Content uploaded by Merve Ertarğın
Author content
All content in this area was uploaded by Merve Ertarğın on Oct 29, 2023
Content may be subject to copyright.
Copyright © European Journal of Technique (EJT) ISSN 2536-5010 | e-ISSN 2536-5134 https://dergipark.org.tr/en/pub/ejt
European Journal of Technique
journal homepage: https://dergipark.org.tr/en/pub/ejt
A Deep Learning Approach for Motor Fault Detection
using Mobile Accelerometer Data
Merve Ertarğın1*, Turan Gürgenç2, Özal Yıldırım3,4 , Ahmet Orhan5
1*Munzur University, Electrical and Electronics Engineering Department, Tunceli, Turkey (e-mail: merveboydak@munzur.edu.tr).
2 Fırat University, Automotive Engineering Department, Elazığ, Turkey. (e-mail: tgurgenc@firat.edu.tr).
3 Fırat University, Software Engineering Department, Elazığ, Turkey. (e-mail: ozal@firat.edu.tr).
4 Fırat University, Artificial Intelligence and Data Engineering Department, Elazığ, Turkey. (e-mail: ozal@firat.edu.tr).
5 Fırat University, Electrical and Electronics Engineering Department, Elazığ, Turkey. (e-mail: aorhan@firat.edu.tr).
1. INTRODUCTION
Electrical machines are indispensable for industry, especially
in power generation, manufacturing and transportation. In these
applications, they play a critical role in the conversion and
control of energy and are essential to meeting the energy
demands of modern society. Due to occasional malfunctions in
electrical machines, they do not operate at high efficiency and
may consume more energy than necessary. This leads to higher
operating costs. Failures that occur can create safety hazards
for workers, such as electric shock or fire. These hazards can
cause injury, equipment damage, or even death. For these
reasons, it is necessary to detect faults before they progress and
to implement preventive maintenance and monitoring
programs.
Electric motors have complex internal structures and
mechanisms that make it difficult for humans to visually
inspect and identify faults. Many faults occur in motor parts or
electrical windings that are not easily accessible or visible
without disassembling the motor. Many motor failures present
as subtle changes in performance or behavior that are not
immediately noticeable to humans [1]. Identifying motor
failures often requires a deep understanding of motor operation,
performance characteristics and failure patterns. Some faults in
motor may occur intermittently or under certain operating
conditions. Detecting such errors in real time requires constant
monitoring of various parameters and being able to analyze
large volumes of data quickly. People can find it difficult to
constantly monitor motors at such high frequencies and to
analyze complex data patterns effectively. In contrast,
machine-learning models can overcome these challenges by
analyzing large amounts of data from engines, detecting fine
patterns, and identifying error signatures more accurately and
efficiently. These models process data in real time, providing
continuous monitoring and timely fault detection, increasing
overall motor reliability and minimizing downtime. Deep
learning methods are based on the use of raw input data, unlike
traditional approaches where it is necessary to manually extract
the properties of the input data. Thus, the need for expert
knowledge is minimized [2]. Due to these advantages, deep
learning models have been applied in many different fields
such as detecting brain abnormalities from magnetic resonance
images (MRI) [3], diagnosing heart diseases from
electrocardiography (ECG) signals [4, 5], face recognition [6],
speech recognition [7], as well as motor fault detection [8-16],
and successful results have been obtained.
The most preferred input data for detecting faults in motor
bearings are current [8-10] and vibration [11-16] data.
Vibration signals are very sensitive to the presence of bearing
ARTICLE INFO
ABSTRACT
Received: Aug., 01. 2023
Revised: Oct., 04. 2023
Accepted: Oct., 27. 2023
Electrical machines, which provide many conveniences in our daily life, may experience
malfunctions that may adversely affect their performance and the general functioning of the
industrial processes in which they are used. These failures often require maintenance or
repair work, which can be expensive and time consuming. Therefore, minimizing the risk
of malfunctions and failures and ensuring that these machines operate reliably and
efficiently play a critical role for the industry. In this study, a one-dimensional convolutional
neural network (1D-CNN) based fault diagnosis model is proposed for electric motor fault
detection. Motor vibration data was chosen as the input data of the 1D-CNN model. Motor
vibration data was obtained from a mobile application developed by using the three-axis
accelerometer of the mobile phone. Three-axis data (X-axis, Y-axis and Z-axis) were fed to
the model, both separately and together, to perform motor fault detection. The results
showed that even a single axis data provides error-free diagnostics. With this fault detection
method, which does not require any connection on or inside the motor, the fault condition
in an electric motor has been detected with high accuracy.
