This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.3007615, IEEE Access

Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000.
Digital Object Identifier 10.1109/ACCESS.2017.Doi Number

Speed control of direct current motor using

ANFIS based hybrid P-I-D configuration
controller
First A. Guo Yanling1, and Second A. Mohamed Elhaj Ahmed Mohamed 1
1
Mechanical & Electrical Eng. College Northeast Forestry University-Harbin-150040, PR China
Corresponding author: Second A. Mohamed Elhaj Ahmed Mohamed (e-mail: mohaj2007@hotmail.com).

ABSTRACT Considering DC motor’s non-linear characteristics, design challenges and mechanical

variation due to the operation conditions, the precise controller cannot be performed using the traditional
control alone. This study presents an innovative method to optimize the direct current motor speed by
decreasing the transient response characteristics such as overshoot, settling time and rise time. In this study,
several PID’s configurations and ANFIS controllers have been performed for direct current motor speed
control. The suggested techniques were compared to each other to validate the superiority of the highest
controller performance. Training data for ANFIS was generated using standard configuration, cascade and
dual configuration. The dual configuration is a hybrid of two controller’s data manipulated using a special
algorithm. The results obtained using Matlab Simulink tools demonstrated the efficiency of ANFIS based
dual PID and ANFIS based PD methods for controlling the motor speed. The overall simulation results
showed that the models have a rapid response, small overshoot, superior dynamics and robustness, and
static performance. The techniques presented herein outperform the traditional control methods for
obtaining highly accurate control.

INDEX TERMS ANFIS, direct current motor, PID, PI, PD, feedforward data and algorithm

I. INTRODUCTION have presented several controllers to enhance the

The primary purpose of control is to allow the model to performance of the DC motor. PID control is the most
operate reliably and enhance its performance. This goal can suitable technique for industrial application, but it requires an
be achieved by implementing the engineering concepts to accurate mathematical model for the control system [1-3].
control the process. Direct current motor speed control The essential concept for designing a PID controller includes
means adjusting the speed to the required value to realize a tuning of the P, I, and D gains. This may also include several
given operation. Thus, each direct current motor requires a other tuning manual methods such as those in the industrial
controller to adjust its direction or the speed to the desired field manual. Furthermore, some methods are a combination
value. In most cases, the control performs manually (by of inelegant methods such as ANFIS, Fuzzy control,
hand) or by applying some traditional/intelligent control computation algorithms, and neural network [4-8]. During
techniques. The motor produces the most needed the operation, the disturbance and changing of the parameters
mechanical motion, which is categorized into direct current can destabilize the system. Due to the recent increase in the
and alternate current motors. Indeed, motors commonly system’s complexity, there are often challenges when tuning
used to convert the direct or alternating current to the PID’s parameters to obtain perfect transient
mechanical energy has been implemented in several characteristics and stable systems. Therefore, control
applications and equipments. The increased precision and designers have switched to intelligent strategies to obtain
accurate motor movement dynamic has resulted in optimal performance [9, 10].
widespread use of high performance direct current motors. As mentioned, control designers have applied intelligent
techniques to avoid drawbacks related to the traditional
Regulation of the DC motor speed is essential as it allows the method. Fuzzy control is one of the modern methods applied
generation of an accurate and precise speed. Previous studies to enhance the control of the DC motor. Recently, an

