sensors

Article
Anomaly Detection in a Smart Industrial Machinery Plant
Using IoT and Machine Learning
Angel Jaramillo-Alcazar , Jaime Govea and William Villegas-Ch *

Escuela de Ingeniería en Ciberseguridad, Facultad de Ingenierías Ciencias Aplicadas,

Universidad de Las Américas, Quito 170125, Ecuador; [email protected] (A.J.-A.);
[email protected] (J.G.)
* Correspondence: [email protected]; Tel.: +593-098-136-4068

Abstract: In an increasingly technology-driven world, the security of Internet-of-Things systems

has become a top priority. This article presents a study on the implementation of security solutions
in an innovative manufacturing plant using IoT and machine learning. The research was based
on collecting historical data from telemetry sensors, IoT cameras, and control devices in a smart
manufacturing plant. The data provided the basis for training machine learning models, which were
used for real-time anomaly detection. After training the machine learning models, we achieved a
13% improvement in the anomaly detection rate and a 3% decrease in the false positive rate. These
results significantly impacted plant efficiency and safety, with faster and more effective responses
seen to unusual events. The results showed that there was a significant impact on the efficiency
and safety of the smart manufacturing plant. Improved anomaly detection enabled faster and more
effective responses to unusual events, decreasing critical incidents and improving overall security.
Additionally, algorithm optimization and IoT infrastructure improved operational efficiency by
reducing unscheduled downtime and increasing resource utilization. This study highlights the
effectiveness of machine learning-based security solutions by comparing the results with those of
previous research on IoT security and anomaly detection in industrial environments. The adaptability
of these solutions makes them applicable in various industrial and commercial environments.

Citation: Jaramillo-Alcazar, A.; Keywords: IoT; machine learning; anomaly detection; operating efficiency
Govea, J.; Villegas-Ch, W. Anomaly
Detection in a Smart Industrial
Machinery Plant Using IoT and
Machine Learning. Sensors 2023, 23, 1. Introduction
8286. https://doi.org/10.3390/
Smart manufacturing has entered suddenly into industrial use. This practice has the
s23198286
transformative potential to redefine the way manufacturing plants operate and deliver high-
Academic Editor: Rodrigo quality products to market. At the center of this revolution is the convergence of the Internet
Román-Castro of Things (IoT) and machine learning [1]. These two disruptive technologies have come
Received: 5 September 2023
together to drive efficiency and security in highly specialized production environments. In
Revised: 25 September 2023
this work, a smart manufacturing plant is considered a space where innovation is found at
Accepted: 2 October 2023 the intersection of industrial machinery, construction equipment, and industrial tools [2].
Published: 7 October 2023 IoT and machine learning are becoming vital tools for meeting critical challenges.
Robust construction equipment, tools, and machinery are produced in an intelligent
manufacturing plant. Every component of this plant plays a vital role in building infrastruc-
ture, industrial development, and improving productivity worldwide. This plant is not just
Copyright: © 2023 by the authors. an example of traditional manufacturing, but paradigmatic of what smart manufacturing
Licensee MDPI, Basel, Switzerland. represents today [3]. Every machine, piece of equipment, and tool are connected via IoT,
This article is an open access article
generating a constant stream of data that becomes valuable information. Sensors embedded
distributed under the terms and
in the machinery capture real-time data on temperature, pressure, speed, and other crucial
conditions of the Creative Commons
parameters for the production process [4]. Data flow through a secure network and reach
Attribution (CC BY) license (https://
machine learning systems capable of analyzing them, identifying patterns, and making
creativecommons.org/licenses/by/
informed decisions.
4.0/).

Sensors 2023, 23, 8286. https://doi.org/10.3390/s23198286 https://www.mdpi.com/journal/sensors

Sensors 2023, 23, 8286 2 of 22

However, the digital transformation of manufacturing is not just a story of efficiency

and profitability. With the power of IoT and machine learning come substantial challenges,
especially regarding security. While robust, the interconnection of devices and systems
also introduces significant risks. Cybersecurity becomes critical, and early detection of
anomalies in the IoT network is essential to maintaining plant security and ensuring that
production runs smoothly [5]. In this scenario, the research problem is apparent: How
can an intelligent manufacturing facility specializing in industrial machinery, construction
equipment, and industrial tools ensure the safety of its operations while optimizing pro-
duction efficiency? To address this challenge, in this paper, we focus on two key aspects:
anomaly detection and security in IoT.
Anomaly detection is essential to identifying any unwanted deviations in the produc-
tion process, ranging from machine temperature fluctuations to unusual tool performance
patterns. The early detection of these anomalies allows for fast and preventive responses,
avoiding unplanned interruptions and reducing maintenance costs [6]. On the other hand,
security in IoT is essential to protecting the integrity of the smart manufacturing plant.
The interconnection of devices and systems opens the door to possible cyberattacks and
vulnerabilities. A top priority is ensuring that production data are secure and that machines
do not fall victim to malicious intrusions.
This paper aims to explore how anomaly detection and IoT security can converge
in a smart manufacturing plant to address these crucial challenges. It examines how
combining advanced sensors, machine learning algorithms, and robust security measures
can improve production efficiency and keep the plant safe from external and internal threats.
To develop the method, the practical implementation of these solutions in an intelligent
manufacturing plant specialized in industrial machinery, construction equipment, and
industrial tools is explored [7]. This process identifies how sensors collect vital data in
real time, how machine learning algorithms analyze these data, and how IoT security
measures protect plant integrity. In addition, we present concrete results that illustrate the
impact of these solutions on plant efficiency and safety, as well as their relevance in today’s
industrial landscape.
The results of this study are promising and highlight the relevance of IoT and machine
learning implementation in an innovative manufacturing plant environment [8]. Real-
time anomaly detection has proven effective in the early identification of machinery and
equipment problems, significantly reducing unplanned downtime and maintenance costs.
In addition, IoT security has ensured the integrity of the network and protection against
potential threats, which has strengthened the reliability of the plant. The relevance of
this study lies in its ability to address critical challenges in the modern manufacturing
industry [9]. Smart manufacturing is a growing trend, and this case study offers valuable
insight into how advanced technologies can improve efficiency and safety in this context.
In an increasingly technology-driven world, the security of IoT systems has become
a top priority. This article presents a comprehensive study on implementing security so-
lutions in an innovative manufacturing plant by leveraging IoT technology and machine
learning. Its key contributions include proposing a comprehensive framework for imple-
menting security solutions in industrial environments using IoT and machine learning
technologies. For this, a detailed analysis of the proposed resolution’s impact on the smart
manufacturing plant’s operational efficiency and safety is carried out. To evaluate its
effectiveness, a comparison of several machine learning algorithms, including support
vector machine (SVM) and convolutional neural network (CNN), is developed to evaluate
their effectiveness in detecting anomalies in real time.
The article is structured as follows: In the Section 2, we describe how the data were
collected and prepared, as well as the model training process. The Section 2.2 section
addresses practical implementation and critical stages. Next, in the Section 3 section, we
present the results obtained and perform a detailed analysis. The Section 4 focuses on the
research’s limitations, makes comparisons with the literature, and examines the relevance
of our study. Finally, in the Section 5 we summarize the main contributions of this research
Sensors 2023, 23, 8286 3 of 22

and highlight the importance of our work in IoT security and machine learning in industrial
environments.

