Neural Computing and Applications

https://doi.org/10.1007/s00521-024-09470-y (0123456789().,-volV)
(0123456789().,-volV)

ORIGINAL ARTICLE

Effective speed control of brushless DC motor using cascade 1PDf-PI

controller tuned by snake optimizer
Emre Çelik1 • Mehmet Karayel2

Received: 29 June 2023 / Accepted: 14 January 2024

 The Author(s) 2024

Abstract
This paper introduces a cascade one proportional derivative incorporating filter (1PDf)-proportional integral (PI) controller
abbreviated as c-1PDf-PI to deal effectively with the speed control issue of brushless DC (BLDC) motors. Two problems
exist with implementing this controller such as iterated integral overflow and derivation-based chattering owing to the
noise. The former is resolved by using an equivalent expression for the integral operation, while the latter is addressed by
putting a first-order filter on the derivative term. To achieve the best performance from the controller, snake optimizer (SO)
is fruitfully employed for optimizing the controller parameters without need for expert knowledge/interpretation. Here, a
more reasonable cost function to assess the candidate solutions is also described. Simulations and laboratory experiments
using DSP of TI TMS320F28335 are performed and the results are presented which show that the reference tracking
performance, torque disturbance capability and robustness of the c-1PDf-PI controller have potential. These results are also
contrasted by those offered by PI and 1PDf speed control schemes individually, affirming the superior performance of our
proposal. As per the results, discussion and observation of this research, we stress that good performance and simplicity are
salient advantages of the c-1PDf-PI controller, rendering it a good alternative over the complicated controller designs.

Keywords Brushless DC motor  Speed control  1PDf-PI control  Snake optimizer  Optimization

1 Introduction sinusoidal counterpart termed as permanent magnet syn-

chronous motor (PMSM), the BLDC motor permits some
Speed control of electric motors is a significant require- important system and operational simplifications while
ment in several industry applications such as CNC, electric generating relatively higher power density [1]. However,
and hybrid vehicles, traction motors, servo systems, pumps the major disadvantage of BLDC motor comes from its
and similar systems that need operation under constant nonlinear model, which makes the controller design task
speed. In line with the technological improvements, the complicated.
needs for higher efficiency and lower size have been The above-described shortcoming of BLDC motor has
growing day-by-day and brushless DC (BLDC) motors inspired many researches reporting advanced control
have gained incontrovertible popularity in this direction, techniques based on sliding modes [2], fuzzy logic [3],
replacing the induction motor (IM) motor and brushed DC adaptive neuro-fuzzy inference system [4], fuzzy neural
motor in several systems. Further, compared with its network PID controller structure [5], nonlinear feedback
controller [6], fractional order concept [7], model predic-
tive control [8], etc. Although these controllers are shown
& Emre Çelik
emrecelik@duzce.edu.tr to yield satisfactory performance, they are based on com-
plex control laws and heavy computational operations,
Mehmet Karayel
mkarayel@karatekin.edu.tr often resulting in an impractical solution for engineering
commissioning. As a result, proposing new control
1
Department of Electrical and Electronics Engineering, schemes that not only provide a practical solution for the
Engineering Faculty, Düzce University, Düzce, Turkey electric drive community but also offer a good controlling
2
Department of Electrical and Energy, Vocational School, effort is always welcome.
Çankırı Karatekin University, Çankırı, Turkey

123
 Neural Computing and Applications

Our rigorous literature survey points out that with researchers in different fields, particularly for the last two
structural variations and modifications on the classical PID decades. The reason behind such a great attention is that
controller, a good option for effective closed-loop control the classical deterministic and gradient-based methods are
can be achieved, though they are recognized as obsolete. In not capable of handling cost functions that are non-convex
this regard, two-degree-of-freedom (2DOF) control is and contain multiple local optimum solutions. The severity
successfully applied to speed control in permanent magnet further increases when the complexity of the system under
brushed DC motor [9] and PMSM [10], which has a non- control grows. These concerns give rise to stagnation in the
linear model. A PI controller is used in [11] to accomplish solutions and it is likely to obtain sub-optimal results while
speed regulation in PMSMs. The controller parameters are more promising solutions are missed out. Nonetheless,
adaptive and adjusted by a fuzzy logic controller (FLC) thanks to the continuous progress made on metaheuristic
with respect to the speed error and its time derivative. algorithms in line with the no-free-lunch theorem [27], new
Similarly, a hybrid method is proposed to achieve an avenues in solution quality, convergence behavior and
effective control of the speed of the BLDC motor [12]. In implementation time have been incorporated to mitigate
the method, PID controller is employed to control the such issues efficiently. Indeed, the literature has been
speed of the motor, whose best gain parameters are tuned witnessing numerous metaheuristic algorithms based on
by student psychology optimization algorithm-based radial inspiration from different concepts and the researchers are
basis function neural network. [13] applied particle swarm found very keen on applying them to acquire unknown
optimization (PSO) algorithm to search for optimal controller parameters of their control system design. Apart
parameters of PID controller in BLDC speed control sys- from the references above, the application of genetic
tem. Experimental results based on TI TMS320F2812 algorithm (GA) to tune all the parameters of FLC was
revealed that PSO optimized PID controller could perform addressed in [28, 29] to enhance speed response profile of
better than conventional PID controller in augmenting the PMSM under speed reference and load torque variations.
step speed response and load impact. A self-tuning PID Parameters of a PI controller in a DSP-based DC servo
control scheme is introduced in [14] to deal with the auto- drive were effectively derived by the SOS algorithm [30].
tuning of PID controller for deriving optimal performance For the same system, PI ? DF technique, as a new control
under wide speed range. It is shown that the proposed scheme, was also suggested and its parameters were tuned
controller is more effective than the classical one with fixed by stochastic fractal search algorithm [31]. Optimal design
parameter gains. On the other hand, gray wolf optimization of a robust parallel Fuzzy P ? Fuzzy I ? Fuzzy D con-
[15]/hybrid firefly algorithm-pattern search [16]/quasi-op- troller based on a GA hybridized by PSO approach was
positional differential search algorithm [17] optimized PI/ introduced for an automatic voltage regulator system [32].
PID, jaya algorithm-invasive weed optimization [18]/optics More recently, a novel imperialist competitive algorithm
inspired optimization [19]/improved stochastic fractal optimized cascade fuzzy-proportional integral derivative
search algorithm [20]/symbiotic organisms search (SOS) incorporating filter (PIDN)-fractional order PIDN (FPIDN-
[21] optimized PID, SOS optimized PIDN [22], hybrid FOPIDN) controller [33], DSA optimized FOPI-FOPD
invasive weed optimization-pattern search optimized controller [34] and salp swarm algorithm optimized fuzzy
2DOF-PID [23], biogeography based optimization-based 1 ? proportional derivative-tilt integral (F1PD-TI) con-
three-degree-of-freedom-PID (3DOF-PID) [24], sine– troller [35] have been also appeared in the literature to
cosine algorithm optimized proportional derivative-pro- better address LFC issue. To our best knowledge, it seems
portional integral derivative with double derivative filter there is no prior research taking advantages of a cascade
(PDPID ? DDF) [25] and dragonfly search algorithm one proportional derivative with filter (1PDf)-proportional
(DSA) tuned (1 ? PD)-PI [26] controllers have also integral (PI) (c-1PDf-PI) controller in any control system.
appeared to resolve load frequency control (LFC) problem As such, we apply this controller to speed control system of
in diverse power systems. Good controlling effort is BLDC motor where the snake optimizer (SO) along with a
achieved by the above controllers although the systems suitable cost function is employed to help in finding the
tested are governed by several nonlinearities. As a result, optimal controller parameters.
by the simple and successful idea of combining the SO is one of the nature-inspired metaheuristic algo-
advantages of different P/I/D combinations, it is possible to rithms that have been recently proposed to handle different
obtain increased dynamic performance. This is the main optimization problems in a better way [36]. As the name
intention and contribution of the current article. implies, the algorithm mimics the special mating concept
It is noticeable from the above discussion that the of snakes. Each snake (male or female) endeavors to
mechanism of evolutionary computation has been engage with the best partner if the available food is suffi-
employed to design effective control techniques using cient and the ambient temperature is low enough. By
metaheuristic algorithms, which have been of interest to modeling such forage and reproduction strategies