VOLUME XX, 2017 1

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.3007615, IEEE Access

advanced technique for tuning scale factor of Fuzzy PD-I suggested controller showed its feasibility and suitability
model combined with anti-windup was developed to avoid under different situation[22].
the overall gain of the model and limiting of the oscillatory in Having considered the points outlined above, various studies
the controlled variables [11]. Model-reference adaptive fuzzy on the control of the speed of DC motors have been
was recommended for the control of the DC motor’s speed; conducted. The studies demonstrate how fuzzy logic control
the controller required a reference model for training FIS, and ANFIS control can provide suitable methodologies to
and an oscillation appeared in the speed response[12]. On the determine the best control. Meanwhile, DC motor speed
other hand, due to the rapid development in the application, could be regulated using several techniques. In this regard,
several studies have used fuzzy logic combined with ANFIS most of the previous studies have some drawbacks. For
to improve control efficiency. In[13], fuzzy-based PID online instance, their main goal is to minimize the transient response
supervised adaptive network FIS based was introduced for parameters, but there are often deviations from this goal, and
controlling the DC motor speed, obtained results using there are significant similarities in most techniques used. In
Simulink toolkits showed perfect performance in transient order to avoid these inefficiencies; this work presents the
response characteristics. ANFIS has been used to regulate speed control of the DC motor by controlling the armature
the speed of the DC motor, and the efficiency was compared voltage. In this method, the control is done using ANFIS
to the traditional PI controller and fuzzy-based PID control based on several PIDs forms as a controller for the DC
[14]. Although the controllers used in the comparison motor. The PID model design is used as a basis to prepare
showed excellent performance in settling time and rise time, training data of the proposed controllers. For comparison
there was a significant overshoot and small steady-state error. purposes, the DC motor model was operated using designed
ANFIS, as a modern technique, has replaced traditional controllers. Furthermore, the simulation model was
methods in many applications. In order to decrease the performed repeatedly to meet different conditions, including
transient response parameters such as the overshoot, rise time sudden change of the speed values or change in the input
and settling time for SEDC motor, an intelligent control signal characteristics. The efficiency of the suggested
based ANFIS has been proposed. Simulation results have controllers were validated using various kinds of speed
demonstrated suitable rise time and sustainable overshoot signals. This study aims at implementing cascade and dual
with oscillation before reaching the steady-state[14]. On the block to control the DC motor speed.
other hand, different researchers have applied trained This paper is organized as follows: The second section
feedforward of ANFIS based on several algorithms. describes the mathematical model of the DC motor, and all
Premkumar introduced ANFIS online-based GA-PSO relevant mathematical equations have been detailed. Part
optimized BLDC motor speed control has been by using a three of this paper discusses the introduced controllers, while
hybrid algorithm [15], while other studies [16, 17] part four illustrates the results and curves attained by
introduced ANFIS based rotor position control for DC motor. introduced controllers using Matlab toolkits. Finally, the
A novel bacterial Foraging PSO and BAT, ANFIS conclusions drawn from this study are shown in part five.
Optimization Algorithms based speed control has been II. MOTOR MODEL
introduced to (MC)-fed brushless dc motor, where the results DC motor used for this study is separately-excited and has
proved superior performance to other techniques under the ability to separately control over the speed and torque by
various operation conditions[18]. regulating the field current and armature current. According
Generating data for training ANFIS is an essential aspect of to figure 1, the developed torque is illustrated using equation
designing the ANFIS controller[19]. In[20] fractional-order 1 and 2, where the armature current is I (t), angular speed is
PID was designed with the GA algorithm to train ANFIS Ꙍ, interior friction is J; the friction coefficient is β, and
data where three ANFIS configurations were formed using torque constant is ka. Armature voltage e (t) is given by
training data to tune the parameters during the operation. In equation 3, where Kb Back emf constant, an applied voltage
other studies, speed control based ANFIS was developed is symbolized by V (t) and can be calculated using equation
using three layers backpropagation algorithm [22]. In this 4. L represents the Armature inductance, whereas R
model, the parameters measured only focused on the rise represents resistance. Equation 5 explains the motor transfer
time and the settling time, which both showed high values for function. Table1 describes the parameters of the dc motor.
the designed controller and controllers introduced for
comparing the performance [22]. Moreover, ANFIS based
PSO and recursive-least-square has been used for DC motor
speed controller as multi-level ANFIS, and the method
provides optimal performance[21]. For the ANFIS and PI-
based MPPT algorithm controllers presented for control
BLDC motor speed, the performance of the designed method
was evaluated by comparing to the observed method. The

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.3007615, IEEE Access

FIGURE 2: The model used to produce training data for the dual
controller.

FIGURE 1. DC motor equivalent circuit.

The PIDs controllers were used to generate training data for
ANFIS configuration. The DC motor was controlled by
d  (t )
T (t )  J   B   (t ) several ANFIS controllers based on different PIDs forms.
dt
The first controller was ANFIS based PID. The transfer
T (t )  K a  i (t ) function of the first PID controller, as well as the PID2
designed to work with the PID1, are explained in tables II and
ea (t )  K a  (t )
III. The PID2 cannot provide a high performance for the
di (t ) control of the DC motor transfer function but only clarifies
v (t )  L   Ri (t )  ea (t )
dt the controller Strict. The controller transfer function was
 s Ka listed in table III. The PID2 controller was built based on the