2. Materials and Methods

Method development uses a practical and systematic approach in order to implement
and execute IoT security and anomaly detection solutions in intelligent manufacturing
plants. For this, an exhaustive review of previous works on security in IoT systems and
detecting anomalies in similar environments is carried out, identifying the best practices
and research gaps. Then, the selection and configuration of the machine learning algorithms
used for anomaly detection are presented, followed by a detailed description of the security
measures implemented in IoT devices and the network [10]. This method is critical to
understanding how the key steps that improved efficiency and safety in the intelligent
manufacturing facility were carried out.

2.1. Review of Similar Works

Security in IoT systems and anomaly detection in similar environments have been crit-
ical research topics in the last decade. Numerous works and projects have been developed
in this context to shed light on the challenges and solutions in these crucial areas [11]. One
of the fundamental aspects of security in IoT systems is protecting data and devices against
cyber threats. Previous research, such as the study by [12], has highlighted the importance
of device authentication and data encryption in IoT environments to guaranteeing the
confidentiality and integrity of the information transmitted. These approaches are based
on the premise that security starts at the most basic level of the IoT infrastructure, and that
any vulnerability could put the entire system at risk.
Another critical aspect is the detection of anomalies in real time in IoT environments.
The study [13] addressed failure detection in industrial machinery by analyzing time
series of sensor data. This research showed how machine learning algorithms, such as
recurrent neural networks, can identify unnatural patterns in the operation of industrial
machinery and prevent unplanned downtime. Furthermore, the study by [14] focused on
intrusion detection in IoT systems through an abnormal behavior analysis-based approach.
This approach makes it possible to identify unauthorized or unusual activities in the IoT
network, which is crucial to ensuring device security and data integrity [15].
Despite the advances outlined above, there is a substantial gap in research in terms
of the context of a specialized smart manufacturing plant for industrial machinery. This
gap lies in the lack of solutions and approaches that are optimally suited to the unique
characteristics of this highly technical environment [16]. Most previous work has addressed
general IoT applications or focused on specific domains like healthcare or home security.
The lack of thorough research on smart manufacturing plants operating industrial machin-
ery has left a gap in terms of understanding the challenges of and optimal solutions for this
context [17,18].
This research gap highlights the critical need for this case study. The smart manufac-
turing plant addressed here represents a highly specialized and challenging scenario where
safety and efficiency are paramount [19]. Previous works on IoT security and anomaly
detection laid the theoretical and methodological foundations. Thus, this case study dives
into the concrete application of these solutions in a real industrial context, addressing the
specific challenges faced by this smart manufacturing plant [20].

2.2. Concepts Used

Anomaly detection is a critical component in managing IoT systems and encompasses
several key concepts:
• Anomaly: An anomaly refers to any unusual deviation or deviation from an expected
or normal pattern in the data. In IoT systems, these anomalies can manifest as unusual
fluctuations in sensor measurements, atypical device behavior, or unexpected ways in
data traffic [21].
Sensors 2023, 23, 8286 4 of 22

• Machine learning for anomaly detection: This phrase refers to the use of machine
learning algorithms, such as support vector machines, random forests, or neural
networks, to identify abnormal patterns in data collected using IoT sensors. These
algorithms learn from historical data and can automatically detect anomalies without
the need for predefined rules.
• False positives and false negatives: In anomaly detection, false positives are cases
where the system identifies an anomaly that is not an anomaly, while false negatives
are cases where the system does not detect a genuine issue. Balancing the rate of
false positives and false negatives is essential to guaranteeing the effectiveness of an
abnormality detection system.
Security in IoT devices and networks involves several fundamental concepts that
protect the integrity and confidentiality of data and devices:
• Encryption: Encryption protects the confidentiality of data transmitted via an IoT
network [22]. The data are converted into an unreadable format for unauthorized
persons, and only those with the proper decryption key can access the information.
• Authentication: Authentication ensures that IoT devices only communicate with
authorized systems. Devices must prove their identity before accessing the network
or exchanging data.
• Access Control: Access control establishes who has permission to perform specific
actions on the IoT network. This ensures that only authorized people and devices can
access specific resources and features.
• Network Security: Network security protects the IoT communications infrastructure
against cyber threats. This includes the use of firewalls, intrusion detection systems,
and network segmentation to prevent unauthorized access.
• Key Management: Key management involves the generation, secure storage, and peri-
odic rotation of encryption keys used in the IoT network [23]. Proper key management
is essential to prevent security breaches.
These concepts and their interpretation are crucial to designing and applying effec-
tive anomaly detection and security solutions in IoT systems, especially in a specialized
environment such as an intelligent manufacturing plant.

2.3. Plant Description

The smart manufacturing facility specializes in producing high-quality industrial
machinery, construction equipment, and industrial tools. Located in a highly automated
environment, this plant is a leading example of smart manufacturing, where efficiency
and quality are paramount. The industrial machinery produced covers a wide range of
equipment, with output ranging from heavy machinery used in construction to precision
tools used in the manufacturing industry. The plant operates 24/7, running production
across multiple assembly lines and manufacturing processes. This highly specialized
environment presents unique challenges related to security and efficiency.
Every machine and device in the plant is equipped with sensors that collect real-time
data on parameters such as temperature, pressure, speed, and other aspects crucial to the
performance and quality of the final product. The IoT infrastructure in the plant is an
essential component of its operation. It includes a wide variety of sensors and devices
that continuously collect data. Sensors embedded in machinery and equipment capture
detailed information about their operation, while IoT devices on the network manage data
transmission and secure storage [24]. Data communication takes place over a secure IoT
network using encrypted communication protocols. Sensor data are transmitted in real time
to a central platform responsible for storing, processing, and analyzing the information.
This platform is the system’s core and is where the machine learning algorithms for anomaly
detection are applied.
The selection of machine learning algorithms is based on the need to detect anomalies
in real time in the data generated by the plant’s sensors. A combination of algorithms,
including convolutional neural networks (CNN) and hidden Markov models (HMM), is
 information. This platform is the system’s core and is where the machine learning algo-
rithms for anomaly detection are applied.
The selection of machine learning algorithms is based on the need to detect anomalies
Sensors 2023, 23, 8286 in real time in the data generated by the plant’s sensors. A combination of algorithms, 5 of 22