123
Neural Computing and Applications

2 3 2 32 3 2 3 2 3
mathematically with simple and straightforward opera- va Ra 0 0 ia Ls 0 0 i
4 vb 5 ¼ 4 0 d a
tions, SO could achieve a good capacity in exploration– Ra 0 54 ib 5 þ 4 0 Ls 0 5 4 ib 5
exploitation balance and convergence speed on various dt
vc 0
2 30 Ra ic 0 0 Ls ic
search spaces. This has been supported by the numerical ea
results and statistical comparisons obtained in several þ 4 eb 5
problems such as LFC problem [37], thermal error forecast ec
model of motorized spindle [38], power-aware task ð1Þ
scheduler [39], feature selection [40], etc.
In the present research, a new c-1PDf-PI controller is where va, vb, vc are phase voltages, ia, ib, ic are phase
applied to speed control in BLDC motor and we show the currents, ea, ea, ec are induced emfs, Ra is stator winding
interesting features of this controller with respect to speed resistance, and Ls is synchronous inductance.
reference change and external torque disturbances. In lieu For the electromagnetic torque Te, we have:
of trial-and-error approach, SO, as the recent and potent p ea ia þ eb ib þ ec ic
Te ¼ ð2Þ
optimization algorithm, has been adopted to find out the 2 xe
best combination of controller parameters. It is believed
where p is the number of poles and xe is the electrical rotor
that the improved performance and simplicity of the
speed in rad s-1. Finally, we can arrive the rotor speed
c-1PDf-PI controller render it attractive for research com-
equation in state-space form as:
munity and commercial BLDC motor drives. Eventually,
   
results of simulation based on C ? ? IDE and real-time d p 2
xr ¼ Te  TL  B xr =J ð3Þ
experiments using DSP of TI TMS320F28335 are dt 2 p
demonstrated to verify the performance of the proposed
control scheme. Moreover, comparisons with PI and 1PDf where J [ 0 and B [ 0 are inertia and viscous friction
speed controllers are also presented. coefficient of the whole rotating system, whereas TL is the
The workflow of this article is as follows: Sect. 2 pre- applied load torque.
sents the drive system and mathematical modeling of In BLDC motor, only two phases are conducted and the
BLDC motor. The c-1PDf-PI controller structure, review of remaining phase is kept open at any instant. Considering
the SO mechanism and SO-based controller design are the emf and corresponding phase current are in phase, the
introduced in Sect. 3. Section 4 is devoted to various total electromagnetic torque becomes equal to combination
comparative results as to reveal the performance of our of those obtained from the two phases. Let E and I be the
proposal. Finally, some concluding marks and future work magnitude of emf and phase current, respectively, Eq. (2)
are given in Sect. 5. is simplified and rewritten as:
EI
Te ¼ p ð4Þ
xe
2 Brushless DC motor model and its drive It is noticeable from Eq. (4) that as long as E and I are
system constant, i.e., free of harmonics, the electromagnetic torque
Te maintains constant inherently.
In a BLDC motor drive system, one way is to regulate
motor currents individually. This needs three current sen-
sors to be placed in the ac-link to sense each of the phase
3 Proposed speed control system
currents. Given the structure of this kind of drive system in
Fig. 1a, rectangular current references shown in Fig. 1b
3.1 c-1PDf-PI controller
with 120 duration are generated as a function of rotor
electrical position he and current reference I*, which is
Although PID controller is perceived as outdated, it con-
obtained from the output of the c-1PDf-PI controller. Based
stitutes a good alternative for effective control system
on the comparison between current references ia;b;c and
when improved by some structural variations and modifi-
measured currents ia;b;c , hysteresis current controllers cations. In this respect, the literature has witnessed enor-
generate the switch signals directly. So, stator currents are mous volume of research work to fulfill challenges in
regulated so that they can strictly track their references different systems, as discussed previously. However, to our
within a predetermined hysteresis band. best knowledge, c-1PDf-PI controller has not been
The stator voltage equation of a BLDC motor can be observed yet in the literature. Thereby, in the present study,
written as: it is recommended to regulate velocity in BLDC motor as
an effective, vigorous and simple controller. The

