant of induction problems," Computer Science & Information
motor," International Journal of the Physical Technology, DOI 10, pp. 51–60, 2014.
Sciences, vol.6, no.3, pp.358-376, February 2011.
[28] PITTMAN_EC121A_Brushless_Servo
[18] M. A. S. K. Khan, M. Azizur Rahman, _Motor,http://www.pittmanmotors.com/Portals/0/Su
"Implementation of a new wavelet controller for bPageImages/BrushlessDCMotors/InstrumentGrade
interior permanent-magnet motor drives," IEEE BrushlessMotors/EC121A/PITTMAN_EC121A_Bru
Transactions on Industry Applications, vol.44, no.6, shless_Servo_Motor.pdf, 2016.
pp. 1957-1965, 2008.
[19] Chong Tan and Christian Nørgaard Storm Pedersen,
"Financial time series forecasting using improved
wavelet neural network," Master of Computer
Science Faculty of Science, Univesity of
Copenhagen, Copenhagen, 2009.
[20] Sung Jin Yoo, Jin Bae Park and Yoon Ho Choi,
"Direct adaptive control using self-recurrent wavelet
neural network via adaptive learning rates for stable
path tracking of mobile robots," IEEE American
Control Conference , vol.1, pp.288-293, June 2005.
[21] M. A. Saldaña-O, L. E. Ramos-V, J. P. Ordaz-O, A.
Garcia-Barrientos, I. Algredo-Badillo, J. F. Balmat
and F. Lafont, "A comparative study of the wavenet
PID controllers for applications in non-linear
systems," In Electrical Engineering, Computing
Science and Automatic Control (CCE), 2015 12th
International Conference on IEEE, pp.33-38, October
2015.
[22] Nishant Kumar, "Optimal design of fir and iir filters
using some evolutionary algorithms," Master of

6243

View publication stats