including convolutional neural networks (CNN) and hidden Markov models (HMM), is
chosen due to their ability to capture complex patterns and subtle changes in sensor data.
CNN models
chosen due to aretheirused to to
ability analyze
capturedata from sensors
complex patternsthat
and represent thermal
subtle changes and visual
in sensor data.
CNN
imagesmodels are used
of industrial to analyze
machinery data
[25]. Thisfrom
allowssensors that
for the represent
detection thermaland
of optical andthermal
visual
images
anomaliesof industrial machinery
in the operation [25]. ThisHMMs
of machines. allowsareforapplied
the detection
to timeofseries
optical and
data thermal
recording
anomalies in the operation
the performance of devicesofover
machines. HMMs
time. This are applied
combination to time series
of algorithms data recording
enables accurate,
the performance ofanomaly
multidimensional devices detection.
over time. This combination of algorithms enables accurate,
multidimensional anomaly detection.
A historical data set collected over a significant period at the plant is used to train the
A historical
machine learningdata set collected
models. These dataover a significant
include detailedperiod at thelogs
operating plantandis sensor
used tomeas-
train
the machine
urements. learning
Cases models.
of known These data
anomalies in theinclude detailed
data are taggedoperating
for use inlogsthe and sensor
supervised
measurements. Cases of
training of the models known
[26]. anomalies
The models arein the dataadjusted
carefully are tagged forthe
to fit useplant
in the supervised
environment
training of the models [26]. The models are carefully adjusted to fit the
and to minimize false positives and negatives. This involves parameter optimization andplant environment
and to minimize to
cross-validation false positives
ensure and negatives.
the accuracy This involves
and reliability parameter
of real-time optimization
anomaly and
detections.
cross-validation to ensure the accuracy and reliability of real-time anomaly
Machine learning implementation is rigorously carried out, and the trained models are detections.
Machine
integratedlearning
into theimplementation is rigorously
plant’s IoT infrastructure, carriedcontinuous
enabling out, and the traineddetection
anomaly models are in
integrated into the
the production plant’s IoTmachinery,
of industrial infrastructure, enabling equipment,
construction continuous anomaly detection
and industrial in the
tools.
production of industrial machinery, construction equipment, and industrial tools.
2.4. Method Design
2.4. Method Design
The method design addresses safety and efficiency challenges in a highly specialized
The method design addresses safety and efficiency challenges in a highly specialized
smart manufacturing facility for producing industrial machinery and other construction
smart manufacturing facility for producing industrial machinery and other construction
equipment. The primary purpose of this process is to strengthen the security in IoT sys-
equipment. The primary purpose of this process is to strengthen the security in IoT systems
tems that support the operation of the plant, identifying and mitigating possible cyber
that support the operation of the plant, identifying and mitigating possible cyber threats
threats and guaranteeing the integrity of the data [27]. In addition, it seeks to improve the
and guaranteeing the integrity of the data [27]. In addition, it seeks to improve the plant’s
plant’s efficiency by implementing anomaly detection solutions based on automatic learn-
efficiency by implementing anomaly detection solutions based on automatic learning
ing algorithms.
algorithms.
The research
The research follows
follows aamethodological
methodologicalapproach,
approach,combining
combiningqualitative
qualitativeand
andquantita-
quanti-
tative analysis elements [28]. This choice is based on the need to address both technical
tive analysis elements [28]. This choice is based on the need to address both technical and
and quantifiable
quantifiable aspects
aspects related
related to security
to security andand efficiency
efficiency in IoT
in IoT systems
systems andand
the the qualita-
qualitative
tive factors
factors that influence
that can can influence the implementation
the implementation of solutions
of solutions in a specialized
in a specialized manufac-
manufacturing
turing environment. Figure 1 shows the general flow of the methodology
environment. Figure 1 shows the general flow of the methodology used. used.

Figure 1. Flowchart illustrating the methodological process of investigation in the smart manufactur-
ing plant.

In the first phase, the analysis of the existing IoT infrastructure in the smart manufac-
turing plant is carried out. This includes identifying sensors, devices, and communication
protocols used in data collection. Subsequently, historical data covering a significant period
are collected [29]. These data are used to train and validate the machine learning models
Sensors 2023, 23, 8286 6 of 22

used in anomaly detection. Implementing security solutions in IoT systems is performed in

parallel with data collection, focusing on IoT network protection, device authentication,
and data encryption. Once the keys are implemented, they are evaluated to ensure their
effectiveness in detecting anomalies and improving plant safety. The results are analyzed,
and the solutions are adjusted as necessary.

2.4.1. IoT Infrastructure Analysis

IoT infrastructure in the smart manufacturing floor is a critical component of the pro-
cess, enabling real-time data collection and informed decision making. This infrastructure
is meticulously reviewed to address the plant’s specific needs and ensure efficiency in
production and security in IoT systems. In this way, it can be identified that each machine
and piece of equipment in the plant is equipped with telemetry sensors that monitor critical
parameters [30]. These sensors include temperature gauges, pressure, accelerometers, and
flow sensors. The data collected by these sensors provide real-time information on the
performance and condition of the machines.
For visual monitoring, IoT cameras are installed in critical areas of the plant. These
cameras capture high-resolution images that can be used to detect visual abnormalities and
assess the quality of manufactured products. In addition to monitoring sensors, the plant
has IoT control devices that allow for two-way communication with the machines. This
facilitates the ability to remotely adjust the settings and operation of machines.
The message queuing telemetry transport (MQTT) protocol efficiently transmits real-
time data from IoT sensors and devices. This protocol is highly efficient regarding band-
width and adapts well to resource-constrained environments. The constrained application
protocol (CoAP) is used for communication between IoT devices and the central manage-
ment platform [31]. CoAP is a lightweight protocol designed specifically for devices with
limited resources. Hypertext transfer protocol secure (HTTPS) is implemented in all data
transmissions to ensure communication security. This provides end-to-end authentication
and encryption. A custom-access control protocol has been implemented to ensure only
authorized devices can communicate with critical plant systems.

2.4.2. Historical Data Collection and Data Balancing

Historical data collection provides the basis for training and validating anomaly
detection models in the smart manufacturing plant. In this environment, historical data
were generated using sensors and IoT devices, allowing us to build a representative and
realistic data set for our research. Historical data were collected from telemetry sensors
that monitor various parameters, including temperature, pressure, acceleration, and flow
rates. These data were collected regularly, typically every second, over several months. We
also used IoT cameras to generate historical visual data, capturing images at scheduled
intervals and labeling them in order to identify typical situations and possible anomalies
in the plant. In addition to monitoring sensor readings, IoT control data were recorded to
document the actions taken on machines and equipment.
Regarding data balance, it is essential to highlight that techniques were implemented
to ensure that the data set used for training machine learning models was balanced in
terms of standard and abnormal data. This was essential to avoid bias in the anomaly
detection models. During the historical data collection process, we ensured that both
regular events and anomalies were captured and appropriately labeled. Additionally,
in order to effectively manage the massive volume of data, a scalable and redundant
storage system was implemented, thus ensuring the integrity and continuous availability
of the data. Data compression techniques were also applied to optimize the use of storage
resources without compromising data quality. These strategies ensured the quality and
balance of the data used in our study, which in turn contributed to the effectiveness of our
anomaly detection models and the robustness of our results.
The total volume of historical data collected was significant. Regarding the telemetry
sensors, approximately 15 terabytes of data were generated over six months. The images
Sensors 2023, 23, 8286 7 of 22

captured using the IoT cameras represented around 50,000 high-resolution images. Data
were collected and stored with a high degree of integrity and accuracy. Quality control
measures were implemented to identify and correct possible errors in data collection.
Properly and accurately labeled data enabled the practical training of machine learning
models. The collection of historical data provided a solid information base for research,
allowing us to advance the implementation of anomaly detection solutions based on
machine learning algorithms and strengthen security in IoT systems on the intelligent
manufacturing floor.
To effectively manage the 15 terabytes of data, a scalable and redundant storage
system was implemented to ensure the integrity and continuous availability of the data.
Additionally, data compression techniques were applied to optimize the use of storage
resources without compromising data quality.
Image data captured by IoT cameras were instrumental in our approach to IoT security.
These images were used for both the real-time diagnosis and training of ML models. In
real time, image analysis algorithms visually processed and evaluated situations on the
manufacturing floor, allowing for more accurate and faster anomaly detection. Images
were also integrated into ML models, significantly improving the algorithms’ ability to
understand and evaluate complex situations.
Image data are justified by the need for a richer and more detailed understanding
of situations on the manufacturing floor. These images provide visual information that
complements the numerical data collected using telemetry sensors. Visual information not
only helps to detect anomalies but also contributes to developing a better understanding
and evaluation of critical situations. In IoT security applications, visual information can
be essential to making informed decisions and responding effectively to unusual events.
Table 1 presents the summary of the values considered in the data collection.