123
 Neural Computing and Applications

Fig. 1 Typical schematics for BLDC motor a drive system b current and emf waveforms

configuration of this controller is depicted in Fig. 2. As high-frequency noise is damped. The practical solution to
seen, the controller is composed of a 1PDf and PI controller the latter problem is to alter the form of the integral term
connected in cascade. That is, the output of 1PDf controller and use the equivalent expression instead. For this, PI
is y, which is the input of PI controller. After acting on the control law in s-domain is written first using PI block
speed error ex, the controller produces control input or diagram in Fig. 2:
current reference I*, which is mathematically given by y ðsÞ
0
yðsÞ
1 I  ðsÞ ¼ K p2 yðsÞ þ K i ð6Þ
zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{ s
B  
B ex Kd sf C C Ki e ðsÞK sf
I  ¼ Bex þ ex Kp1 þ C  Kp2 þ ð5Þ where yðsÞ ¼ ex ðsÞ þ ex ðsÞK p1 þ x sþf d as derived via
@ sþf A s
1PDf block diagram in Fig. 2. Substituting y(s) in Eq. (6),
we can easily arrive at Eq. (5), which relates I* to ex. The
where Kp1 and Kp2 are proportional gains, Ki is integral time-domain form of Eq. (6) is also written as:
gain, Kd is derivative gain, and f is the filter coefficient. Z t
Interestingly, this controller has superb performance, but I  ðtÞ ¼ K p2 yðtÞ þ K i yðtÞdt ð7Þ
0
it has some implementational problems, such as chattering
due to the noise and integral windup due to the iterated Taking the time derivative of both sides of Eq. (7) gives
accumulation overflow. The former issue can be resolved the following equation:
by putting a first-order filter on the derivative part and I  ðkÞ  I  ðk  1Þ ¼ K p2 ½ yðkÞ  yðk  1Þ þ Ki TyðkÞ ð8Þ
tuning the respective gains, i.e., Kd and f, such in a way that
where (k) and (k - 1) are, respectively, the current and
previous samples for the given variable, and T is the
sampling time greater than zero. Equation (8) can be
rewritten as:
DI  ðkÞ ¼ K p2 DyðkÞ þ K i TyðkÞ ð9Þ
Notice that Eq. (9) results in the change in current ref-
erence and it does no longer have integration term, elimi-
nating the risk of excessive accumulation and unwanted
problem of windup. Eventually, to calculate the actual

Fig. 2 The configuration of the c-1PDf-PI speed controller

123
Neural Computing and Applications

current reference I*(k), DI*(k) needs to be added to the 3.2.1 Initialization

previous current reference DI*(k - 1) as:
I  ðkÞ ¼ DI  ðkÞ þ I  ðk  1Þ ð10Þ SO begins its search by generating a random population
from uniform distribution, similar to other metaheuristic
The parameters to be tuned for the proposed c-1PDf-PI algorithms. The random initialization is done by:
controller are 5, which are Kp1, Kp2, Ki, Kd and f. To ensure
Si ¼ Smin þ rand  ðSmax  Smin Þ ð11Þ
the best performance for the controller, its parameters are
optimized using SO through a more appropriate cost where Si is the position of the ith snake, rand is a random
function. number in the range [0, 1], Smin and Smax are the problem
constraints, standing for the minimum and maximum limits
3.2 Snake optimizer of decision variables, respectively.

As the name implies, SO is an optimizer recently proposed 3.2.2 Splitting the population into two equal groups:
to deal with optimization challenges alternatively [36]. The males and females
main inspiration source is special mating concept of
snakes. Mating between males and females occurs under In original SO, it is assumed that the population of snakes
some conditions. For example, the temperature should be is equally split into two groups: male group and female
low enough. Moreover, availability of food also affects the group. To do this, Eq. (12) is used.
mating process. If the conditions are satisfied, opponent N m  N=2
males will fight each other to be a king and accordingly ð12Þ
Nf ¼ N  Nm
impress females. If the female decides to mate, mating
takes place and the female begins to lay its eggs in a nest or where N is the total number of solutions, Nm and Nf denote
burrow. Once the eggs come out, the female leaves the the number of male and female individuals, respectively.
location. Fighting of snakes, their mating and egg-laying
are shown in Fig. 3. 3.2.3 Evaluation of each group based on temperature
The mathematical modeling of the identified behavior of and food quantity
snakes is described next.
The best individual in each group is evaluated and assigned
as fbest,m for the best male and fbest,f for the best female. The
food position ffood is also gained.

Fig. 3 Snakes in nature a snake

fight b mating process c egg-
laying

123
 Neural Computing and Applications

Using the iteration index t and maximal iteration num- the best individual, and c3 is a constant whose value is
ber T, the temperature is altered in an exponentially tuned to 2.
decreasing trend in the algorithm as: If Temp falls below 0.6 (treated as cold), snakes will
t fight or mate. In case they fight, the following position
Temp ¼ exp ð13Þ update rules are governed:
T
As concerned to food quantity, it is calculated by: Si;m ðt þ 1Þ ¼ Si;m ðtÞ þ c3  FM  rand

   Q  Sbest;f  Si;m ðtÞ ð20Þ
tT
Q ¼ c1  exp ð14Þ
T where Si,m is the ith male position, Sbest,f is the best position
in female group, and FM is the fighting ability of the male
where c1 is a constant equal to 0.5. individual. It is assumed that similar fight also takes place
among the female snakes and Eq. (20) is modified as in
3.2.4 Exploration phase with no food Eq. (21) for the individuals in female group.
 