Keywords:
Motor Fault Diagnosis
Deep learning
1D-CNN
Corresponding author: Merve Ertarğın
ISSN: 2536-5010 | e-ISSN: 2536-5134
DOI:
Copyright © European Journal of Technique (EJT) ISSN 2536-5010 | e-ISSN 2536-5134 https://dergipark.org.tr/en/pub/ejt
defects or anomalies. Within the realm of deep learning
models, convolutional neural networks (CNN) excel at learning
features from mechanical vibration signals. As a result, many
studies have utilized CNNs for intelligent fault diagnosis of
machines [11-14].
Jia et al. [11] proposed an approach called deep normalized
convolutional neural network (DNCNN) to solve the problem
that CNNs do not take into account the unbalanced distribution
of machine health conditions. In this approach, normalized
layers based on weight normalization strategy and ReLU
activation function are used to improve the training process. A
weighted softmax loss has been developed to deal with the
unbalanced distribution data problem. In addition, a neuron
activation maximization (NAM) algorithm was developed to
understand how DNCNN learns features from vibrational
signals.
Machine learning models trained with data previously
collected from another machine may not perform satisfactorily
when the environment and operating conditions change on
different machine instances. Asutkar et al. [12] presented a
transfer-learning model to address this deficiency. With 1D-
CNN and transfer learning, it has been determined that the
accuracy rates are high even if datasets from different machines
are used in training and testing. Shen et al. [14] developed an
approach that embed the physical knowledge of bearing faults
into the model training process. Fault detection has been
successfully achieved with this deep learning approach, which
consists of a simple threshold model and CNN model for error
detection. In addition, generative adversarial networks (GAN)
[15], long-short-term-memory (LSTM) [16] models were also
used in motor fault diagnosis and motivating results were
obtained.
Various sensor equipment and platforms installed around
the motor are used to obtain vibration signals for motor fault
diagnosis [10]. These platforms are both costly and impractical
to use. In this study, motor vibration data were collected with a
non-invasive mobile application in order to evaluate motor
health with an easy method that does not require the use of
expensive sensors and minimizes the need for expert
knowledge. Today, the possibilities of smartphones, which are
available to almost everyone, are used in motor fault diagnosis
and the motor health status is evaluated without any cost. With
the CNN model, which is one of the deep learning methods and
has proven to be successful in diagnosis and classification in
many areas, motor fault diagnosis has been carried out without
error. Thus, a low-cost and practical method for the problem is
presented.
2. MATERYAL VE METOD
In this study, a mobile application has been developed to detect
motor failures from vibration data with 1D-CNN model.
Illustration of the flowchart to build proposed approach is given
in Figure 1. The phone, on which the mobile application was
installed, was placed on the motor and data acquisition was
performed in three axes (X, Y, Z). The data is segmented and
divided into train set, validation set and test set. The 1D-CNN
model was trained with the vibration data received, and then
the performance of the model was evaluated with the test data.
Segmentation
Vibrations signals
from smartphone
Deep 1D-CNN
Model
Healthy Faulty
Smart Phone
Train Set
Valid Set
Test Set
Figure 1. Illustration of the flowchart to build proposed approach.
2.1. Mobile Application
The mobile application used to get vibration data from the
electric motor was realized with Flutter based on Dart
language. Developed in 2011 by Google, Dart is defined as an
object programming language. Flutter, developed by Google,
makes it possible to develop applications for Android, iOS and
web through a single toolkit. The reason why Flutter
environment was preferred in this study is that Flutter enables
the development of applications for different operating systems
and devices through a single code base. The interface of the
mobile application is as shown in Figure 2. Vibration data in
the X-, Y- and Z-axes can be easily obtained by placing the
phone with the application installed on it on an electric motor,
opening the application screen and pressing the "Start
Recording" button shown in Figure 2 (a). After starting the
application, the application can show the vibrations in the X-,
Y- and Z-axes both graphically and numerically as shown in
Figure 2 (b). When the "Stop Recording" button is pressed, the
application stops receiving vibration data and saves the
received data in an excel spreadsheet. To delete the received
data from the excel table, press the "Clear Table" button. Thus,
the application becomes ready again to receive new data.
(a) (b)
Figure 2. Visual interfaces of the mobile application (a) Application opening
screen (b) When receiving real-time vibration data.