Table 1. Parameters considered for the data collection phase.

Parameter Value
Telemetry sensors 50 sensors
Sampling rate 1 time per second
Data format CSV and JPEG image format
Historical data volume 15 terabytes (sensors)
Captured images 50,000 images (cameras)
Data quality High integrity and precision

2.4.3. Training Machine Learning Models

The machine learning model training process was based on extensive historical data
collected in the smart manufacturing plant. The data were carefully divided into training
and test sets to ensure an accurate evaluation of the models. A training data set consisting
of approximately 80% of the total volume of historical data was used. This represented
a data volume of 12 terabytes (sensors) and 40,000 images (cameras). The test data set
comprised 20% of the historical data, resulting in a data volume of 3 terabytes (sensors)
and 10,000 images (cameras). Training and test data sets were split temporarily to ensure
that models were trained on earlier data and evaluated on later data. This ensured that the
models were capable of detecting anomalies in future situations.
Data augmentation was applied to the camera images in order to improve the robust-
ness of the models and ensure that they were capable of handling diverse situations. This
included the use of random rotations, displacements, and scaling to generate a more varied
training data set. Regarding camera images, variables related to visual characteristics, such
as textures, colors, and patterns, were considered [32]. The data from telemetry sensors
included variables related to temperature, pressure, acceleration, and flow rates.
Parameters such as kernel size, learning rate, and number of convolutional layers
were adjusted for the CNN. In the support vector machines (SVM) algorithm with the
RBF kernel, the C and gamma parameters of the RBF kernel were adjusted. For the
Sensors 2023, 23, 8286 8 of 22

variational autoencoders (VAE), the number of units in the latent layers and the learning
rate were adjusted.
Since sensor data anomalies were relatively rare compared to everyday situations, class
balancing techniques were used during training. This technique ensures that the model is
not biased towards classifying everyday situations to the detriment of abnormalities. The
training process was carried out in a high-performance computing environment to handle
the large volume of data and the complexity of machine learning models [33]. Detailed
model evaluation results are presented in the results section of the article, along with
specific performance metrics.
Data augmentation to camera images is carried out through several techniques,
such as:
• Random rotation: Random rotations were applied to the images to different orienta-
tions and perspectives. This helped the models recognize anomalies from multiple
angles.
• Random shift: Images were randomly shifted in different directions in response to
variations in the location of objects. This improved the model’s ability to detect
anomalies at various positions.
• Random rescaling: Random rescaling was performed on the images, allowing the
model to identify anomalies in both large and small objects, regardless of their relative
size in the picture.
• Horizontal mirroring: By applying horizontal mirroring, mirror versions of the images
were created, helping models to recognize mirror anomalies or object inversions.
• Noise addition: Random noise was added to images of varying lighting and visibility
conditions in the manufacturing plant environment.
In addition to these techniques, other data augmentation methods were used to
diversify the training set. These data augmentation strategies ensured that the model was
trained in various situations and conditions, improving its ability to detect anomalies in
the smart manufacturing plant environment.

2.4.4. Implementation of Security Solutions in IoT

The implementation of security solutions in the IoT environment of the intelligent
manufacturing plant is carried out in order to strengthen protection against threats and
ensure the safe and reliable operation of the system [34]. This implementation is based on
pre-trained machine-learning models used for anomaly detection. Figure 2 outlines the
critical steps in the implementation process for IoT security solutions.
The first stage involves integrating the previously trained anomaly detection models
into the plant’s IoT infrastructure. This is possible when deploying a machine learning
model inference server that is an intermediary between the IoT devices and the models.
For this, a GPU-equipped high-performance server ensures efficient real-time inference.
Once the models are online, a continuous monitoring system is established to monitor all
incoming data from IoT sensors and cameras [35]. This system is based on a real-time
data processing architecture that allows for the ingestion and analysis of data as they are
generated. Technologies such as Apache Kafka and Apache Flink are used for development
to facilitate the transmission and processing of data in real time.
Anomaly detection is performed by constantly evaluating the data received by ma-
chine learning models. These models are designed to identify patterns and abnormal
behaviors in data from sensors and cameras. If an anomaly is specified, the system automat-
ically proceeds to generate alerts. The system creates automatic alerts when an abnormality
or unusual behavior is detected [36]. Signals are generated in real time and contain precise
details about the monster’s nature and exact plant location. These alerts are sent to plant
managers and security personnel through secure communication channels.
 Sensors 2023, 23, x FOR PEER REVIEW 9 of 22
Sensors 2023, 23, 8286 9 of 22

Figure 2. Implementation flow of machine learning-based security solutions in an IoT-enabled smart