If Q is smaller than a threshold value, i.e., 0.25, SO Si;f ðt þ 1Þ ¼ Si;f ðtÞ þ c3  FF  rand  Q  Sbest;m  Si;f ðtÞ
emphasizes on exploration where the snakes look for food ð21Þ
source by choosing a random solution from the population
and updating their position relative to it. This is governed where Si,f is the ith female position, Sbest,m is the best
by: position in male group, and FF is the fighting ability of the
female individual. FM and FF are calculated as:
Si;m ðt þ 1Þ ¼ Srand;m ðtÞ  c2  Am  
 ðSmin þ rand  ðSmax  Smin ÞÞ ð15Þ f best;f
FM ¼ exp ð22Þ
fi
where Si,m is the ith male position, Srand,m is the position of  
random male, c2 is a constant equal to 0.05 and Am is the f best;m
FF ¼ exp ð23Þ
male ability to discover food, which is calculated as: fi
!
f rand;m where fbest,f and fbest,m refer to the fitness of Sbest,f and
Am ¼ exp ð16Þ Sbest,m, respectively.
f i;m
In case the snakes are mating, the following position
where frand,m indicates the fitness of Srand,m, whereas the update rules are used:
fitness of Si,m is given by fi,m. Si;m ðt þ 1Þ ¼ Si;m ðtÞ þ c3  M m  rand

Similarly, Eqs. (15) and (16) are also applied to the  Q  Si;f ðtÞ  Si;m ðtÞ ð24Þ
group of females as in Eqs. (17) and (18).  
Si;f ðt þ 1Þ ¼ Si;f ðtÞ þ c3  M f  rand  Q  Si;m ðtÞ  Si;f ðtÞ
Si;f ðt þ 1Þ ¼ Srand;f ðtÞ  c2  Af
 ðSmin þ rand  ðSmax  Smin ÞÞ ð17Þ ð25Þ
! where Si,m is the ith male position, Si,f is the ith female
f rand;f
Af ¼ exp ð18Þ position, Mm and Mf are the mating ability of male and
f i;f female, respectively. They are calculated by:
!
where Si,f is the ith female position, Srand,f is the position of f i;f
random female, Af is the female ability to discover food, M m ¼ exp ð26Þ
f i;m
frand,f is the fitness of Srand,f, and fi,f is the fitness of Si,f. !
f i;m
3.2.5 Exploitation phase with food M f ¼ exp ð27Þ
f i;f

If Q is greater than the threshold value and Temp is higher If egg hatches, then the worst male and female indi-
than a threshold value (0.6 treated as hot), snakes will viduals are replaced as follows:
locate themselves around the food using Eq. (19).
  Sworst;m ¼ Smin þ rand  ðSmax  Smin Þ ð28Þ
Si;j ðt þ 1Þ ¼ Sfood  c3  Temp  rand  Sfood  Si;j ðtÞ
Sworst;f ¼ Smin þ rand  ðSmax  Smin Þ ð29Þ
ð19Þ
where Sworst,m and Sworst,f refer to the position of individ-
where Si,j is the position of the ith snake in the jth uals in male group and female group, respectively. The
dimension (either male of female), Sfood is the position of operator ± used in the above equations is named as flag

123
Neural Computing and Applications

direction operator or diversity factor, which adds to or solved by fourth-order Runge–Kutta technique and trape-
subtracts from solutions’ positions to alter the direction of zoidal method is used to numerically calculate the integral
search agents so that the given search space is sampled and operations needed in the controller. The source codes of SO
scanned thoroughly in different directions. To better are also written in another MATLAB/m-file script and the
understand the concept of SO, its pseudocode description is two scripts are linked with each other in order to solve our
provided in Algorithm 1. problem for acquiring optimal parameters of the c-1PDf-PI
controller.
Algorithm 1 Pseudocode of SO

In the original paper [36], SO has been benchmarked There are five parameters required for the design of c-
against several popular and newly developed algorithms 1PDf-PI controller, which are Kp1, Kp2, Ki, Kd and f. At the
such as L-SHADE, LSHADE-EpSin, coyote optimization beginning of the algorithm, a number of snakes (search
algorithm, moth-flame optimization, Harris hawks opti- agents) are generated randomly, each of which has there-
mizer, grasshopper optimization algorithm and whale fore different controller parameters. The search agents are
optimization algorithm on 29 CEC 2017 benchmark func- sent to the drive system script to simulate and obtain speed
tions and four practical engineering problems. The com- response from the controller. Then, a fitness value is
parison results affirm the performance of SO in terms of computed for each search agent by employing a more
solution accuracy and convergence rate—the interested appropriate cost function as:
readers are recommended to kindly refer to reference [36]. Z ts Z ts
2
As such, replicating already reported results would not only J ¼a es dt þ b  jes jdt þ c  I rt ð30Þ
0 0
be redundant but also repetitive within the context of this
article. where es is speed error between the reference and actual
speed, ts is simulation time in sec, a&b&c are the weight
3.3 SO-based controller design factors deciding the significance of each term in Eq. (30).
To find a suitable set of these weight factors, we first cal-
In the present research, the BLDC motor, equivalent drive culate each term in Eq. (30), revealing that the first term
system and c-1PDf-PI controller are modeled in a has a value far greater than those of the second and the
MATLAB/m-file script. The differential equations are third term in order. This allows the algorithm to pay more

