A Machine-Learning Approach for Prognosis of Oscillating Water
Column Wave Generators
IZASKUN GARRIDO, JON LECUBE, FARES MZOUGHI, PAYAM ABOUTALEBI, IRFAN AHMAD,
SALVADOR CAYUELA, AITOR GARRIDO
Automatic Control Group, ACG, Department of Automatic Control and Systems Engineering, University of the
Basque Country—UPV/EHU, Bilbao, SPAIN
Abstract: Wave excitations cause structural vibrations on the Oscillating Water Columns (OWC) lowering the power generated
and reducing the life expectancy. The problem of generator deterioration has been considered for the Mutriku MOWC plant
and a machine learning-based approach for prognosis and fault characterization has been proposed. In particular, the use of k-
Nearest Neighbor (kNN) models for predicting the time to failure of OWC generators has been proposed. The analysis is based
on data collected from sensors that measure various operational parameters of the turbines. The results demonstrate that the
proposed kNN model is an excellent choice for reducing maintenance costs by enabling maintenance scheduling months in
advance. The model's high accuracy in predicting generator failures allows for timely and cost-effective maintenance,
preventing costly breakdowns and improving turbine efficiency. The results highlight the potential of machine learning-based
approaches for addressing maintenance challenges in the energy sector and underscore the importance of proactive maintenance
strategies in reducing operational costs and maximizing energy production.
Keywords: Machine learning, oscillating water column, wave energy.
Received: May 19, 2022. Revised: July 17, 2023. Accepted: August 29, 2023. Published: September 19, 2023.
1. Introduction
Based on data from global energy forecast, it is projected
that the demand for energy will witness a significant surge of
4.6% in 2030, primarily due to climate change and the growth
of emerging and developing economies [1]. Consequently, the
global energy market is shifting its focus towards sustainable
energy sources to cater for the basic energy requirements.
Despite the availability of multiple renewable energy options,
ocean energy, and wave in particular, have observed a
substantial increase in their adoption in the last decade, as
depicted in Figure 2. In line with these environmentally
conscious policies, several studies have been conducted on
ocean energy resources, such as [2-3].
As per the energy roadmap, Europe is under the obligation to
establish a marine energy infrastructure capable of meeting
roughly 10% of its energy consumption through wave and tidal
energy by 2050 [4]. In the course of this development, Wave
Energy Converters (WEC) have acquired significant importance
[5]. In particular, by 2050, it is expected that 337 GW will be
harnessed from the oceans throughout the world, and the
technology needed will be developed by then [6]. In case of wave
energy, it will be possible to generate 16 PWh of energy per year.
Approximately 50% of the expected energy by 2040 would be
achieved by means of wave energy.
In the case of Basque Country, the Mutriku Wave Power
Plant uses the Oscillating Water Column (OWC) principle to
generate electricity from
waves. This working
principle is quite simple.
It works as a result of
oscillation of the internal
water column within a
chamber, which has an
opening below the water
level. The incoming and
outgoing waves make the
internal water column
oscillate, and
consequently the air
within the chamber (see
Figure 1) is compressed
and decompressed.
Therefore, there are
pressure gradients across
the turbine. The turbines
deployed are
unidirectional, and in this particular article Well´s turbines. For
this reason, the generated bidirectional air flow passes through the
Figure 1: Capture chamber for an
OWCs in Mutriku MOWC
WSEAS TRANSACTIONS on COMPUTERS
DOI: 10.37394/23205.2023.22.15
Izaskun Garrido, Jon Lecube,
Fares Mzoughi, Payam Aboutalebi, Irfan Ahmad,
Salvador Cayuela, Aitor Garrido
E-ISSN: 2224-2872
120
Volume 22, 2023
unidirectional turbine, thus generating electricity (Garrido et al.,
2022). Only onshore devices, such as the multiple OWC Mutriku
Wave Power plant in the Basque Country, have shown consistent
power generation and can be classified as TRL 8 [7].
Effective monitoring and maintenance strategies are essential
for achieving high availability, capacity factor, and Annual
Energy Production (AEP) in power plants. Good maintenance
practices can help maintain steady operations, which has a strong
influence on reducing downtime and increasing availability,
power production, capacity factor, and AEP. Therefore, reducing
Operational and Maintenance (O&M) costs is a critical approach
to controlling the Levelized Cost of Energy (LCoE) [8].