Figure 2. Implementation
manufacturing flow of machine learning-based security solutions in an IoT-enabled smart
plant environment.
manufacturing plant environment.
The incident response protocol is activated as soon as alerts are received. Security
Anomaly
personnel detection
have access to a is performeddashboard
centralized by constantly evaluating
that provides the data
detailed received by
information ma-
about
chine
the learning
anomaly, models.relevant
including These models
data logs areand
designed
root-causeto identify
analysis.patterns and abnormal
This allows for a quick be-
haviors
and in data
accurate from sensors
assessment of theand cameras.
situation andIf an
theanomaly is specified,
implementation the system
of corrective automat-
measures
asically proceeds
necessary. A to generate alerts.
continuous The system
updating creates automatic
and improvement strategyalerts
was when an abnormal-
implemented to
ity or unusual
maintain behavior
the anomaly is detected
detection models’[36].effectiveness
Signals are generated
[37]. Newindata realare
time and contain
collected and
precisein
labeled details
orderabout
to train thethe
monster’s
models nature
regularly.and In exact plant location.
addition, These alerts are
the hyperparameters sent
of the
models
to plantare reviewed
managers andand adjusted
security to adaptthrough
personnel them tosecurethe changing conditions
communication of the plant
channels.
environment.
The incident In theresponse
last stage,protocol
an exhaustive evaluation
is activated as soonof the impactare
as alerts of the implemented
received. Security
security
personnel solutions is carried
have access to aout. Results regarding
centralized dashboard reduced security incidents,
that provides improved
detailed information
response
about the times, and increased
anomaly, includingoverall
relevant system reliability
data logs are measured.
and root-cause analysis. This allows for a
Implementing real-time security solutions is
quick and accurate assessment of the situation and the implementation crucial in smart manufacturing environ-
of corrective
ments. In this study, real-time anomaly detection was
measures as necessary. A continuous updating and improvement strategy was imple- achieved through continuously
monitoring
mented to and analyzing
maintain data fromdetection
the anomaly telemetrymodels’
sensors,effectiveness
IoT cameras, [37]. and control
New data devices in
are col-
the manufacturing
lected and labeled plant.in order Theto data
trainwere transmitted
the models regularly.and processed
In addition,inthe real time using the
hyperparameters
message queuing
of the models aretelemetry
reviewedtransport
and adjusted(MQTT) protocol,
to adapt themwhich
to theischanging
widely recognized
conditionsfor of its
the
efficiency and speed in smart industry applications.
plant environment. In the last stage, an exhaustive evaluation of the impact of the imple-
MQTT
mented has become
security solutions anisefficient andResults
carried out. reliableregarding
communicationreducedprotocol
security in the smart
incidents, im-
industry due to its ability to handle real-time data transmission
proved response times, and increased overall system reliability are measured. effectively. It allows for the
publication and transmission
Implementing real-timeofsecurity
messages between
solutions is IoT devices
crucial and manufacturing
in smart monitoring systems, fa-
environ-
cilitating
ments. In thethis
transmission of critical
study, real-time data quickly
anomaly and reliably.
detection In the context
was achieved through of this research,
continuously
MQTT
monitoring and analyzing data from telemetry sensors, IoT cameras, and controland
enabled agile communication between the telemetry sensors, IoT cameras, the
devices
machine learning-based anomaly detection system. This ensured
in the manufacturing plant. The data were transmitted and processed in real time using data were analyzed in
real time, leading to rapid responses to unusual events on
the message queuing telemetry transport (MQTT) protocol, which is widely recognized the manufacturing floor.
Theefficiency
for its efficiencyand of MQTT
speed in played
smarta industry
critical role in the anomaly detection system’s ability
applications.
to operate in real time, enabling faster and
MQTT has become an efficient and reliable communicationmore effective responsesprotocol
to strange inevents
the smartin thein-
smart manufacturing plant environment. This technology is an essential component in the
dustry due to its ability to handle real-time data transmission effectively. It allows for the
IoT communication infrastructure, supporting efficiency and security in the smart industry.
publication and transmission of messages between IoT devices and monitoring systems,
facilitating the transmission of critical data quickly and reliably. In the context of this
Sensors 2023, 23, 8286 10 of 22

2.4.5. Continuous Assessment

Continuous evaluation enables the constant improvement of security solutions imple-
mented in an IoT-enabled smart manufacturing plant environment. To carry out ongoing
evaluation, a systematic data collection process is established. Data were collected on
incidents, alerts generated, responses to incidents, and any corrective action taken. These
data were recorded and stored in a centralized database [38]. Specific performance metrics
were then defined, which served as critical indicators for evaluating the effectiveness of
security solutions.
• Anomaly detection rate: The proportion of correctly detected anomalies compared to
the total number of abnormalities in the system.
• False positive rate: The proportion of alerts generated that turned out to be false
alarms compared to the total signals generated.
• Average response time: The average time from detecting an anomaly to implementing
corrective measures.
• Incident reduction: The quantitative decrease in security incidents in the plant envi-
ronment compared to periods before the implementation of security solutions.
• Resource efficiency: The efficiency in using resources, including hardware resources
and security personnel.
A regular analysis cycle and periodic reports were established to evaluate the perfor-
mance metrics and results obtained. The collected data were used to generate detailed
reports in order to provide a complete overview of the security system’s performance.
Based on the results of the periodic reports, iterative adjustments and improvements
were made to the security solutions. These adjustments included the optimization of
detection algorithms, modifications to the alert configuration, or updates to security poli-
cies. Continuous evaluation also addressed the scalability of security solutions and the
ability of solutions to adapt to an increase in the number of IoT devices and data without
compromising their performance.
A regular update strategy was implemented in order to maintain the anomaly detec-
tion models’ effectiveness. The models were retrained using new data to ensure that they
kept up with the latest trends in the plant environment. The continuous improvement
process ensured that high security and reliability were always maintained.

2.4.6. Results and Analysis

This section outlines the results of implementing machine learning-based security
solutions in the IoT-enabled smart manufacturing plant environment. Additionally, the
methods used to obtain the results and the ways in which they are analyzed to assess the
effectiveness of the answers are described.
The overall results address security solutions’ impact in reducing incidents and im-
proving response times, indicating their overall effectiveness in detecting anomalies. The
results are obtained through data collection in real time. This is performed a significant
period after the implementation of the security solutions. Data on incidents, generated
alerts, response times, and other relevant indicators are recorded for this purpose. These
data are analyzed using statistical analysis and data visualization techniques.
The analysis of results should focus on evaluating the performance metrics defined
in the model design. Researchers should calculate anomaly detection rates and false
favorability rates, study trends over time, and compare the results with the reference data
in order to assess the impact of security before the implementation of solutions should be
considered.
In addition to performance metrics, it is necessary to collect specific examples of
incidents detected and resolved thanks to security solutions. These examples illustrate
real situations where the solutions proved their value and contributed to safety in the
plant. The results should support the effectiveness of machine learning-based security
solution implementation in the IoT-enabled smart manufacturing plant environment. The
 considered.
In addition to performance metrics, it is necessary to collect specific examples of in-
cidents detected and resolved thanks to security solutions. These examples illustrate real
situations where the solutions proved their value and contributed to safety in the plant.
Sensors 2023, 23, 8286 The results should support the effectiveness of machine learning-based security11solution
of 22

implementation in the IoT-enabled smart manufacturing plant environment. The reduc-

tion in incidents, the improvement in response times, and the high efficiency in detecting
reductionindicate
anomalies in incidents,
these the improvement
solutions’ inimpacts
positive responseon
times,
plantand the high efficiency in
safety.
detecting anomalies indicate these solutions’ positive impacts on plant safety.
2.4.7. Adjustments and Optimization
2.4.7. Adjustments and Optimization
The
Thetuning
tuningand
and optimization sectiondescribes
optimization section describes
thethe technical
technical and continuous
and continuous processprocess
of
of improving
improvingand andoptimizing
optimizing thethe security
security solutions
solutions implemented
implemented in the in the IoT-based
IoT-based intelli-
intelligent
gent manufacturing
manufacturing plant
plant environment.
environment. ThisThis process
process is based
is based on on real-time
real-time datadata collection,
collection,
performance
performancemonitoring,
monitoring,and and the constant
constantevaluation
evaluationofof critical
critical metrics.
metrics. Figure
Figure 3 shows
3 shows
thethe stages
stages consideredfor
considered for adjustment
adjustment andandoptimization.
optimization.

Figure 3. 3.
Figure IoT Security
IoT SecurityTuning
Tuning and
and Optimization Flow.
Optimization Flow.