123
 Neural Computing and Applications

attention to minimizing the squared error (first term) rather run 20 times and the results obtained over 20 independent
than the others. To have a balanced optimization process runs are analyzed statistically. In this analysis, the best,
wherein every contribution in Eq. (30) is influential and worst, average and standard deviation of FOD values cal-
considered sufficiently by the algorithm, coefficient of the culated for c-1PDf-PI, PI and 1PDf controllers are obtained
term with high value should be small, whereas that with and tabulated in Table 1. Note that all these results are
small value needs to be increased. Given this, setting found when T = 100 and N = 30 in the algorithm. As seen
a = 0.01, b = 1 and c = 5 results in good results consid- in Table 1, the best FOD value is attained with c-1PDf-PI
ering the minimization of both es and Irt simultaneously. controller (FOD = 148.2831) compared to PI (FOD =
Here, Irt is the ripple rate in current reference and calcu- 148.5584) and 1PDf (FOD = 51.1239) controllers. Of
lated by: course, the smaller the value of FOD, the better the per-
I max  I min formance we achieve. Table 1 also states that c-1PDf-PI is
I rt ¼ ð31Þ better than its cousins in terms of other statistical metrics.
I avg
Overall, considering the standard deviation values which
where Imax, Imin and Iavg are the maximum, minimum and are all closer to zero, one can conclude that SO yields
average current reference measured over a time period. robust performance for the studied controllers in speed
Notice that the first two terms in Eq. (30) measure and aim regulation of BLDC motor.
at reducing the speed error in order to have an augmented The best c-1PDf-PI, PI and 1PDf controller parameters
speed response profile. As concerned to the third term, it offered by SO and the associated FOD values are given in
helps increase the controller robustness to noise come up Table 2. Critical observation of Table 2 points out that Kp2
with the derivative term in the controller. It is worth is far smaller than Kp1 because it is multiplied by Dy in
explaining that during the cost function evaluation, the Eq. (9). To ensure DI* in Eq. (9) to accommodate new
speed reference x and motor load torque TL are assumed speed reference hastily, Kp2 should be small. Kd and f are
to change from 0 to 3500 rpm and from 0 to 3 Nm in steps also seen to be small so that the derivative will not amplify
of 500 rpm and 0.5 Nm, respectively (the maximum limits the high-frequency noise, but smooths it out sufficiently.
are selected based on the motor nameplate). So, there are As concerned to other controller parameters, they are
54 different combinations of these two variables and the optimized simultaneously by SO in a way that the FOD
final fitness pertaining to one search agent is calculated as value is minimized.
the mean of these 54 costs using: Twenty convergence profiles of SO during the opti-
! mization of the c-1PDf-PI controller parameters, i.e., when
X
NC
FOD ¼ J i =NC ð32Þ minimizing FOD, are depicted in Fig. 4. As seen, at the
i¼1 start of the optimization, the value of FOD is large and SO
tries to minimize it by a set number of iterations using the
where NC denotes the number of different cases equal to 54
search concept of its own. It is noticeable from Fig. 4 that
and FOD is the final fitness, termed as figure of demerit.
each curve stars from a different point because SO is ini-
After the overall performance of the search agents that are
tialized randomly at the beginning of each run. After a
evaluated by Eq. (32), they are fed back to the optimization
dramatic decrease, SO stabilizes and converges around the
script to evolve based on the SO concept for the next
global optimum, which takes between 20 and 40 iterations
iterations. This process of switching between the drive
only. Thus, this justifies our selection of maximum itera-
system script and the optimization script continues unless
tion number as T = 100. When the algorithm stops at
t [ T.
T = 100, the c-1PDf-PI controller parameters yielding the
In order to avoid the impact of stochasticity and show
minimum value of FOD, i.e., 148.2831 are assumed as
that the improvements do not take place by chance, SO is
optimal and used in simulation and experiments.

Table 1 Statistical minimization results of SO for different

controllers Table 2 Controller parameters and their FOD values offered by SO
Best Worst Average Standard deviation Kp1 Kp2 Ki Kd f FOD

c-1PDf-PI 148.2831 148.3001 148.2899 2.01E - 10 c-1PDf-PI 5.676 0.0541 4.735 0.3522 0.0117 148.2831
PI 148.5584 148.9514 148.7512 1.08E - 08 PI – 0.1635 7.0311 – – 148.5584
1PDf 151.1239 152.1215 151.5898 3.08E - 06 1PDf 2.8082 – – 0.0175 0.0702 151.1239

123
Neural Computing and Applications

Table 3 Nominal parameters of the motor model Sanyo Denki

P50B08100D
Variable Value

Rated power 1000 W

Rated armature current 6.7 Arms
Rated revolution speed 3000 rpm
Rated torque 3.185 Nm
Phase armature resistance (R/ ) 0.41 X
Electrical time constant (te ) 5.9 ms
Induced voltage constant (ke ) 19.3 V/krpm
Inertia (including wiring-saved INC) (J) 2.651 9 10-4 kg m2
Fig. 4 Some convergence curves of SO during the minimization of Mechanical time constant (tm ) 1.1 ms
FOD Number of poles (p) 4
Encoder resolution 2000 ppr
4 Simulation and experimental results