v(s) k kb  ( Js  B )  ( Ls  R )
a
parameters, as illustrated in figure 9. In the last step, the two
TABLE I PIDs merged to form the dual PID controller, which showed
THE PARAMETERS OF THE DC MOTR a high performance, although the dynamic was still too slow.
The function of the dual PID controller is described in table
Parameter Value
IV. For the slow speed of the controller response, a controller
Armature inductance L 0.005 H was implemented to create data for training ANFIS model to
Armature resistance R 0.1 Ω solve the problem of the response speed. However, the data
Rotor inertia J 0.1 N.m2
Viscous friction coefficient B 0.05 Nms/rad generated was too big, and the extracted data was therefore
Back emf constant Kb 0.004Vs/rad minimized in order to increase the controller speed response,
Torque constant Ka 0.0036 Nm/A leading to high performance.
The artificial neural network is a new control strategy, which
III. PROPOSED CONTROL TECHNIQUE manipulates the numerical entities to perform the divergence
To produce suitable training data for ANFIS controller, this and convergence among them. The incorporation of fuzzy
study was conducted in many PIDs configurations to prove and neural network produced a modern method that gained
the perfect configuration for training ANFIS. The the features of both strategies and resulted in a significant
configurations implemented the PI, PD and PID. Cascade improvement in nonlinear mapping, modelling, learning and
compensated for previous controllers which we were labelled pattern realization. In this context, several models are
as dual PI, dual PD and dual PID. The dual-controller was designed to generate ANFIS training data and to demonstrate
then manipulated using an algorithm to conduct the the impact of the data in the performance of the controller.
convergence and divergence among the two models. The models that have been built are commonly known as the
In this stage, the DC motor function was operated under PI, PD and the PID. Moreover, new forms of PIDs have been
different controllers. The linearization Continuous-time presented, such as dual PI, dual PD and dual PID. Each dual-
transfer function of the dc motor used to apply the control block is a hybrid of the two configurations, which were later
was as follows: manipulated using the convergence and divergence
algorithm, as explained in the flowchart in figure4.
7.2
The efficiency of standard configurations is predictable, but
--------------------
the performance of the newly presented methods might be
s^2 + 20.5 s + 10.03
more optimized. By observing the motor behaviour, the data
In this strategy, the PIDs configurations were used to was generated and loaded in the neuro-fuzzy designer. FIS
generate training data for ANFIS control. The DC motor was was created based on the grid partition with function two
controlled by several ANFIS controllers based on different inputs (linear type- each one is four triangle membership
PIDs forms. The first controller was ANFIS based PID. The function). The optimization method for training the FIS data
model used to produce training data for the dual controller is was back propagation with error tolerance zero and epoch 10,
illustrated in figure 2. and the data was subsequently tested for error. ANFIS in this
method was implemented using 53 nodes, and the total

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.3007615, IEEE Access

number of parameters was 16 linear parameters and 24 =(Ak ꓵBk )). Sk was used to set out the symmetric difference
nonlinear parameters. The number of data pairs were then among the two data sets, and it was represented in the
checked using the 16 fuzzy rules, and by adjusting the inputs algorithm as Sk={ j:( j Є Ak )or (j Є Bk ). The variable Rk was
rules to get an optimal response. used to carry the union of the two data items (Rk =Uk ꓴ Sk).
The numerical entities were loaded in FIS to track the I/O for The counter k increased until the completion of the criterion.
the provided data, and on-line was performed to identify the In this controller, ANFIS are extracted using models
objects during the operation, which needs the neuron network explained previously in tables II, III and IV, where the
structure suitable for the learning strategy. The most double controller has implemented previously and as PIPD
important factor is the speed of the learning mechanism, and and proved a good performance[28].In the ANFIS process,
if the configuration is too complicated it could affect the firstly the training data is loaded from the model used to
adapting speed, making it difficult to trace the objects. For all generate it. It is secondly converting this training data to
the suggested controllers, an FIS based Takagi-Sugeno model fuzzy set form based on suitable member function; this step
performed an ANFIS configuration with two inputs, where called fuzzification and membership function selection. Then
one output was later utilized as a control signal[23]. creating a fuzzy rule frame(if-the-else)corresponding to the
Design of (ANFIS) controller is the most approved intelligent fuzzified input weight. Fourth, updating the parameters using
controls strategies [24]. ANFIS is a combination of Fuzzy the training data algorithm based on error tolerance. In the
logic, and neural network, both of them have some final step, which is a defuzzification step, the fuzzified data
components and objectives. Both techniques attempt to figure will be reorganized as numerical data to run as an ANFIS
out the functioning of human knowledge. In both methods, model.
the theorem is to store the expert’s knowledge to make a By considering a fuzzy inference system with two inputs (x,
suitable decision. In essence, adaptive neural fuzzy is a kind y), and one output (f), The rule base should be created with
of fuzzy logic enhanced by a neural network to develop the two Takagi–Sugeno fuzzy rules type [22].ANFIS rules form
performance of the control [25].ANFIS is fuzzy rules based on Takagi–Sugeno is given as in equations below,
enhanced by a neural network to carry out the system
behavior based on I/O data. The combination of these Rule 1: If X is A1 and y is B1 , then
intelligent strategies hybridizes the Takagi–Sugeno with the
learning capabilities. The learning process role is to regulate f1  p1 x  q1 y  r1 ;
the fuzzy inference system parameters.
The main objective of ANFIS is to implement the
configuration of a neural network to develop the process of Rule 2: If X is A2 and y is B2 , then
fuzzy parameters to represent the neuron’s weight..
Therefore, ANFIS can automatically recognize by regulating f2  p 2 x  q2 y  r2 ;
the MF and modifying the neural-network weight using a
suitable algorithm. As a consequence, the neural network Where fuzzy antecedent sets A1,A2,B1 and B2 are the
architecture can be optimizing the membership function, consequent crisp function. Fz is the output resulted by rule
since the fuzzy configuration built based three parts, fuzzy z,pz,qz and rz are the parameters resulted by rule z. Layer1
input, inference mechanism, which is known as fuzzy rules has N nodes; these nodes numbers are equal to the variables
and the output of the system, where several techniques can number of that transmitted data to each node of the next
implement all these parts. Usually, ANFIS uses Takagi– layer. Each neuron in this layer uses its transfer function to
Sugeno, where it is applicable in controlling and modelling calculate a certain degree of a specific fuzzy subset. In this
complicated process. In this design of this controller as layer executing the fuzzification strategy to carry out the
presented in figure 2, this structure mainly contains five output according to the equation below,
layers which are rule base, database, decision unit,
fuzzification and defuzzification unit. ANFIS’s every layer is
1
corresponding to the Takagi Sugeno FIS[26, 27]. oi1   A1 (  )  b1
The flowchart of the algorithm, as in figure 4, is describing   x  c 2 
1   
  a1  
all steps of manipulating data generated to train the ANFIS 1