AA constantdata
constant data collection
collection process
processisisestablished in real
established timetime
in real by implementing
by implementingsecurity
secu-
solutions. This process allows for the continuous acquisition of information about
rity solutions. This process allows for the continuous acquisition of information about in- incidents,
generated alerts, responses to incidents, and any changes in system configuration. It is
cidents, generated alerts, responses to incidents, and any changes in system configuration.
essential to implement a performance monitoring system in real time. This system actively
It is essential to implement a performance monitoring system in real time. This system
monitors the behavior of the security system and generates performance data in real time.
actively
Specificmonitors the behavior
performance of the
metrics such securitydetection
as anomaly system andrate, generates performance
false positive data in
rate, and other
real time.
vital Specificwere
indicators performance
defined formetrics such These
the process. as anomaly
metricsdetection rate, false
are constantly positive
evaluated overrate,
and other vital indicators were defined for
time to measure the security system’s performance.the process. These metrics are constantly eval-
uated over
One time
of thetocritical
measure the security
approaches system’s performance.
to optimization is tuning the hyperparameters of
One ofdetection
anomaly the critical approaches
models. Systematic to experiments
optimizationareisusually
tuningcarried
the hyperparameters
out to find the of
optimal combination of hyperparameters for maximizing detection
anomaly detection models. Systematic experiments are usually carried out precision and
to mini-
find the
mizing false positives. A periodic update strategy is implemented to ensure that anomaly
detection models are kept up to date. The models are retrained with new data collected
from the plant environment to ensure they are aware of the latest trends in abnormal
behavior. The adjustment and optimization process must be carried out continuously to
maintain high effectiveness and efficiency in IoT security solutions. These actions allowed
us to proactively address challenges that arose over time and ensured that the security
system remained adapted to the changing conditions of the plant environment.
Sensors 2023, 23, 8286 12 of 22

3. Results
The results obtained in this study represent a significant step towards improving
security in IoT systems in the context of a smart manufacturing plant. Through continuous
tuning and optimization, notable advances have been made in anomaly detection and
the protection of IoT infrastructure. This section will present and analyze the critical
data obtained during the research, highlighting how these results align with previously
established research objectives. In addition, the impact of these improvements on the
efficiency of the plant and the security of operations will be discussed, thus supporting the
relevance of this study in the security field in IoT systems and machine learning.

3.1. Data Presentation

During the process of fine-tuning and optimizing security solutions in the IoT-enabled
smart manufacturing plant, significant results were obtained. These demonstrate the
substantial improvements seen in system performance. These results are presented in
Table 2, which compares key metrics before and after the optimizations:

Table 2. Comparison of metrics before and after applying the optimization.

Metrics Initial Value Value after Optimization

Detection Rate (%) 85% 96%
False Positive Rate (%) 12% 3%
Average Response Time (seconds) 4.5 s 1.8 s
Alert Frequency (per hour) 120 65
False Positive Rate (%) 12% 3%

These values provide a quantitative view of how IoT security solutions evolved
and improved throughout the study. There was a significant increase in the anomaly
detection rate, which went from 85% in the initial phases of the study to a robust 96%
after optimizations. In addition, the false positive rate was significantly reduced to 3%,
indicating a substantial improvement in system accuracy. One of the most notable changes
was the reduction in the average response time, which was reduced to less than 2 s. This
enhancement ensured a fast and efficient response to abnormal situations, which is essential
for security on the smart manufacturing floor. The frequency of alerts generated by the
system also decreased, indicating a reduction in non-critical situations that require attention.
These results highlight the positive impact of tuned and optimized security solutions on
the plant environment.

3.2. Analysis of Data

We present detailed analysis of the results obtained before and after implementing
security solutions based on machine learning algorithms in the smart manufacturing plant.
This section focuses on evaluating the effectiveness of solutions in detecting anomalies and
improving security in IoT systems. For this, it is essential to highlight that the data used in
this study were collected over six months in the smart manufacturing plant. The data were
divided into two sets: one for training and one for model evaluation.
• Total volume of historical data collected: 15 terabytes (sensors) and 50,000 images
(cameras).
• Telemetry sensors: 50 sensors monitoring parameters such as temperature, pressure,
acceleration, and flow rates. Sampling frequency: 1 time per second.
• Images captured by IoT cameras: High-resolution images used for historical visual
data.
• Data format: CSV for sensors and JPEG image format for cameras.
• Data quality: High integrity and accuracy with quality control measures in place.
During the training process, different combinations of hyperparameters were exper-
imented with in order to achieve the best performance in anomaly detection. Here is an
Sensors 2023, 23, 8286 13 of 22

example of two hyperparameter configurations for the SVM model with an RBF kernel and
their respective results after the test set. These examples demonstrate how hyperparameters
can affect model performance. Configuration B achieved a higher anomaly detection rate,
although this was with a slightly lower false positive rate in this case.
Configuration A:
• Parameter C: 1.0
• Gamma parameter: 0.01
• Results in the test set:
• False positive rate: 5%
• Rate of true positives: 90%
Configuration B:
• Parameter C: 0.5
• Gamma parameter: 0.001
• Results in the test set:
• False positive rate: 3%
• Rate of true positives: 95%
The analysis is divided into categories of sensors and IoT devices, allowing us to
assess the performance of security solutions in each category. Table 3 summarizes the data
analysis results by type of sensor and IoT device before and after the implementation of
optimization. These results show each category’s anomaly detection rate and false positive
rate, along with the change in these metrics after optimizations.
• Before optimizations, telemetry sensors achieved an anomaly detection rate of 78%.
After the optimizations, this rate increased significantly to 91%, representing a 13%
increase in anomaly detection capability. Additionally, the false positive rate decreased
from 9% to 6%, indicating a 3% improvement in detection accuracy.
• For IoT cameras, the anomaly detection rate before optimizations was 88%. After
optimizations, this rate increased to 94%, which equated to a 6% increase in anomaly
detection capability. The false positive rate also decreased from 7% to 5%, representing
a 2% improvement in accuracy.
• IoT control data showed an 82% anomaly detection rate before optimizations. After
optimizations, this rate increased to 90%, reflecting an 8% increase in anomaly de-
tection capability. The false positive rate decreased from 10% to 7%, indicating a 3%
improvement in detection accuracy.

Table 3. Results of the Analysis by Category of Sensors and IoT Devices.

Anomaly Anomaly Change in

Sensors and False Positive False Positive Change in
Detection Rate Detection Rate Anomaly
Devices Rate (%) before Rate (%) after False Positive
(%) before (%) after Detection Rate
Category Optimizations Optimizations Rate (%)
Optimizations Optimizations (%)
Telemetry
78 91 +13 9 6 −3
Sensors
IoT cameras 88 94 +6 7 5 −2
IoT monitoring 82 90 +8 10 7 −3

The results show significant improvements in anomaly detection capability after

optimizations across all IoT sensor and device categories. This demonstrates the positive
impact of solutions based on machine learning algorithms on the efficiency and safety of
the intelligent manufacturing plant. Improved anomaly detection rates and reduced false
favorable rates support the effectiveness of these solutions and their ability to strengthen
security in IoT systems in an industrial environment.
Sensors 2023, 23, 8286 14 of 22

3.3. Impact on Efficiency

The implementation of security solutions based on machine learning algorithms had a
significant impact on the efficiency of the smart manufacturing plant. The early detection
of anomalies made it possible to address potential problems before they became critical.
This resulted in reduced machine and equipment downtime by an average of 17%, which
in turn increased overall plant efficiency. Preventive maintenance was carried out in a more
effective and scheduled way, reducing unplanned interruptions in production.
In addition, the ability to predict and prevent equipment failures allowed for the
more efficient use of resources and assignment of maintenance personnel. This translated
into significant savings in operating costs, with a 22% reduction and higher overall plant
productivity. Table 4 presents the key findings about efficiency and safety in our innovative
manufacturing plant before and after optimizations. One of the metrics we use to evaluate
impact is “Preventive Maintenance Efficiency”. This metric refers to the plant’s ability to
perform effective preventive maintenance on its equipment and machinery.