2.2. Proposed 1D-CNN Model
The CNN model proposed in this study is realized with an end-
to-end learning structure. With this model, which does not
require any feature extraction step, it is aimed to detect the
motor health status. Since the vibration signals are one-
dimensional, a 1D-CNN model is used.
The designed deep network model consists of 13 layers.
The model has 1D Convolution (Conv1D), MaxPooling
(MaxPool), flatten and dense layers. Figure 3 shows the
structure of the proposed model for electric motor fault
detection. Table I shows the parameters of the model in detail.
Copyright © European Journal of Technique (EJT) ISSN 2536-5010 | e-ISSN 2536-5134 https://dergipark.org.tr/en/pub/ejt
Conv1D (32,5)
Conv1D (64,3)
Conv1D (128,5)
Conv1D (256,3)
Conv1D (256,7)
Conv1D (32,3)
MaxPool (2)
MaxPool (2)
MaxPool (2)
MaxPool (2)
Vibration Signals
Flatten
Dense (128 Unit)
2, SoftMax
Healthy
Faulty
Figure 3. Architecture of proposed 1D-CNN model
TABLE I
DETAILED LAYERS AND PARAMETERS OF THE PROPOSED 1D-CNN MODEL
Layer
Layer Name
Kernel×Unit
Other Layer
Parameters
1
Conv1D
5×32
Activation = ReLu,
Strides = 1
2
MaxPooling1D
-
Strides = 2
3
Conv1D
3×64
Activation = ReLu,
Strides = 1
4
MaxPooling1D
-
Strides = 2
5
Conv1D
5×128
Activation = ReLu,
Strides = 1
6
MaxPooling1D
-
Strides = 2
7
Conv1D
3×256
Activation = ReLu,
Strides = 1
8
MaxPooling1D
-
Strides = 2
9
Conv1D
7×256
Activation = ReLu,
Strides = 1
10
Conv1D
3×32
Activation = ReLu,
Strides = 1
11
Flatten
-
-
12
Dense
1×128
ReLu
13
Dense
1×2
Softmax
Convolutional layers are the fundamental building blocks
of CNNs. Convolutional layers consist of filters that slide over
the input image, scanning for relevant patterns and features.
Pooling layers reduce the spatial dimensions of feature maps
while preserving important information. Flatten layer flattens
the feature maps into a 1D vector before transferring the data
to the dense layers. The dense layer, also known as the fully
connected layer, connects every neuron (or node) in the
previous layer to every neuron in the current layer, creating a
dense, fully connected network of neurons. In the last layer of
the network, the softmax layer is used to predict the class to
which the input signals belong. The optimizer selected was the
Adam optimizer, and loss function was selected as the binary
cross-entropy. After developing the model, the layer numbers,
types and parameters of the deep algorithm are changed by
brute force technique and the performance of the CNN model
are observed.
2.3. Dataset
A three-phase, two-pole, 50 Hz, 5.5 kW asynchronous motor
was selected for data acquisition. Firstly, data was obtained
from the faulty motor and then the motor was repaired and data
was obtained from the healthy motor in three axes (X, Y, Z).
At 40 Hz operating frequency, vibration data of 64000×3 (1280
seconds) from the faulty motor (F) and 64000×3 (1280
seconds) from the healthy motor (H) were taken. These data
were segmented in 500×3 dimension with 50 sample shifts.
Thus, 1270 samples were obtained from each of the H and F
classes, 2540 data samples in total. Then 80% of all data was
used for training, 10% for validation and 10% for testing.
Figure 4 shows the X-, Y- and Z-axes vibration signal
samples from the faulty and healthy motor. When the vibration
samples are analyzed, it is seen that the peak value of the
vibration amplitude of the defective motor is approximately 1.7
and the peak value of the vibration amplitude of the healthy
motor is approximately 0.9.
Faulty
0100 200 300 400 500
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8 YX Z
020 40 60 80 100
0.1
0.3
0.5
0.7
0.9
020 40 60 80 100
0.1
0.3
0.5
0.7
020 40 60 80 100
0
0.4
0.8
1.2
1.6
Samples
Amplitude
X-axis
Y-axis
Z-axis
Amplitude
Amplitude
Amplitude
Samples
Samples
Samples
(a)
Healthy
0100 200 300 400 500
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9 YX Z
020 40 60 80 100
0
0.2
0.4
0.6
020 40 60 80 100
0
0.2
0.4
0.6
0.8
020 40 60 80 100
0.1
0.3
0.5
0.7
0.9
Samples
Amplitude
Samples
Samples
Amplitude
Amplitude
Amplitude
Samples
X-axis
Y-axis
Z-axis
(b)
Figure 4. Vibration samples a) Faulty motor b) Healthy motor
In this study, the performance of the proposed CNN model
in motor fault detection is tested with four different cases:
Case 1: Motor fault detection using X-axis data.