controller. The algorithm established，three groups（ i， k

 
， j） were initially were set to zero = 0 for use as index and
counter, where（ i ）and（ j ）are the data set of two Where the parameters a and c using to modify the MF shape.
controllers (cascade controller and master controller), and（ ANFIS model has two inputs with four trmf and one linear
k） is the new hybrid data set (dual controller). The variables output type. 16 FIS rules were implemented. The FIS was
Ak ,Bk ,Uk ,Sk ,Rk are defined . Uk was used to set the optimized using backpropagation for training the numerical
intersection between two inputs that represent the data in Ak entities. According to previous factors, the implemented
and Bk , where they were symbolled in the algorithm as (Uk neural network was as demonstrated in figure 3.

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.3007615, IEEE Access

FIGURE 3：The architecture of ANFIS network.

The second layer of ANFIS controller is calculating the MF
and the weight facture values using explained equation
bellow, which is presenting the fuzzy-sets, where each
neuron is a rule premise.

wi   A1 (  )* B 1 ( y)
The normalization of the relation between the fuzzy inputs
and realizing of fuzzy operation it is the responsibility of the
third layer.

wi
wi  (i  1, 2)
w1  w2
In the fourth layer, the nodes compensate for the operation
and execute the rules of the fuzzy system. The output of this
layer has explained in equation bellow.

oi4  wi * fi  wi *( p1 x  q1 y  r )
The defuzzification process is performed in the fifth layer,
where all input signals summation are summed in this layer
as explained in the equation below.

w * f i i
o   w i * fi
5
 i

w
i
i ii
i FIGURE 4：The algorithm of convergence and divergence.

The suggested controllers were adjusted correctly and

investigated to ensure their efficiency. They were also
validated to make the data set output predictable. In order to
increase the speed of the dynamic response, generated data
was minimized to the size that is able to gain the target. The
models were operated using different input signals and speed
in order to increase the efficiency. An optimal controller was
obtained through several modifications of the error tolerance,
membership functions and optimization methods. The
schemes used to perform ANFIS based dual PID are
demonstrated in figures 5 and 6, respectively. Figure 7
demonstrates the Simulink model ANFIS dual controller.

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.3007615, IEEE Access

possible. The specifications of the PD2 are illustrated in

figure 9. For the dual PD that was operated as in previous
controllers, the transfer function of the dual PD controller is
illustrated in table IV. The parameters obtained using the
dual PD controller are shown in figure 10.
TABLE II
THE TRANSFER FUNCTIONS OF SUGGESTED CONTROLLERS.
Configuration First configuration
transfer
function
PI 10.24 s + 5.601
FIGURE 5：Dual PID training errors for ANFIS. ---------------
s
PD 7.598e05 s + 1.704e07
---------------------
s + 2.502e04

PID 326.1 s^2 + 7762 s + 5345

-------------------------
s^2 + 850.9 s

TABLE III
THE TRANSFER FUNCTIONS OF SUGGESTED CONTROLLERS.
Configuration Cascade configuration
transfer
function
PI 3.888 s + 0.8321
FIGURE 6：dual PID training data for ANFIS. ----------------
s
PD 1.893 s + 169.3
---------------
s + 100