Table 4. Impact on efficiency and safety: key findings.

Parameter before Optimizations after Optimizations Change

Average Downtime (%) 22 5 −17
Operating Costs (%) 65 43 −22
General Plant Productivity (%) 78 95 +17
Reduction of Unplanned
27 5 −22
Interruptions (%)
Efficiency in Preventive
68 92 +24
Maintenance (%)

Preventive maintenance efficiency is calculated as follows:

Number of preventive maintenance performed

Preventive Maintenance Efficiency (%) = × 100 (1)
Total number of scheduled preventive maintenance
This metric measures the proportion of preventive maintenance that was successfully
carried out compared to the total amount scheduled. A higher value indicates greater
efficiency in performing this critical maintenance.
During the study, a significant increase in “Preventive Maintenance Efficiency” was
observed after the optimizations. This value rose from 68% before optimizations to 92%
after optimizations, marking a 24% improvement. This increase reflects the plant’s im-
proved ability to perform preventive maintenance more effectively, reducing unscheduled
downtime and saving maintenance costs.
Importantly, this metric is essential to ensuring the availability and reliability of
equipment and machinery in the smart manufacturing plant. High efficiency in preventive
maintenance helps to prevent critical failures and maintains high productivity and safety
in the manufacturing environment.
Regarding security in IoT systems, the implemented solutions have effectively de-
tected threats and mitigated risks. Specific examples of incidents include the detection of
unauthorized access attempts to IoT devices, denial of service attacks, and sensor data
tampering. A notable case was the early detection of an attempted intrusion into the teleme-
try sensor network. The security system identified abnormal behavior in an IoT device
trying to establish unauthorized communications. Immediate action was taken, including
isolating the device and notifying security teams. As a result, potential unauthorized access
and exploitation of the sensor network was prevented.
Another incident involved detecting a distributed denial of service (DDoS) attack
targeting IoT servers. The security solution identified an unusual overload in network
traffic and activated mitigation protocols to block the attack before it significantly affected
the availability of services. This resulted in a 30% reduction in successful DDoS attack
attempts. Table 5 presents several examples of how threats and security solutions based on
Sensors 2023, 23, 8286 15 of 22

machine learning can be detected and rapidly generate responses to threats in real time,
thus strengthening the security of IoT systems at the manufacturing plant.

Table 5. Impact on Security in IoT Systems: Incident Cases.

Detection Detection
Incident Threat Type Action Taken
Before After
Unauthorized Access Isolation and
Intrusion No Yes
Attempt to Device Notification
Denial of Service Traffic Activation of Mitigation
No Yes
(DDoS) attack overload Protocols
Sensor Data Sensor Data Data Restoration and
No Yes
Manipulation Manipulation Enhanced Security

The table presents a series of security-related incidents in IoT systems, details their
detection, and describes actions taken before and after optimizations. These incidents
represent situations evaluated in an innovative manufacturing plant environment, where
specific tests were conducted to assess the performance of the safety and anomaly detection
system. However, it is essential to note that the lack of detection of these attacks in the
past should not be attributed solely to security or software deficiencies. In many cases,
detecting cyber threats and anomalies in IoT systems is a complex challenge that requires the
implementation of advanced machine learning algorithms and updated security measures.
It is important to note that the incidents mentioned are not actual events at the smart
manufacturing plant during its regular operation. Instead, these are carefully designed
scenarios to test our system’s ability to detect cyber threats and anomalies in a controlled
environment. These tests are essential to evaluating the effectiveness of our security solu-
tions and ensuring that they can adequately respond to potential hazards in the production
environment. The actions taken after detecting these incidents represent recommended re-
sponses and preventative measures in the event of an attack. These actions are designed to
ensure the integrity and security of the smart manufacturing plant and have been optimized
due to our research and testing in simulated situations.
The average response time is critical for evaluating efficiency and safety in a smart
manufacturing environment. After optimizing our security solutions based on IoT and
machine learning, we significantly reduced the average response time to less than 2 s for
anomaly detection. This improvement was instrumental in proactively addressing unusual
events and minimizing potential negative impacts on production and safety.
However, this decrease in average response time was identified as raising questions
about the ability of security personnel to respond effectively in such a short period. To
address this issue, additional measures were implemented to ensure an appropriate and
timely response:
• Response automation: Part of our optimization strategy involved implementing au-
tomated responses to predefined, less severe anomalies. This means that specific
corrective actions are performed automatically, without human intervention, when cer-
tain common abnormalities are detected. This frees up security personnel to address
more critical events.
• Response hierarchy: A response hierarchy system was established to classify anomalies
based on their severity and potential risk. Critical anomalies are prioritized and
assigned to security personnel appropriately. This ensures that human resources focus
on events that require immediate attention.
• Training and drills: Security personnel have received specific training to respond
effectively to high-pressure, low-latency situations. In addition, regular drills are
carried out to ensure that staff are prepared to act quickly and accurately when
anomalies are detected.
Sensors 2023, 23, 8286 16 of 22

These measures ensure that security personnel can access the centralized dashboard
and take corrective action effectively, despite the speedy required response time. Automa-
tion and hierarchical response management have proven valuable in balancing detection
speed with security personnel’s responsiveness.

3.4. Evaluation of the Efficiency of the Algorithms

This section evaluates the efficiency of the anomaly detection algorithms implemented
in the intelligent manufacturing plant. Various metrics are used here to measure the
performance of the models and their ability to detect anomalies in the data accurately. Key
metrics include:
• Accuracy;
• Recall (sensitivity or rate of true positives);
• F1-Score;
• False positive sate;
• Positive predictive value or precision in positions;
• Area under the ROC curve (AUC-ROC);
• Area under the PR curve (AUC-PR).
We use test data separate from the historical data sets to evaluate these parameters.
We calculate these metrics before and after the optimizations in order to compare the
performance of the algorithms and demonstrate the improvement achieved. The results
show a significant improvement in the efficiency of the algorithms after the optimizations.
Precision, recall, and F1-Score increased, indicating better anomaly detection capabilities.
Additionally, the false positive rate decreased, reducing false alarms. The optimistic predic-
tion values also improved, as shown in Table 6. The AUC-ROC and AUC-PR increased,
reflecting the higher discrimination ability of the models. These results demonstrate the
positive impact of tuned and optimized security solutions on efficiency and security im-
provement in IoT systems used on the smart manufacturing floor.