In this section, the performance of the speed control

scheme presented in this work is assessed. A comparison and parameter uncertainty. Note that all the results were
study is also established with respect to a PI and 1PDf acquired by keeping the controller parameters same as in
controller to see whether the cascade connection of them Table 2. Simulations performed for this research were
has better performance. The brushless AC motor model conducted in Dev-C ? ? IDE and the results were plotted
Sanyo Denki P50B08100D is the motor to be controlled. A using MATLAB R2017b, running on AMD Ryzen 9
hysteresis brake from Magtrol is coupled to BLDC motor 5900HX 3.3 Ghz CPU and 32 GB RAM.
shaft to apply external step torque disturbances. The writ-
ten motor control algorithm is implemented in DSP of TI 4.1 Speed regulation performance
TMS320F28335. In the system shown in Fig. 5, there is
also a built-in tachogenerator with a voltage constant of In this test, firstly speed regulation was realized when the
7 V/krpm, which is used to display speed responses in load torque is 1Nm and the speed reference signal is step
Tektronix DPO 2024B oscilloscope. set to x* = 2000 rpm. The resulting responses are pre-
The brushless AC motor has the nominal parameters as sented in Fig. 6. In the experimental results, the red and
in Table 3. turquoise traces indicate the desired speed and measured
In order to evaluate the credibility of the c-1PDf-PI speed, x*, x, respectively, with a speed scaling factor of
controller, the BLDC motor was tested using three condi- 714.3 rpm/div. It is clear that the c-1PDf-PI controller
tions such as speed regulation, torque disturbance rejection responds better because after a mild overshoot, the speed
settles to its reference hastily without unwanted oscilla-
tions. Response obtained with the 1PDf speed controller
does not have any overshoot, but it yields significant speed
error and speed ripple sustained in steady state. It is also
seen that the simulated responses are comparable with the
experimental results.
Secondly, the c-1PDf-PI controller is utilized to track a
reference trajectory. The reference trajectory selected for
this exercise is x* = 1700 9 (sin(t) ? cos(0.5t))rpm. The
load torque is in its rated value. Figure 7 depicts the tra-
jectory tracking for the reference trajectory for all the
controllers. As it can be seen in Fig. 7, all the controllers
are able to give more or less the same responses, where the
actual speed tracks the desired trajectory well. Nonetheless,
some steady state error and notable speed ripple are
observed in the case of the 1PDf controller.

Fig. 5 Experimental system

123
 Neural Computing and Applications

Fig. 6 Simulated and experimental responses to a step speed reference a comparison in simulation b c-1PDf-PI c PI d 1PDf

In the last exercise for the speed regulation test, the 4.2 Torque disturbance rejection
system is run under random step speed references. The
speed reference signal is configured as increasing up to 2 s. This test aims to check disturbance rejection capabilities of
Each increment occurs after every 0.4 s of time duration the controllers. For this, the load torque is assumed zero
with a step change of 500 rpm in speed reference. The initially while the motor rotates at 2500 rpm. Then, step
simulation and experimental responses under such a case load torque disturbance of 3.185Nm (rated torque) is
are portrayed in Fig. 8. It is evident from this figure that applied at t = 0.75 s and removed from the BLDC motor
when the system is controlled by the c-1PDf-PI controller, shaft at t = 1.5 s. For clarity, these instants are, respec-
better controlling effort is achieved where the motor speed tively, indicated by the marks ; and : on the experimental
is able to track its reference with less overshoot and no responses. When observing the resulting responses in
steady state error. It is also noted that important speed Fig. 9, one can notice that although torque disturbances
ripple and steady state error still persist with the 1PDf seem to be rejected fast by the 1PDf controller, it is not
controller. capable to offer acceptable steady state response, where the

123
Neural Computing and Applications

Fig. 7 Simulated and experimental tracking performance of controllers for the given reference trajectory a comparison in simulation b c-1PDf-PI
c PI d 1PDf

actual speed could not settle to the reference again after the 4.3 Robustness to variation in motor parameters
load application. The PI controller eliminates the steady
state error effectively, but it causes large deviations and Finally, robustness of the c-1PDf-PI controller is tested
longer settling time for the response to recover. On the when there is a change in the induced voltage constant (ke),
other hand, response with the c-1PDf-PI controller does not phase armature resistance (R/) and mechanical time con-
have steady state error and is recovered quickly following stant (tm) in the presence of load torque disturbances.
torque disturbances. Here, it is noteworthy to emphasize Carrying out experimental work for this test is impossible
that some speed error and considerable speed ripple occur because there is no any direct methodology to alter these
in the experimental response with 1PDf controller. parameters in practice. However, we believe that the con-
sidered motor parameters tend to change by temperature
increase during operation in above experimentations, thus
already proving the robustness of our proposal experi-
mentally. Here, ke, which is related to the motor torque

123
 Neural Computing and Applications

(b) (c)

(d)
Fig. 8 Simulated and experimental responses to random step speed references a comparison in simulation b c-1PDf-PI c PI d 1PDf

constant also, is assumed to change in the range of ± 40% test, it is affirmed that c-1PDf-PI controller performs
and R/ in the range of ± 75%. To obtain different robustly to uncertainty in the motor parameters.
mechanical time constants, we consider the friction coef- Our simulation and experimental results in previous
ficient B very close to zero, and change of inertia J three sections clearly revealed that c-1PDf-PI controller could
times greater/lower than the nominal inertia. Setting the provide better system performance by combining the dis-
step speed command to x* = 2000 rpm and introducing tinct features and advantages of the compared approaches.
some torque disturbances of ?/- 2Nm, the resulting step While it provides performance betterment, its time com-
responses under parameter changes and torque disturbance plexity is unknown, which is important from the view point
are depicted in Fig. 10. Critical observation of Fig. 10 of practical implementation. To address this aspect, time
points out that speed regulation is preserved in each con- consumption of each control scheme is measured in DSP
dition and load changes are rejected satisfactorily. It is also and visualized in Fig. 11. Expectedly, minimum time is
found that phase resistance is the parameter that has the required by the PI controller, which is followed by the
least impact on the time-domain characteristics of the 1PDf and c-1PDf-PI controller, respectively. It may be
response comparing with the others. As the result of this noted that the time consumption increases notably in the

123
Neural Computing and Applications

Fig. 9 Simulated and experimental disturbance rejection capacity under sudden load variation a comparison in simulation b c-1PDf-PI c PI
d 1PDf