PID 2.139e07 s^2 + 1.378e10 s + 1.807e12

------------------------------------
s^2 + 2.395e06 s

FIGURE 7: the Simulink model ANFIS dual controller. TABLE IV

For the PI controller, similar procedures of PID were THE TRANSFER FUNCTIONS OF SUGGESTED CONTROLLERS.
followed. First, the DC motor function was tuned using PI1 Configur- Dual configuration
controller, and the TF is illustrated in table II. The controller ation
transfer
was adjusted to the parameters, which is illustrated in figure function
8. For the PI2 controller, the transfer function is shown in PI 39.8 s^2 + 30.3 s + 4.661
table III, and the parameters are demonstrated in figure 9. -------------------------
s^2
Designing of the dual PI controller achieved using the
parameters is demonstrated in table IV. The transfer function PD 1.438e06 s^2 + 1.609e08 s + 2.885e09
of the PI2 controllers are shown in figure 9. For the dual ------------------------------------
controller’s blocks, no warnings or incompatibilities were s^2 + 2.512e04 s + 2.502e06
identified in the linearization, and each block was accurately PID 6.975e09 s^4 + 4.658e12 s^3 + 6.963e14 s^2 + 1.4
linearized. The dual PI controllers were tuned to the 1e16 s + 9.659e15
parameters as in figure 10. ------------------------------------------------------------
-----
Similar steps were repeated for the PD controller to confirm s^4 + 2.396e06 s^3 + 2.038e09 s^2
the performance effect of the dual control. The PD1 controller
was implemented with the transfer function as shown in table
II. The controller was tuned to the P and D action, which
were obtained, as explained in figure 8. The PD2 controller
transfer function only used the proportional action because
when the dual PD utilized the double derivative action, the
system became unstable, and the transfer function is shown
in table III. The derivative action was minimized as low as

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.3007615, IEEE Access

FIGURE 11: The step response of the dc motor function without control.

In this study, the error for each controller was measured and
the motor behavior observed in order to return the controller
FIGURE 8: The parameters of the first configuration and generate suitable data for training ANFIS and
subsequently ensuring optimal performance. For ANFIS
based PI1 and PID1, a significant error was observed because
the controllers adjusted to maximum performance and were
capable of working with the second controller. For the
ANFIS based PD1 reaction, the controller showed minimum
error. The error response of the three controllers is illustrated
in figure 12.

:
FIGURE 9 The parameters of the cascade configuration

FIGURE 12: The error rate of three ANFIS configurations based PI1, PD1,
and PID1

The cascade configuration's controllers were operated

separately from the master controllers, for the cascade
ANFIS based PI2 and PD2, the significant error was
generated by these controllers in the error response (Figure
13).

:
FIGURE 10 The parameters of the dual configuration

IV. RESULTS AND DISCUSSION

First, the dc motor was operated without control to determine
the effect of the control and the response showed inefficiency
and sizeable steady-state error (Figure 11).

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.3007615, IEEE Access

controllers, where the PD1 controller demonstrated better

performance. For the ANFIS based PI1 as PID1, the speed
response showed a delay in settling time and rise time. The
transient parameters were listed in table III and figure 22.

FIGURE 13: Error produced by the cascade PI2 and PD2

The second PID2 controller showed the error produced by the FIGURE 16: The speed response of the ANFIS controller-based PI1, PD1,
PID2 without the master PID (Figure 14). and PID1.

In the next operation, the dc motor control performance was

evaluated using the cascade controller, which was
subsequently termed as the second controller. Figure 17
shows the response of the ANFIS based PID2 separated from
the PID1, while table V and figure 22 show the transient
parameters of the PID2 controller. The controller was built to
support the first one, so the response was quite different from
the desired speed.

FIGURE 14: Error produced by the cascade PID2

In the last level of tracking, an error was generated for the

dual controllers, as shown in figure 15.

FIGURE 17: The response of the ANFIS based PID2

With respect to the ANFIS based, the PI2 controller alone

showed a varied response to the step input signal (figure18).
The response was as a result of using the PI2 controller far
from the ideal response, but the objective of the controller is
to work with the PI1 controller. The P and I action is a
FIGURE 15: Error produced by the dual controllers.
suitable asset value and can be achieved after repeating the
tests many times. For the ANFIS based, PD2 controller
After controlling the DC motor speed using ANFIS based
transfer function only used the proportional action because in
PID1 controller, the response was different from that of the
case of utilizing double derivative action. It made the system
input signal, but after repeated tests, there was improved
unstable, so the transfer function is P=1.693, and the
performance, making it suitable for controlling again with the
derivative action minimized as small as possible. The effect
PID2 controller. Figure 16 shows the controller response. The
of the PD2 alone was shown in table 5 and figure 22, while
transient parameters of the controller are explained in table 3
figure 18 illustrates the response of the ANFIS based PD2
and figure 22. Figure 16 also demonstrates the dc motor
controller.
speed response of the first ANFIS based PI1 AND PD1