Case 2: Motor fault detection using Y-axis data.
Case 3: Motor fault detection using Z-axis data.
Case 4: X-axis, Y-axis, Z-axis data were given to the
deep learning model as three different features and
motor fault detection was performed.
3. EXPERIMENTAL RESULTS
The 1D-CNN model was first trained on each axis data
separately to obtain loss and accuracy values. Figure 5 shows
the changes in the accuracy and loss values of the model over
10 epochs for the cases where X-axis, Y-axis and Z-axis data
are used, respectively. Looking at the performance graphs, it is
seen that the model does not have an overfitting problem.
Copyright © European Journal of Technique (EJT) ISSN 2536-5010 | e-ISSN 2536-5134 https://dergipark.org.tr/en/pub/ejt
1 2 3 4 5 6 7 8 9 10
0.82
0.84
0.86
0.88
0.9
0.92
0.94
0.96
0.98
1
1 2 3 4 5 6 7 8 9 10
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
Accuracy Value
Train
Validation
Epochs
Case 1
(X)
Loss Value
Train
Validation
Epochs
Case 1
(X)
(a)
1 2 3 4 5 6 7 8 9 10
0.82
0.84
0.86
0.88
0.9
0.92
0.94
0.96
0.98
1
Accuracy Value
Epochs
Case 2
(Y)
Train
Validation
1 2 3 4 5 6 7 8 9 10
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
Loss Value
Train
Validation
Epochs
Case 2
(Y)
(b)
1 2 3 4 5 6 7 8 9 10
0.85
0.9
0.95
1
Accuracy Value
Epochs
Case 3
(Z)
Train
Validation
1 2 3 4 5 6 7 8 9 10
0
0.05
0.1
0.15
0.2
0.25
0.3
Loss Value
Train
Validation
Epochs
Case 3
(Z)
(c)
Figure 5. Accuracy and loss graphs for each case a) X-axis, b) Y-axis, c) Z-
axis
The training performances of the model on each axis data
were quite successful. However, considering the motor fault
types, it was thought that providing all axis data to the model
input would provide an even superior performance. In line with
this idea, X-, Y- and Z-axes signals were combined and the
training performance of the model was observed. Figure 6
shows the performance graphs obtained by combining X-axis,
Y-axis and Z-axis signals and feeding them to the model input.
As can be seen in the graphs, combining X-axis, Y-axis and Z-
axis data provided similar performance in the performance
measures of the model.
1 2 3 4 5 6 7 8 9 10
0.86
0.88
0.9
0.92
0.94
0.96
0.98
1
Accuracy Value
Epochs
Case 4
(X,Y,Z)
Train
Validation
1 2 3 4 5 6 7 8 9 10
0
0.05
0.1
0.15
0.2
0.25
0.3
Loss Value
Train
Validation
Epochs
Case 4
(X,Y,Z)
Figure 6. Accuracy and loss graphs for X-axis, Y-axis, Z-axis together
Accuracy, the most widely used performance evaluation
metric, is used to evaluate the performance of the model. The
accuracy value is calculated as in Equation 1:
𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦 (%)= 𝑇𝑃 +𝑇𝑁
𝑇𝑃 +𝑇𝑁 +𝐹𝑃 +𝐹𝑁 ×100 (1)
In the equation, TP represents true positives and TN
represents true negatives. Similarly, FP represents false
positives and FN represents false negatives. Table Ⅱ shows the
validation accuracy values of the model at each epoch for the
cases generated. When these values are analyzed, it is seen that
the model quickly learns the motor fault condition.