1PDf implementation, attributable to the calculation of 5 Conclusion

derivative. Further, although the c-1PDf-PI controller
combines the two controllers in cascade, its time require- In this assessment, a c-1PDf-PI controller was presented to
ment differs only by a slight increase comparing with that handle the speed regulation issue of BLDC motors effec-
of 1PDf controller. As a result of this analysis, it is eluci- tively. The proposed controller brings together and
dated that that our proposal is capable of offering improved involves the distinct features of 1PDf controller and PI
system performance without compromising on time com- controller by connecting them in cascade. To attain the best
plexity much. performance from this controller, SO was adopted and the
controller parameters were optimized through a reasonable
cost function. Performance acquired was tested via

123
 Neural Computing and Applications

Fig. 10 Step response and load disturbance rejection of the c-1PDf-PI controller under a change in a ke b R/ c tm

• The resulting controller provides a simple and efficient

control technique for adjustable speed in BLDC motor
drives.
• 1PDf controller allows to have a fast response and has
the ability to mitigate high-frequency noise that is
normally amplified by the derivative.
• PI controller plays an important role in eliminating
speed error in steady state. Thanks to the integration,
Fig. 11 Time-consuming of DSP implementation for each control which accumulates the speed errors, we have zero or
scheme
negligible steady error in the speed response profile.
• The equivalent form of PI control law used in this
simulation and experimental work where a comparison
research does not present integral windup problem
with the 1PDf and PI controllers was also presented. The
owing to excessive integral accumulation.
results witnessed that the c-1PDf-PI controller was able to
• The cascade connection of 1PDf and PI controllers
offer eminent reference tracking performance and torque
enhances the rejection capacity to disturbance before it
disturbance rejection with a superb robustness to parameter
propagates to other parts of the control system.
changes. In the following, the technical features and
• There are only five controller parameters whose values
effectiveness of our proposal were elaborated.
are acquired effectively by SO without the need for
human knowledge/interpretation.

123
Neural Computing and Applications

• In all conditions tested, satisfactory and robust speed 6. Li ZB, Lu W, Gao LF, Zhang JS (2021) Nonlinear state feedback
regulation performance is attained wherein the acquired control of chaos system of brushless DC motor. Procedia Com-
puter Sci 183:636–640
controller parameters need not be reset. This is 7. Vanchinathan K, Selvaganesan N (2021) Adaptive fractional
attributed to the way the optimization process is order PID controller tuning for brushless DC motor using artifi-
implemented in which various motor working condi- cial bee colony algorithm. Results Control Optim 4:100032
tions are taken into account. 8. Ubare P, Ingole D, Sonawane DN (2021) nonlinear model pre-
dictive control of BLDC motor with state estimation. IFAC-
Several search avenues could be recommended as future PapersOnLine 54(6):107–112
9. Umeno T, Hori Y (1991) Robust speed control of DC servomo-
research. Other metaheuristic algorithms instead of SO
tors using modern two-degree-of-freedom controller design.
may be tried out to minimize the cost function further and IEEE Trans Ind Electron 38(5):363–368
accordingly escalate the system performance. It is also 10. Mendoza-Mondragón F, Hernández-Guzmán VM, Rodrı́guez-
interesting to apply the c-1PDf-PI controller to other con- Reséndiz J (2018) Robust speed control of permanent magnet
synchronous motors using two-degrees-of-freedom control. IEEE
trol applications in lieu of speed servo system.
Trans Industr Electron 65(8):6099–6108
11. Çelik E, Dalcali A, Öztürk N, Canbaz R (2013) An adaptive PI
Acknowledgements Authors thank the Scientific Research Projects
controller schema based on fuzzy logic controller for speed
Coordination Unit of Düzce University for supporting this research
control of permanent magnet synchronous motors. In: 4th inter-
work partially by Grant 2022.06.03.1291.
national conference on power engineering, energy and electrical
drives. 715–720
Funding Open access funding provided by the Scientific and Tech-
12. Arivalahan R, Venkatesh S, Vinoth T (2022) An effective speed
nological Research Council of Türkiye (TÜBİTAK).
regulation of brushless DC motor using hybrid approach. Adv
Eng Softw 174:103321
Data availability The research data presented in the work can be made
13. Chen GY, Perng JW, Ma LS (2016) DSP based BLDC motor
available on reasonable request.
controller design with auto tuning PSO-PID algorithm.In: 2015
IEEE/SICE International symposium on system integration. SII
Declarations 766–770
14. Gadekar K, Joshi S, Mehta H (2020) Performance improvement
Conflict of interest No conflict of interest is declared in this work. in BLDC motor drive using self-runing PID controller. Pro-
ceedings of the 2nd international conference on inventive
Open Access This article is licensed under a Creative Commons research in computing applications ICIRCA 2020. 1162–1166
Attribution 4.0 International License, which permits use, sharing, 15. Guha D, Roy PK, Banerjee S (2016) Load frequency control of
adaptation, distribution and reproduction in any medium or format, as interconnected power system using grey wolf optimization.
long as you give appropriate credit to the original author(s) and the Swarm Evolut Comput 27:97–115
source, provide a link to the Creative Commons licence, and indicate 16. Sahu RK, Panda S, Padhan S (2015) A hybrid firefly algorithm
if changes were made. The images or other third party material in this and pattern search technique for automatic generation control of
article are included in the article’s Creative Commons licence, unless multi area power systems. Int J Electr Power Energy Syst
indicated otherwise in a credit line to the material. If material is not 64:9–23
included in the article’s Creative Commons licence and your intended 17. Guha D, Roy PK, Banerjee S (2016) Quasi-oppositional differ-
use is not permitted by statutory regulation or exceeds the permitted ential search algorithm applied to load frequency control. Eng Sci
use, you will need to obtain permission directly from the copyright Technol, an Int J 19(4):1635–1654
holder. To view a copy of this licence, visit http://creativecommons. 18. Oozeer MY, Ramjug-Ballgobin R (2019) A jaya-based invasive
org/licenses/by/4.0/. weed optimization technique for load frequency control, In: smart
and sustainable engineering for next generation applications—
Proceeding of the second international conference on emerging
trends in electrical, electronic and communications engineering
References (ELECOM 2018) 561 11–21
19. Ozdemir MT, Ozturk D (2017) Comparative performance anal-
1. Çelik E, Öztürk N (2018) Attenuating saturated-regulator oper- ysis of optimal PID parameters tuning based on the optics
ation effect of brushless DC motors through genetic-based fuzzy inspired optimization methods for automatic generation control.
logic estimator. Turk J Electr Eng Comput Sci 26(6):3207–3223 Energies 10(12):2134
2. Gan MG, Zhang M, Zheng CY, Chen J (2018) An adaptive 20. Çelik E (2020) Improved stochastic fractal search algorithm and
sliding mode observer over wide speed range for sensorless modified cost function for automatic generation control of
control of a brushless DC motor. Control Eng Pract 77:52–62 interconnected electric power systems. Eng Appl Artif Intell
3. Jegajothi B, Geethamahalakshmi G, Raja A, Mahendran N (2022) 88:103407
An efficient metaheuristic optimization based fuzzy controller for 21. Yılmaz ZY, Bal G, Çelik E, Öztürk N, Güvenç U, Arya Y (2021)
brushless DC drives lifetime expansion. Mater Today: Proc A new objective function design for optimization of secondary
56(6):3343–3348 controllers in load frequency control. J Fac Eng Archit of Gazi
4. Premkumar K, Manikandan BV (2015) Speed control of brush- Univ 36(4):2053–2067
less DC motor using bat algorithm optimized adaptive neuro- 22. Guha D, Roy PK, Banerjee S (2018) Symbiotic organism search
fuzzy inference system. Appl Soft Comput 32:403–419 algorithm applied to load frequency control of multi-area power
5. Zhang R, Gao L (2022) The brushless DC motor control system system. Energy Syst 9(2):439–468
based on neural network fuzzy PID control of power electronics 23. Manoharan N, Dash SS, Rajesh KS, Panda S (2017) Automatic
technology. Optik 271:169879 generation control by hybrid invasive weed optimization and