To achieve optimal maintenance, scheduling adequate
maintenance frequency and implementing the best maintenance
strategy is crucial. Frequent maintenance can be costly, but
neglect can lead to higher failure rates and longer downtime. An
optimal maintenance system can reduce O&M costs by 11% to
18% [9]. Predictive maintenance is critical for identifying
potential failures before they occur, and analyzing data plays a
vital role in this regard. Collecting and analyzing data on turbo
generator performance can enable the development of predictive
models for scheduling maintenance proactively, reducing
downtime, minimizing repair costs, and improving operational
efficiency. In ocean industries that heavily rely on equipment,
such as manufacturing and transportation over a narrow time-
window, predictive maintenance is particularly important. By
analyzing data, valuable insights into equipment performance can
be obtained, enabling proactive measures to ensure optimal
offshore device operation and minimize the risk of unexpected
failures.
Maintenance strategies are classified into reactive, proactive,
and opportunistic categories based on the timing of maintenance
tasks. Reactive maintenance strategy, also known as corrective
maintenance, is a failure-based maintenance method that involves
performing maintenance only after a failure has occurred. This
strategy is efficient for small farms with high reliability, where
downtime-related maintenance operations are negligible and can
achieve high availability [10]. On the other hand, proactive
maintenance strategy is an approach that involves scheduling
inspections and replacements before the occurrence of failures to
avoid small faults from developing into major failures.
Preventive, condition-based and predictive maintenance are
examples of proactive maintenance strategies [11]. Opportunistic
maintenance strategy is the grouping of different planned
preventive and corrective maintenance actions with unplanned
preventive tasks that were meant for some worn-out components
in the future [12].
To develop and implement an adequate maintenance
strategy in onshore and offshore power plants, time-based and
sensor-based information is gathered. However, processing
this data is complicated due to the enormous amount of data
gathered and the number of variables measured. Feature
extraction is used to reduce redundant information and
dimensionality in many fields [13-14]. including maintenance.
Principal Component Analysis (PCA) is the most common
feature extraction algorithm, which extracts important
information from data and represents it as a set of new
orthogonal variables called principal components [15].
Another well-known feature extraction method is Linear
Discriminant Analysis (LDA), which involves finding the
projection hyperplane that minimizes the interclass variance
and maximizes the distance between the projected means of
the classes [16].
In Section II a comprehensive overview of the
manipulation and analysis of data from the Mutriku MOWC
turbo generators will presented. Initially, data is collected by
the PLCs using the data acquisition system, which must be
imported, formatted, and stored in appropriate files.
Subsequently, the data from each turbine is analyzed using
various group statistics, and different data sets may be merged.
The modified data can then be utilized in Section III to train a
kNN classification model that predicts the health status of the
turbo generator. The performance of the model will be
evaluated in Section IV and any necessary improvements and
future work will be presented in the Conclusions section that
ends the article.
Figure 2: Renewable electricity generation growth by technology by 2050
WSEAS TRANSACTIONS on COMPUTERS
DOI: 10.37394/23205.2023.22.15
Izaskun Garrido, Jon Lecube,
Fares Mzoughi, Payam Aboutalebi, Irfan Ahmad,
Salvador Cayuela, Aitor Garrido
121
Volume 22, 2023
2. Import PLC Data
In order to incorporate data from the output file of a
Programmable Logic Controller (PLC) into the programming
language of preference, a suitable method must be established.
This may be accomplished by defining distinct sets of tables
for each turbine, associated with specific time frames, thereby
facilitating the import of data from the PLC into tables. Each
column of the table represents a variable, allowing for a
straightforward analysis and interpretation of the data.
2.1 Bearing Analysis
In this section, a systematic approach is delineated to
cluster and preprocess data from a turbine, with the aim of
rendering it amenable for classification as exhibiting bearing
deterioration. To accomplish this, the data tables are
methodically refined through the elimination of any row that
contains an undefined or missing value, as well as those rows
or columns deemed extraneous for the purpose of the analysis.
The statistical analysis of each turbine on a specific day
can be determined by calculating the mean of the grouped
values to the first power for each generated pressure.
Subsequently, the resulting tables for different days can be
combined by joining only those pressure values that appear on
all tables. This process ensures that the analysis is consistent
and accurate across all the observed days.
As observed in Figures 3 and 4, there exists an optimal
operating point at approximately 6.5kW, which is
characterized by high output power and low levels of
vibration. The existence of this is optimal operating point is
further underlined when the vibrations are plotted against the
pressure, grouped by power output, as illustrated in Figure 5.