TABLE Ⅱ
VALIDATION ACCURACY VALUES OF THE 1D-CNN MODEL AT EACH EPOCH
(%)
X-axis
Y-axis
Z-axis
X,Y,Z-axis
Epoch 1
0.9724
0.9606
0.9685
0.9724
Epoch 2
1.0
0.9881
0.9921
1.0
Epoch 3
1.0
1.0
1.0
1.0
Epoch 4
1.0
0.9960
1.0
1.0
Epoch 5
1.0
1.0
1.0
1.0
Epoch 6
1.0
1.0
1.0
1.0
Epoch 7
1.0
1.0
0.9960
1.0
Epoch 8
1.0
1.0
1.0
1.0
Epoch 9
1.0
1.0
1.0
1.0
Epoch 10
1.0
1.0
1.0
1.0
The trained model was run on 254 test data. It was observed
that the proposed model achieved 100% performance on the
test data in all cases.
4. DISCUSSION
In this study, a deep learning model is trained using data
obtained from a mobile platform to determine the motor fault
status. The biggest advantage of the study is that it enables fault
diagnosis only with the help of a smartphone without the need
for any external sensor connection. Thus, fault conditions can
be detected without the need for any platform installation inside
or around the motor. The 1D-CNN model used in the study
eliminates the need for any feature extraction step by providing
end-to-end learning. The 1D-CNN model trained on the data
obtained from the developed mobile application provided
100% accurate detection. In addition, fault recognition can be
achieved by using any of the X-, Y- and Z-axes for the motor
used.
In addition to its advantages, this study has several
limitations. First of all, a single motor dataset was used for the
study. Since the number of records in the dataset is limited, the
number of data was increased with the 50-sample sliding
window method. If more records are obtained, higher and more
reliable accuracy values can be achieved. A single motor type
was used in the study. The use of electric motors of different
power and types will be useful in evaluating the
generalizability of the proposed model.
In this study, only faulty and healthy motor diagnostics
were performed. No classification of the type of failure was
performed. The detection of different types of faults with the
vibration information received from the mobile phone will be
the subject of future studies.
5. CONCLUSION
In this study, motor vibration data is obtained from a mobile
application and the health status of a motor is evaluated with a
1D-CNN model. The accuracy of the proposed 1D-CNN model
is tested by first using the X-axis, Y-axis and Z-axis vibration
data from the mobile application individually and then feeding
these three axes data to the model simultaneously. In each case,
the 1D-CNN model, which does not require any feature
extraction and is easy to implement, performed an accurate
classification with 100% accuracy rate. With this study, an
Copyright © European Journal of Technique (EJT) ISSN 2536-5010 | e-ISSN 2536-5134 https://dergipark.org.tr/en/pub/ejt
experimental study is presented that the accelerometer sensors
in mobile phones are useful for evaluating the motor health
status, and that healthy and faulty motor states can be detected
without the need for any sensor or vibration meter.
REFERENCES
[1] D. Neupane and J. Seok, "Bearing Fault Detection and Diagnosis Using
Case Western Reserve University Dataset With Deep Learning
Approaches: A Review," in IEEE Access, vol. 8, pp. 93155-93178,
2020, doi: 10.1109/ACCESS.2020.2990528.
[2] S. Zhang, S. Zhang, B. Wang and T. G. Habetler, "Deep Learning
Algorithms for Bearing Fault Diagnostics—A Comprehensive Review,"
in IEEE Access, vol. 8, pp. 29857-29881, 2020, doi:
10.1109/ACCESS.2020.2972859.
[3] M. Talo, U. B. Baloglu, O Yildirim, and U. R. Acharya, “Application of
deep transfer learning for automated brain abnormality classification
using MR images,” in Cognitive Systems Research, vol. 54, pp. 176-188,
2019.
[4] U. B. Baloglu, M. Talo, O. Yildirim, R. S. Tan, and U. R. Acharya,
“Classification of myocardial infarction with multi-lead ECG signals
and deep CNN,” in Pattern recognition letters, vol. 122, pp. 23-30,
2019.
[5] O. Yildirim, P. Pławiak, R. S. Tan, and U. R. Acharya, “Arrhythmia
detection using deep convolutional neural network with long duration
ECG signals.,” in Computers in biology and medicine, vol. 102, pp. 411-
420, 2018.
[6] M. Coşkun, A. Uçar, O. Yildirim and Y. Demir, "Face recognition based
on convolutional neural network," 2017 International Conference on
Modern Electrical and Energy Systems (MEES), Kremenchuk, Ukraine,
2017, pp. 376-379, doi: 10.1109/MEES.2017.8248937.
[7] G. Hinton et al., "Deep Neural Networks for Acoustic Modeling in
Speech Recognition: The Shared Views of Four Research Groups," in
IEEE Signal Processing Magazine, vol. 29, no. 6, pp. 82-97, Nov. 2012,
doi: 10.1109/MSP.2012.2205597.