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.3007615, IEEE Access

FIGURE 19: Speed response using ANFIS based dual PI, PD and PID
FIGURE 18: Speed response using ANFIS based PI2 and PD2

Finally, the two controllers, i.e. the master and cascade were
merged. For ANFIS based dual PID. The controller showed
high performance. The ANFIS based dual PID controller was
performed to generate data for training ANFIS to solve the
problem of the response speed. The data generated was too
large; thus, it was reduced to the least possible amount to
increase the controller speed response, which subsequently
resulted in high performance. The results of the controller
response using ANFIS based dual PID controller was
illustrated in figure 19. The transient parameters of the
ANFIS based dual PID controller was listed in table V and
figure 22. For the designing of the ANFIS based dual PI FIGURE 20: Speed response using ANFIS based dual PD and PID in the
controller, the controller showed varied response for speed range 0.1 second

(figure 19). As it is apparent in the figures, the ANFIS based

dual PI controller showed better settling time and rise time
than the ANFIS based PI1 controller, but simultaneously
gained significant overshoot and oscillations.
Furthermore, any attempts to decrease the overshoot resulted
to an increase in the rise time and settling time. Therefore,
the option of a dual PI controller should be ignored if the
overshoot is not an important factor. As evident in table V
and figure 19, ANFIS based dual PD controller shows an
improvement in all transient response aspects. Figure 20
shows the superiority of the ANFIS based dual PID and PD
at time 0.1 seconds. The ANFIS based dual PID and PD were
operated under sudden change of the speed, where the speed
at time 0.25 second rose up to 56 rad/sec, and slowed down
to 52 rad/sec at time 0.74, proving the efficiency of the FIGURE 21: Speed sudden change response using ANFIS based PI2and
controllers (Figure 21). ANFIS based dual PID showed PD2
optimal performance in all aspects.
TABLE V
THE TRANSIENT CHARACTERISTICS OF THE PROPOSED CONTROLLERS
Rise time Settling
The controller time(second) Overshoot %
(second)
ANFIS-PI1 0.474 0.809 0.609
ANFIS-PD1 0.00966 0.0162 0.562
ANFIS-PID1 0.587 2.31 2.75
ANFIS- PI2 4.16 13.5 0
ANFIS- PD2 1.92 3.46 0
ANIF- PID2 1.66 6.08 7.77
ANFIS-dual PI 0.1.6 0.385 12.1
ANFIS-dual PD 0.00573 0.00984 0.376
ANFIS-dual PID 0.000107 0.000209 0