2.2 Resonance Analysis
Analogous to Section A, we present a systematic approach
for clustering and preprocessing data obtained from a turbine
with the goal of facilitating its classification as experiencing
resonance. The data tables are subjected to a methodical
refinement process involving the elimination of any row
containing undefined or missing values, as well as those rows
or columns deemed extraneous to the analysis. For each
turbine and day, the relevant statistics are computed in a
similar manner to the previous case. Additionally, the tables
obtained from different days are combined and subjected to a
joint analysis.
The study of the data presented in Figure 6 reveals that the
turbine subject to vibrations resulting from resonance is
comparatively less severely impacted when contrasted with
the turbine affected by bearing wear off. This finding is further
validated by the information presented in Figure 8, which
indicates that the turbine experiencing resonance generates a
higher production rate at the same pressure values as the other
turbines. These observations provide valuable insight into the
differential effects of distinct types of turbine vibration, and
Figure 3: Relationship between the power output (kW) and the
amplitude of the vibration (mmps)
Figure 4: Relationship between the power output (kW) and the
amplitude of the vibration (mmps)
Figure 6: Relationship between the power output (kW) and the
amplitude of the vibration (mmps)
Figure 5: Relationship between the vibration amplitude (mmps) and
the pressure across the turbine (daPa) grouped by power output (kW)
WSEAS TRANSACTIONS on COMPUTERS
DOI: 10.37394/23205.2023.22.15
Izaskun Garrido, Jon Lecube,
Fares Mzoughi, Payam Aboutalebi, Irfan Ahmad,
Salvador Cayuela, Aitor Garrido
122
Volume 22, 2023
underscore the importance of implementing targeted
maintenance and repair strategies that are tailored to the
specific nature and severity of the observed vibration
phenomena.
Additionally, it is evident from Figures 6 and 7 that there
exists an optimal operating point at approximately 15 kW.
This point is distinguished by its ability to produce high output
power while simultaneously minimizing levels of vibration.
This optimal operating point is further emphasized when the
vibrations are graphed against the pressure, categorized by
power output, as illustrated in Figure 8.
2. 3 Unbalance Analysis
Finally, we propose a comprehensive methodology for
clustering and preprocessing data collected from a turbine,
aimed at facilitating its classification as experiencing turbine
unbalance. To achieve this objective, the collected data
undergo a meticulous refinement process that entails the
elimination of any row containing undefined or missing
values, as well as the exclusion of those rows or columns
deemed irrelevant to the analysis. Subsequently, for each
turbine and day, pertinent statistics are computed in a manner
similar to the previous case. Furthermore, the tables obtained
from different days are combined and subjected to a joint
analysis. The proposed approach presents a systematic and
rigorous methodology for preprocessing and clustering
turbine data, with the ultimate goal of improving the accuracy
and reliability of turbine unbalance classification.
The analysis of the data presented in Figure 9 reveals that
the turbine vibrations caused by unbalance exhibit a more
pronounced linear relationship with the generated power when
compared to those vibrations resulting from resonance or
bearing wear off. This assertion is supported by the findings
presented in Figure 10, which show that the unbalance turbine
generates a superior production rate at the same pressure
values as the other turbines.
In the context of this specific failure, a definitive optimal
operating point for the turbo generator module is difficult to
ascertain from the information presented in Figures 9 and 10.
Morover, a more evident linear relationship between the
vibrations and the pressure, grouped by power output, can be
discerned from the data plotted in Figure 11. This
visualization highlights the complexity of the underlying
factors contributing to the failure, and underscores the
importance of employing comprehensive and multifaceted
analyses to diagnose and address such issues in a rigorous and
effective manner.
The aforementioned observations serve as compelling
evidence for the potentially significant impact of turbine
unbalance on the efficiency and productivity of the overall
system. These findings underscore the critical importance of
implementing timely and effective maintenance interventions
to mitigate this issue and minimize any adverse effects on the
system's performance. Such interventions may include the
implementation of regular monitoring and inspection
Figure 9: Relationship between the power output (kW) and the
pressure across the turbine (daPa)
Figure 7: Relationship between the power output (kW) and the
pressure across the turbine (daPa)
Figure 8: Relationship between the vibration amplitude (mmps) and
the pressure across the turbine (daPa) grouped by power output (kW)
Figure 10: Relationship between the power output (kW) and the
amplitude of the vibration (mmps)
WSEAS TRANSACTIONS on COMPUTERS
DOI: 10.37394/23205.2023.22.15
Izaskun Garrido, Jon Lecube,
Fares Mzoughi, Payam Aboutalebi, Irfan Ahmad,
Salvador Cayuela, Aitor Garrido
123
Volume 22, 2023
procedures, the incorporation of predictive maintenance
strategies, and the utilization of advanced diagnostic tools and
techniques to facilitate the early detection and remediation of
turbine unbalance issues.