[8] T. Ince, S. Kiranyaz, L. Eren, M. Askar and M. Gabbouj, "Real-Time
Motor Fault Detection by 1-D Convolutional Neural Networks," in IEEE
Transactions on Industrial Electronics, vol. 63, no. 11, pp. 7067-7075,
Nov. 2016, doi: 10.1109/TIE.2016.2582729.
[9] Y. Li, Y. Wang, Y. Zhang, and J. Zhang, “Diagnosis of inter-turn short
circuit of permanent magnet synchronous motor based on deep learning
and small fault samples,.” in Neurocomputing, vol. 442, pp. 348-358,
2021.
[10] R. N. Toma, A. E. Prosvirin, and J. M. Kim, “Bearing fault diagnosis of
induction motors using a genetic algorithm and machine learning
classifiers,” in Sensors, vol. 20(7), 1884, 2020.
[11] F. Jia, Y. Lei, N. Lu, and S. Xing, “Deep normalized convolutional
neural network for imbalanced fault classification of machinery and its
understanding via visualization,” in Mechanical Systems and Signal
Processing, vol. 110, pp. 349-367, 2018.
[12] S. Asutkar, C. Chalke, K. Shivgan, and S. Tallur, “TinyML-enabled
edge implementation of transfer learning framework for domain
generalization in machine fault diagnosis,” in Expert Systems with
Applications, vol. 213, 119016, 2023.
[13] M. Ertargin , O. Yildirim and A. Orhan , "Motor Yataklarında Meydana
Gelen Arızaları Tespit Etmek için Yeni Bir Tek Boyutlu Konvolüsyonel
Sinir Ağı Modeli", Fırat Üniversitesi Mühendislik Bilimleri Dergisi, vol.
35, no. 2, pp. 669-678, Sep. 2023, doi:10.35234/fumbd.1292390.
[14] S. Shen, et al., “A physics-informed deep learning approach for bearing
fault detection,” in Engineering Applications of Artificial Intelligence,
vol. 103, 104295, 2021.
[15] K. Xu, X. Kong, Q. Wang, S. Yang, N. Huang, and J. Wang, “A bearing
fault diagnosis method without fault data in new working condition
combined dynamic model with deep learning,” in Advanced
Engineering Informatics, vol. 54, 101795, 2022.
[16] D. K. Soother, S. M. Ujjan, K. Dev, S. A. Khowaja, N. A. Bhatti, and T.
Hussain, “Towards soft real-time fault diagnosis for edge devices in
industrial IoT using deep domain adaptation training strategy,” in
Journal of Parallel and Distributed Computing, vol. 160, pp. 90-99,
2022.
BIOGRAPHIES
Merve Ertarğın obtained her BSc degree in Electrical and Electronics
Engineering from Firat University in 2016. In 2018 she joined the Electrical
and Electronics Engineering Department, Munzur University as a research
assistant. She received MSc. diploma in Electrical and Electronis Engineering
from Firat University in 2019. Currently, she continues her doctorate education
in the Department of Electrical and Electronics Engineering at Fırat University.
Her research interests are electrical machines, power electronics, machine fault
diagnosis and deep learning.
Turan Gürgenç received the MSc. diploma in Mechanical Engineering from
Erciyes University in 2013. He obtained his PhD diploma from the Mechanical
Engineering Department from Firat University in 2017. He is an Associate
Professor in Automotive Engineering Department at Firat University. His
research interests include surface coating, wear analysis, friction,
manufacturing, and machine learning.
Özal Yıldırım received the MSc. Diploma in Computer Engineering from Firat
University in 2010. He obtained his PhD diploma in Electrical and Electronics
Engineering from Firat University in 2015. He is currently working as an
Associate Professor in the Software Engineering Department at Firat
University. He has published over 60 articles in international peer-reviewed
journals and conference proceedings. As a result of the research carried out on
seven million researchers under the coordination of Stanford University, he
was included in the "World's Most Influential Scientists" list. His main research
interests include deep learning and medical signal and image processing.
Ahmet Orhan received the MSc. and PhD diploma in Electrical and
Electronics Engineering from Firat University in 1987 and 1999 respectively.
He is working as Associate Professor in Department of Electrical and
Electronics Engineering at Firat University. His current research interests
include synchronous motors, synchronous generators, induction motors, PM
motors and generators, matrix converters.