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.3007615, IEEE Access

Conversion and Management, vol. 52, no. 1, pp. 264-273,

2011.
[4] K. S. Alli, "A LabVIEW-based online DC servomechanism
control experiments incorporating PID controller for students’
laboratory," The International Journal of Electrical
Engineering & Education, p. 0020720919868142, 2019.
[5] S. Peicheng, W. Suo, Z. Rongyun, and X. Ping, "Study on the
fuzzy proportional–integral–derivative direct torque control
strategy without flux linkage observation for brushless direct
current motors," International Journal of Advanced Robotic
Systems, vol. 16, no. 3, p. 1729881419853141, 2019.
[6] H. Hu, T. Wang, S. Zhao, and C. Wang, "Speed control of
brushless direct current motor using a genetic algorithm–
optimized fuzzy proportional integral differential controller,"
Advances in Mechanical Engineering, vol. 11, no. 11, p.
1687814019890199, 2019.
[7] B. Hekimoğlu, "Optimal Tuning of Fractional Order PID
Controller for DC Motor Speed Control via Chaotic Atom
Search Optimization Algorithm," IEEE Access, vol. 7, pp.
38100-38114, 2019.
[8] J. Agarwal, G. Parmar, R. Gupta, and A. Sikander, "Analysis of
grey wolf optimizer based fractional order PID controller in
speed control of DC motor," Microsystem Technologies, vol.
24, no. 12, pp. 4997-5006, 2018/12/01 2018.
[9] D. Puangdownreong, (in En), vol. 13, 01/01 2018.
[10] C.-H. Kim, "Multi-loop PID control method of brushless DC
FIGURE 22: The transient characteristics of the proposed controllers motors via convex combination method," transfer, vol. 100, p.
9, 2017.
V. Conclusion [11] A. R. Sartorius, J. de Jesus Moreno, O. Pinon, and A. E. Ruiz,
This study evaluated the effect of the feed-forward data used "A new approach for adjusting scale factor in fuzzy PD+ I
controllers with anti-windup," Journal of Intelligent & Fuzzy
to set ANFIS control by visualizing the DC motor speed Systems, vol. 27, no. 5, pp. 2319-2326, 2014.
behavior. For the training data for ANFIS controller, several [12] H.-P. Wang and Y.-T. Liu, "Integrated design of speed-
models of PIDs were implemented and operated under sensorless and adaptive speed controller for a brushless DC
motor," IEEE Transactions on Power Electronics, vol. 21, no.
various conditions to investigate the controllers' performance. 2, pp. 518-523, 2006.
Based on Matlab Simulink tools, several characteristics were [13] K. Premkumar and B. Manikandan, "Fuzzy PID supervised
observed, such as steady-state error, the rise time, overshoot, online ANFIS based speed controller for brushless dc motor,"
Neurocomputing, vol. 157, pp. 76-90, 2015.
and the settling time, which are all important in developing [14] P. Tripura and Y. Babu, "Intelligent speed control of DC motor
the dc’s motor performance. ANFIS showed simplicity in using ANFIS," Journal of Intelligent & Fuzzy Systems, vol. 26,
execution and learning susceptibility. Furthermore, the no. 1, pp. 223-227, 2014.
implementation of the rules was easy compared to the [15] K. Premkumar and B. Manikandan, "GA-PSO optimized online
ANFIS based speed controller for Brushless DC motor,"
retuning of the traditional control methods. The results Journal of Intelligent & Fuzzy Systems, vol. 28, no. 6, pp. 2839-
demonstrated that ANFIS based dual PID has superior 2850, 2015.
performance, high robustness, and excellent accuracy without [16] M. J. Prabu, P. Poongodi, and K. Premkumar, "Fuzzy
supervised online coactive neuro-fuzzy inference system-based
oscillation for the speed control of the DC motor. The rotor position control of brushless DC motor," IET Power
controller also showed high efficiency and distinct Electronics, vol. 9, no. 11, pp. 2229-2239, 2016.
performance in tracking the motor speed compared to other [17] C. Navaneethakkannan and M. Sudha, "Analysis and
implementation of ANFIS-based rotor position controller for
suggested and previously studied controllers. The constraints BLDC motors," Journal of Power Electronics, vol. 16, no. 2,
of the method are that the performance of any pair of pp. 564-571, 2016.
controllers is not equal, and the response of the conventional [18] T. Sivarani, S. J. Jawhar, and C. A. Kumar, "Novel bacterial
model used to generate data is too slow. foraging-based ANFIS for speed control of matrix converter-
fed industrial BLDC motors operated under low speed and high
torque," Neural Computing and Applications, vol. 29, no. 12,
REFERENCES pp. 1411-1434, 2018.
[1] R. Akkaya, A. Kulaksız, and Ö. Aydoğdu, "DSP [19] G. Yanling and M. E. A. Mohamed, "Study on the Extent of the
implementation of a PV system with GA-MLP-NN based Impact of Data Set Type on the Performance of ANFIS for
MPPT controller supplying BLDC motor drive," Energy Controlling the Speed of DC Motor," Journal of Engineering
Conversion and Management, vol. 48, no. 1, pp. 210-218, and Technological Sciences, vol. 51, no. 1, pp. 83-102, 2019.
2007. [20] H. Arpaci and O. F. Ozguven, "Design of adaptive fractional-
[2] O. Aguilar-Mejía, R. Tapia-Olvera, A. Valderrabano-González, order PID controller to enhance robustness by means of
and I. Rivas Cambero, "Adaptive neural network control of adaptive network fuzzy inference system," International
chaos in permanent magnet synchronous motor," Intelligent Journal of Fuzzy Systems, vol. 19, no. 4, pp. 1118-1131, 2017.
Automation & Soft Computing, vol. 22, no. 3, pp. 499-507, [21] H. Suryoatmojo, M. Ridwan, D. C. Riawan, E. Setijadi, and R.
2016. Mardiyanto, "HYBRID PARTICLE SWARM
[3] M. Demirtas, "Off-line tuning of a PI speed controller for a OPTIMIZATION AND RECURSIVE LEAST SQUARE
permanent magnet brushless DC motor using DSP," Energy ESTIMATION BASED ANFIS MULTIOUTPUT FOR BLDC
MOTOR SPEED CONTROLLER."