3. Train a Model.
Supervised Learning
The findings of the preceding section serve as a motivation
to propose a classification model for the purpose of
prognostication, aimed at efficiently classifying the data
output derived from the Programmable Logic Controller
(PLC). The PLC generates various numerical statistics for
each turbine, and following the creation and importation of
diverse datasets in the preceding section, these datasets shall
be segregated into training and testing sets. Subsequently, a k-
nearest neighbor (kNN) algorithm will be employed to
construct a model capable of classifying the operational state
of the turbine based on a set of PCL data. The classification
model shall be trained using the training set and then
leveraged to make predictions for the testing set.
The k-Nearest Neighbors (kNN) is a supervised machine
learning technique that was initially introduced by Evelyn Fix
and Joseph Hodges in 1951 [17] and later expanded by
Thomas Cover [18]. In kNN classification, the input data
consists of the k closest training examples in a dataset. The
output is a class membership assigned to the object being
classified. The algorithm works by taking a plurality vote of
its neighbors, with the object being assigned to the class that
is most common among its k nearest neighbors. Since kNN
relies on distance for classification, it is important to
normalize the training data if the features come in vastly
different scales. This normalization can significantly improve
the accuracy of the algorithm.
Initially, the data undergoes the customary procedures of
cleansing and scaling as a primary step. Within this specific
system, absolute values will be adopted to consider the
pressure, as both pressure differentials induce a unidirectional
rotation. Moreover, it is assumed that a given pressure
differential will yield similar power generation by the turbine.
The distance metric that measures the similarity between
two data points in the feature space, is chosen to be the
Euclidean distance because it presents an excellent
performance in this case. Therefore, the Nearest Neighbors
model calculates the distance between the new turbine data
output and the data in the training set using the Euclidean
distance formula as follows:
󰇛 󰇜󰇛 󰇜󰇛 󰇜 (1)
where , and are the pressure, power and vibrations of
the new turbine, and , and are the pressure, power and
vibrations of the ith turbine in the training set.
Given a new turbine data, the kNN method has the
capability of classifying a new turbine data by associating it
with the most commonly occurring type among its k nearest
neighbors. This technique is rooted in the principle of
similarity, whereby the classification of a data point is based
on the identities of its closest neighbors as defined by equation
(1) in a high-dimensional space. Through this approach, the
kNN algorithm seeks to classify the new turbine data as
belonging to the same type as the turbines that have the highest
frequency of occurrence among its nearest k neighbors.
4. Simulation and Validation
Using the kNN method, we can calculate the distances to
each turbine in the training set and select the type of those
turbines with the shortest distance. Choosing the optimal value
of the hyperparameter k, the number of nearest neighbors to
be considered, is a critical aspect in the algorithm. Large
values tend to smooth out the decision boundary or prediction
surface, while small values ender the system more sensitive to
noise and overfitting. This value has been tuned to k=5 in
order achieve optimal performance on the validation set.
The Hold-out validation method has been used to estimate
the performance of the model, randomly partitioning the
available dataset into two subsets: a training set with 70% of
the data and a validation set with 30%. This technique has
been used because the available dataset has 21710464 entries,
so that it is large enough to support a random partition into
training and validation sets. This method allows for a quick
estimate of the performance, as the model may be trained only
once and then evaluated on the validation set.
The evaluation of a k-Nearest Neighbor (kNN) classifier's
accuracy involves determining the number of correct
predictions made and dividing that by the total number of
observations within the test set as follows
 (2)
where is a vector of predicted labels generated by the
classifier for the test set, is a vector of true labels for the
test set, and  represents the total number of labels
within the test set. Upon performing this evaluation, the
resulting accuracy score of the kNN classifier (2) was found
to be 0.9129. This score, which is indicative of the model's
effectiveness, can be deemed as excellent.