123
 Neural Computing and Applications

pattern search tuned 2-DOF PID controller. Int J Comput Com- 33. Arya Y, Dahiya P, Çelik E, Sharma G, Gözde H, Nasiruddin I
mun & Control 12(4):533–549 (2021) AGC performance amelioration in multi-area intercon-
24. Rahman A, Saikia LC, Sinha N (2016) AGC of dish-stirling solar nected thermal and thermal-hydro-gas power systems using a
thermal integrated thermal system with biogeography based novel controller. Eng Sci Technol, Int J 24(2):384–396
optimised three degree of freedom PID controller. IET Renew 34. Çelik E (2021) Design of new fractional order PI–fractional order
Power Gener 10(8):1161–1170 PD cascade controller through dragonfly search algorithm for
25. Satapathy P, Debnath MK, Mohanty PK (2018) Design of PD- advanced load frequency control of power systems. Soft Comput
PID controller with double derivative filter for frequency regu- 25:1193–1217
lation. In: 2nd IEEE international conference on power elec- 35. Çelik E, Öztürk N (2022) Novel fuzzy 1PD-TI controller for
tronics, intelligent control and energy systems (ICPEICES) AGC of interconnected electric power systems with renewable
1142–1147 power generation and energy storage devices. Eng Sci Tech, Int J
26. Çelik E, Öztürk N, Arya Y, Ocak C (2021) (1?PD)-PID cascade 35:101166
controller design for performance betterment of load frequency 36. Hashim FA, Hussien AG (2022) Snake optimizer: a novel meta-
control in diverse electric power systems. Neural Comput Appl heuristic optimization algorithm. Knowl-Based Syst 242:108320
33(22):15433–15456 37. Rameshar V, Sharma G, Bokoro PN, Çelik E (2023) Frequency
27. Wolpert DH, Macready WG (1997) No free lunch theorems for support studies of a diesel–wind generation system using snake
optimization. IEEE Trans Evol Comput 1:67–82 optimizer-oriented PID with UC and RFB. Energies 16:3417
28. Öztürk N, Çelik E (2012) Speed control of permanent magnet 38. Dai Y, Pang J, Li Z, Li W, Wang Q, Li S (2022) Modeling of
synchronous motors using fuzzy controller based on genetic thermal error electric spindle based on KELM ameliorated by
algorithms. Int J Electr Power Energy Syst 43(1):889–898 snake optimization. Case Studies in Thermal Engineering
29. Öztürk N, Çelik E (2014) An educational tool for the genetic 40:102504
algorithm-based fuzzy logic controller of a permanent magnet 39. Yousri R, Elbayoumi M, Soltan A, Darweesh MS (2023) A
synchronous motor drive. Int J Electr Eng & Educ 51(3):218–231 power-aware task scheduler for energy harvesting-based wear-
30. Çelik E, Öztürk N (2017) Optimal setting of PI parameters for able biomedical systems using snake optimizer. Analog Integr
direct current motor drives by symbiotic organisms search algo- Circuits Signal Proc 115(2):183–194
rithm. J Inf Techol 10(3):311–318 40. Al-Shourbaji I, Kachare PH, Alshathri S, Duraibi S, Elnaim B,
31. Çelik E, Gör H (2019) Enhanced speed control of a DC servo Elaziz MA (2022) An efficient parallel reptile search algorithm
system using PI ? DF controller tuned by stochastic fractal and snake optimizer approach for feature selection. Mathematics
search technique. J Franklin Inst 356(3):1333–1359 10(13):2351
32. Shayeghi H, Younesi A, Hashemi Y (2015) Optimal design of a
robust discrete parallel FP ? FI ? FD controller for the auto- Publisher’s Note Springer Nature remains neutral with regard to
matic voltage regulator system. Int J Electr Power Energy Syst jurisdictional claims in published maps and institutional affiliations.
67:66–75

123