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.3007615, IEEE Access

[22] A. A. Hepzibah and K. Premkumar, "ANFIS current–voltage of China Association of popular science writers, and standing director of
controlled MPPT algorithm for solar powered brushless DC material Committee of World 3D Printing Association. She has presided
motor based water pump," Electrical Engineering, pp. 1-15, over more than 20 projects of NSFC, published more than 150 papers, and
2019. obtained 8 authorized invention patents and 30 utility models. She presided
[23] M. R. Faieghi and S. M. Azimi, "Design an Optimized PID over four provincial teaching reform projects and won three provincial
Controller for Brushless DC Motor by Using PSO and Based awards as the first adult. In the research and development of new 3D
on NARMAX Identified Model with ANFIS," in 2010 12th printing equipment and CNC system of special machine tools, he has many
International Conference on Computer Modelling and research results. It is proposed that biomass composite materials are used
Simulation, 2010, pp. 16-21. as 3D printing consumables, which are successfully used in laser sintering
[24] H. Bevrani and S. Shokoohi, "An Intelligent Droop Control for additive manufacturing raw materials. It is called the fourth kind of
Simultaneous Voltage and Frequency Regulation in Islanded additive manufacturing materials besides metal, inorganic non-metal and
Microgrids," IEEE Transactions on Smart Grid, vol. 4, no. 3, polymer materials by the industry and domestic and foreign media.
pp. 1505-1513, 2013.
[25] S. S. Chong, A. A. B. A. Raman, S. W. Harun, and H. Arof, SECOND A. MOHAMED ELHAJ AHMED
"Dye Concentrations Measurement Using Mach–Zehner MOHAMED. (M’74) received the B.S. degree in
Interferometer Sensor and Modeled by ANFIS," IEEE Sensors computer engineering from technologic and
Journal, vol. 16, no. 22, pp. 8044-8050, 2016. Science University, Khartoum, Sudan, in 2002 and
[26] A. Senthilkumar and P. A.-D.-V. Raj, "ANFIS and MRAS-PI the M.S. degree in computer and network
controllers based adaptive-UPQC for power quality engineering from the department of electrical
enhancement application," Electric Power Systems Research, engineering, Khartoum University, Sudan, in
vol. 126, pp. 1-11, 2015. 2007. He is currently pursuing the PhD degree in
[27] K. Premkumar and B. Manikandan, "Adaptive neuro-fuzzy mechatronic engineering at Northeast Forestry
inference system based speed controller for brushless DC University, School of mechanical and electrical
motor," Neurocomputing, vol. 138, pp. 260-270, 2014. engineering-Harbin –china.
[28] M. Elhaj, A. Mohamed, and G. Yanling, "Study on the extent of From 2002 to 2006, he was an Assistant teacher and labs director at
the impact of data set type on the performance of anfis for Krrary University, Khartoum, Sudan. From 2006-2014, he worked as
controlling the speed of DC motor," Journal of Engineering lecturer in the department of electrical and electronic engineering, Karrary
and Technological Sciences, vol. 51, no. 1, pp. 83-102, 2019. University, Sudan. Mr. Mohamed’s research interest includes the
development of control of dc motor and applying intelligent control
techniques using developed special algorithms.
Mr. MOHAMED. Won china government scholarship, he has also
received an award for the best performance in Chinese language, and has
published a few articles in developing dc motor control and electrical
vehicles motor control.

FIRST A. GUO YANLING Northeast

Forestry University, School of mechanical
and electrical engineering-Harbin –china, the
discipline of mechanical and electronic
engineering, Professor, doctoral supervisor.
Experience: 2008 / 12-2009 / 12, visiting
scholar, Wu Xianming manufacturing
technology center, University of Michigan,
United States, CO Tutor: Professor Ni Jun.
2001 / 09-2003 / 12, Harbin University of
technology, the postdoctoral mobile station of
mechanical engineering / postdoctoral
workstation of Harbin Huawei Group,
postdoctoral, cooperation Supervisor: Professor Zhao Wansheng. 1997 /
09-2001 / 03, doctor, Department of machinery manufacturing and
automation, Harbin University of Technology, supervisor: Professor Zhao
Wansheng. 1987 / 09-1990 / 01, Department of mechanical manufacturing
and automation, Harbin University of Technology, master's degree, and
tutor: Professor Lu Jipei. 1980 / 09-1984 / 07, Department of machinery,
Dalian Institute of Technology, bachelor's degree, tutor: Jiang Xiuying,
associate professor.
Research experiences:2004 / 09 – until now, doctoral supervisor, School
of mechanical and electrical engineering, Northeast Forestry
University.2003 / 09 – till now, Professor, School of mechanical and
electrical engineering, Northeast Forestry University.1996 / 09-2003 / 08,
associate professor, master's guide, School of mechanical and electrical
engineering, Northeast Forestry University.1990 / 09-1996 / 09, lecturer,
School of mechanical and electrical engineering, Northeast Forestry
University.1985 / 09-1990 / 08, Department of machinery, Northeast
Forestry College, assistant.
Professor Guo Yanling, doctoral supervisor, first-class discipline leader
of mechanical engineering, director of 3D printing materials and
technology research and development center of Northeast Forestry
University. Standing member of China Electrical processing
society/deputy director of the International Exchange Committee, director

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.