4.1 Validation Results
The validation has been carried out once the kNN model
has been designed and trained. A confusion chart, has been
used to evaluate the performance of a classification model. It
Figure 11: Relationship between the vibration amplitude (mmps) and
pressure across the turbine (daPa) grouped by power output (kW)
WSEAS TRANSACTIONS on COMPUTERS
DOI: 10.37394/23205.2023.22.15
Izaskun Garrido, Jon Lecube,
Fares Mzoughi, Payam Aboutalebi, Irfan Ahmad,
Salvador Cayuela, Aitor Garrido
124
Volume 22, 2023
shows the number of true positives (TP), true negatives (TN),
false positives (FP), and false negatives (FN) for each class in
a classification task.
In a confusion chart as seen in Figure 12, the rows
correspond to the true class labels, and the columns
correspond to the predicted class labels. Each cell in the table
represents the number of predictions that were classified as a
certain class. The diagonal cells represent the number of
instances that were correctly classified, while the off-diagonal
cells represent the number of instances that were
misclassified.
In this example, it may be read in Figure 13 that the model
correctly predicted correctly 10539388 instances for the
bearing problem, while misclassifying 583275 instances as
resonance and 179615 as unbalance. It also predicted correctly
8088515 instances for the resonance problem, while
misclassifying 868993 instances as bearing and 103003 as
unbalance. Finally, it predicted correctly 1192596 instances
for the unbalance problem, while misclassifying 104716
instances as bearing and 50363 as resonance. The confusion
chart produced by this kNN classification model reveals that
the model has achieved an accuracy of approximately 90%,
indicating that it has correctly predicted a substantial majority
of the test set labels. This performance is indicative of a
reliable model that may be useful for the intended application.
5. Conclusions
In this article, the authors have presented a study on the
development and evaluation of machine learning models for
prognosis and fault characterization of oscillating water
columns (OWCs) using Mutriku data. The data collection
involved the use of sensors to measure the mechanical and
aerodynamic properties of the entire OWC system. A kNN
model has been proposed for the replication of the OWC
system behavior and structural performance. The model has
been trained with appropriate parameters while adhering to a
low Mean Squared Error (MSE) target function. The efficacy
of the model has been successfully tested on a validation set
to ascertain its computational efficiency, validity, and
accuracy. The presented work has potential implications for
improving the prognosis and fault characterization of OWCs
through machine learning-based approaches.
The results of the evaluation indicate that the proposed
kNN model outperformed existing methods in accurately
predicting turbine failures, further underscoring its potential
for enhancing the prognosis and fault characterization of
OWCs.
Acknowledgment
This work was supported in part through grant IT1555-22 funded by the
Basque Government and through grants PID2021-123543OB-C21 and
PID2021-123543OB-C22 funded by
(MCIN/AEI/10.13039/501100011033/FEDER, UE). Margarita Salas
MARSA22/09 and María Zambrano MAZAM22/15 funded by (UPV-
EHU/MIU/Next Generation, EU), and through grant PIF20/299 (UPV/EHU).
References
[1] Chen, P., Wu, Y., Meng, J., He, P., Li, D., Coffman, D.,
Liang, X., Guan, D., 2022. The heterogeneous role of
energy policies in the energy transition of asia–pacific
emerging economies. Nature Energy, 1–9.
[2] Lekube, J., Garrido, A.J., Garrido, I., Otaola, E., Maseda,
J. Flow control in wells turbines for harnessing
maximum wave power
(2018). Sensors, 18 (2), 535. DOI: 10.3390/s18020535.
[3] Garrido, A.J., Garrido, I., Barambones, O., Alkorta, P.,
Maseda, F.J. Simple linear models for plasma control in
Tokamak Reactors International Conference on Control,
Automation and Systems, ICCAS 2008, 4694261, pp.
2429-2432. DOI: 10.1109/ICCAS.2008.4694261.
[4] Khojasteh, D., Lewis, M., Tavakoli, S., Farzadkhoo, M.,
Felder, S., Iglesias, G., Glamore, W., 2022. Sea level rise
will change estuarine tidal energy: A review. Renewable
and Sustainable Energy Reviews 156, 111855.
[5] Windt, C., Davidson, J., Faedo, N., Penalba, M.,
Ringwood, J. V., 2022. On the importance of high–
fidelity numerical modelling of ocean wave energy
converters under controlled conditions. Floating
Offshore Energy Devices: GREENER 20, 31.
[6] Uihlein, A., Magagna, D. (2016). Wave and tidal current
energy—A review of the current state of research beyond
technology. Renewable and Sustainable Energy
Reviews, vol. 58, pp. 1070-1081.
[7] Magagna, D., Ocean Energy Technology Development
Report 2018, EUR 29907 EN, European Commission,
Luxembourg, 2019, ISBN 978-92-76-12428-3,
doi:10.2760/158132, JRC118296.
[8] Ren, Zhengru, Amrit Shankar Verma, Ye Li, Julie JE
Teuwen, and Zhiyu Jiang. "Offshore wind turbine
operations and maintenance: A state-of-the-art review."
Renewable and Sustainable Energy Reviews 144 (2021):
110886.
[9] Zhu, Wenjin, Bruno Castanier, and Belgacem Bettayeb.
"A dynamic programming-based maintenance model of
offshore wind turbine considering logistic delay and
weather condition." Reliability Engineering & System
Safety 190 (2019): 106512.
[10] Karyotakis, A., and R. Bucknall. "Planned intervention
as a maintenance and repair strategy for offshore wind
turbines." Journal of marine engineering & technology 9,
no. 1 (2010): 27-35.
[11] Jiang Peng. Maintenance of wind turbine. Electr Equip
2011;28(6):68–71.
[12] Thomas, Édouard, Éric Levrat, and BenoÎt Iung.
"Overview on opportunistic maintenance." IFAC
Proceedings Volumes 41, no. 3 (2008): 245-250
[13] Sklansky, Jack. "Image segmentation and feature
extraction." IEEE Transac on Systems, Man, and
Cybernetics 8, no. 4 (1978): 237-247.
[14] Boonyakitanont, Poomipat, Apiwat Lek-Uthai,
Krisnachai Chomtho, and Jitkomut Songsiri. "A review
of feature extraction and performance evaluation in
epileptic seizure detection using EEG." Biomedical
Signal Processing and Control 57 (2020): 101702.
Figure 12: Relationship between the vibration amplitude (mmps) and
pressure across the turbine (daPa) grouped by power output (kW)
WSEAS TRANSACTIONS on COMPUTERS
DOI: 10.37394/23205.2023.22.15
Izaskun Garrido, Jon Lecube,
Fares Mzoughi, Payam Aboutalebi, Irfan Ahmad,
Salvador Cayuela, Aitor Garrido
125
Volume 22, 2023
[15] Hasan, Basna Mohammed Salih, and Adnan Mohsin
Abdulazeez. "A review of principal component analysis
algorithm for dimensionality reduction." Journal of Soft
Computing and Data Mining 2, no. 1 (2021): 20-30.
[16] Izenman, Alan Julian. "Linear discriminant analysis." In
Modern multivariate statistical techniques, pp. 237-280.
Springer, New York, NY, 2013
[17] Fix, Evelyn; Hodges, Joseph L. (1951). Discriminatory
Analysis. Nonparametric Discrimination: Consistency
Properties (PDF) (Report). USAF School of Aviation
Medicine, Randolph Field, Texas.
https://apps.dtic.mil/dtic/tr/fulltext/u2/a800276.pdf (ava.
March 2023)
[18] Cover, Thomas M.; Hart, Peter E. (1967). "Nearest
neighbor pattern classification" (PDF). IEEE
Transactions on Information Theory. 13 (1): 21–27.
doi:10.1109/TIT.1967.1053964.
Contribution of Individual Authors to the
Creation of a Scientific Article (Ghostwriting
Policy)
The authors equally contributed in the present
research, at all stages from the formulation of the
problem to the final findings and solution.
Sources of Funding for Research Presented in a
Scientific Article or Scientific Article Itself
Conflict of Interest
The authors have no conflicts of interest to declare
that are relevant to the content of this article.
Creative Commons Attribution License 4.0
(Attribution 4.0 International, CC BY 4.0)
This article is published under the terms of the
Creative Commons Attribution License 4.0
https://creativecommons.org/licenses/by/4.0/deed.en
_US
This work was supported in part through grant IT1555-22 funded by the
Basque Government and through grants PID2021-123543OB-C21 and
PID2021-123543OB-C22 funded by
(MCIN/AEI/10.13039/501100011033/FEDER, UE). Margarita Salas
MARSA22/09 and María Zambrano MAZAM22/15 funded by (UPV-
EHU/MIU/Next Generation, EU), and through grant PIF20/299 (UPV/EHU).
WSEAS TRANSACTIONS on COMPUTERS
DOI: 10.37394/23205.2023.22.15
Izaskun Garrido, Jon Lecube,
Fares Mzoughi, Payam Aboutalebi, Irfan Ahmad,
Salvador Cayuela, Aitor Garrido
126
Volume 22, 2023