Table 6. Evaluation of the Efficiency of the Algorithms.

before after
Metrics Change
Optimizations Optimizations
Precision 0.87 0.92 +0.05
recall 0.78 0.91 +0.13
F1-Score 0.82 0.91 +0.09
False positive rate 0.12 0.07 −0.05
Positive prediction Value 0.89 0.94 +0.05
Area under the ROC Curve
0.92 0.96 +0.04
(AUC-ROC)
Area under the PR Curve
0.85 0.92 +0.07
(AUC-PR)

3.5. System Efficiency Assessment

The efficiency of an IoT system is a crucial aspect, especially in industrial environments
where response speed and processing capacity can be critical to ensuring security and
operational efficiency. To assess system efficiency, we consider three key parameters:
system response time, processing speed, and overall system efficiency.
• Response time: Response time refers to the time it takes for the system to respond
to a request or to detect and notify researchers of an anomaly. After optimizations,
significant improvements in response time are observed. Data Take 2 reduces the
response time to 45 milliseconds, an improvement over Data Take 1 (50 ms).
• Processing speed: Processing speed measures the number of transactions or processes
the system can handle in one second. After the optimizations, the processing speed
increased remarkably. Data Collection 2’s processing speed reached 135 transactions
per second, compared to the 120 transactions per second seen in Data Collection 1.
Sensors 2023, 23, 8286 17 of 22

• System efficiency: The system’s overall efficiency is evaluated by accurately and

promptly detecting anomalies. After the optimizations, the system efficiency remained
high, exceeding 90%. In Data Collection 2, the system’s efficiency was 92%, indicating
that the optimizations did not compromise the system’s anomaly detection capacity.
Table 7 shows the system performance after implementing the optimizations, including
response time, processing speed, and system efficiency. To better demonstrate how these
metrics improved after the optimizations, the first row presents the performance values
before the optimizations.

Table 7. System Performance After Optimizations.

Response Time Processing Speed System Efficiency

Data Collection
(Milliseconds) (Transactions per Second) (%)
Before Opt. 60 100 85
Take 1 50 120 95
Take 2 45 135 92
Take 3 48 128 94
Take 4 52 115 91
Take 5 46 140 93

Before the optimizations, the system had a response time of 60 milliseconds, a process-
ing speed of 100 transactions per second, and an efficiency of 85%. After implementing
the optimizations, we saw significant performance improvements. The response time was
reduced to an average of 48 milliseconds across data collections, while the processing speed
increased to 129 transactions per second. The system’s efficiency improved by around 7%
on average, reaching values close to 94%. This direct comparison between performance be-
fore and after optimizations provides a clear view of how the improvements have positively
impacted system efficiency and responsiveness, resulting in a higher quality of service and
a better user experience in a smart manufacturing plant environment.
These results demonstrate that the optimizations applied to the smart manufacturing
plant’s IoT system improved security and increased operational efficiency. Reducing
response time and increasing processing speed are especially beneficial in critical situations
where anomaly detection and decision making must occur in real time. The system’s
high efficiency ensures that anomalies can be detected and addressed effectively without
compromising overall plant performance. These results support the implementation of
machine learning-based security solutions in IoT-based industrial environments.

3.6. Comparison of Machine Learning Algorithms

To deepen the validation of the system, a comparison of several machine learning
algorithms that were tested in the study was carried out in order to to evaluate which of
them worked best in the context of anomaly detection in the smart manufacturing plant, as
shown in Table 8. Key metrics such as precision, recall, F1-Score, and false positive rate
were considered to evaluate the performance of the algorithms.

Table 8. Comparison of Machine Learning Algorithms.

Accuracy Recall F1-Score False Positive

Algorithm
(%) (%) (%) Rate (%)
Support Vector Machine (SVM) 92.3 90.8 91.5 4.1
Random Forest 91.5 88.6 90.0 4.9
CNN 94.2 91.7 93.0 3.2

The results show that all three evaluated machine learning algorithms achieved out-
standing performances when detecting anomalies in the smart manufacturing floor. How-
ever, the CNN algorithm stood out with the highest accuracy, recall, and F1-Score, along
Sensors 2023, 23, 8286 18 of 22

with the lowest false positive rate. This indicates that the CNN algorithm is especially
effective at detecting anomalies in this IoT-based industrial environment. It is important
to note that the choice of machine learning algorithm may depend on the specifics of the
environment and security goals. In this case, the high performance of the CNN algorithm
demonstrates its ability to improve both efficiency and security in IoT systems in a smart
manufacturing plant. The comparison between algorithms provides valuable information
for decision making in implementing anomaly detection solutions in similar environments.
Notably, CNN has demonstrated an exceptional performance in terms of accuracy
and F1-Score in anomaly detection. However, it is true that CNN requires more significant
computational resources than SVM and random forest techniques. This difference in
computational power requirements must be considered when selecting an algorithm for
implementation in a real-time application.
To implement CNNs in real-time applications, despite their considerable computa-
tional power requirements, several strategies should be considered:
• Model optimization: Model optimization techniques can be applied to reduce the
complexity of the neural network without significantly compromising its performance.
This may include reducing layers or adjusting hyperparameters.
• Use of specialized hardware: In cases where a high performance is required, you can
choose to use specialized hardware, such as graphics processing units (GPU) or tensor
processing units (TPU), designed to accelerate deep learning tasks.
• Requirement-based decision making: The choice between SVM, random forest, and
CNN should be based on the application’s specific requirements. If accuracy is critical
and adequate computational resources are available, CNN can be a solid choice.
However, if real-time responses are needed in a situation with limited resources, SVM
or random forest may be more suitable.
• Parallel implementation: Parallel CNN implementation can help improve processing
efficiency and speed, essential for real-time applications. The use of distributed
systems or parallelization of tasks can be crucial considerations.
The choice of an anomaly detection algorithm must balance performance with avail-
able computational resources and real-time requirements. Despite its computational power
requirements, CNN can be successfully implemented in real-time applications using op-
timization techniques, specialized hardware, and decisions based on specific application
requirements. This selection must be carefully evaluated according to the individual needs
of each case.

4. Discussion
During the development of this research, we faced certain limitations and challenges
that it is essential to recognize. One of the significant challenges was the availability and
quality of historical data. Although considerable effort was to generate representative data,
there is always the possibility that the data may vary in its complexity and distribution [39].
Additionally, the lack of labeled data in some anomaly situations could affect the model’s
ability to detect these anomalies [40].
The results obtained in this work compare favorably with those of previous research
related to IoT security and anomaly detection in similar environments. Implementing
tight and optimized security solutions has significantly improved efficiency and security in
IoT systems in a smart manufacturing plant [41]. These findings support and expand the
existing knowledge base in IoT security and provide additional evidence for the efficacy of
machine-learning approaches in anomaly detection. While no specific case studies were
included in this article, it is essential to note that the results and methodologies presented
apply to various IoT environments [42]. The adaptability of the proposed solutions allows
for their implementation in different industrial and commercial settings, which broadens
their relevance and applicability.
This study is relevant in the current context of the increasing adoption of IoT in in-
dustrial and commercial environments. Security in IoT systems is a critical concern, and
Sensors 2023, 23, 8286 19 of 22

this study contributes to this field by providing a practical implementation of machine

learning-based security solutions [2,14]. The improvement in efficiency and the ability to
detect anomalies significantly impact the safe and continuous operation of a smart manu-
facturing plant. Furthermore, this work highlights the importance of addressing security
challenges in IoT systems by implementing solutions based on machine learning. Although
limitations were faced, the results obtained support the relevance and contribution of this
study to the field of IoT security.
One of the fundamental aspects at play when considering the implementation of an
IoT and machine learning-based diagnostic framework is the evaluation of the associated
costs. This research addresses this issue by analyzing the expenses related to infrastructure,
hardware acquisition, and algorithm implementation.
The creation of the infrastructure required to collect and process data from telemetry
sensors, IoT cameras, and monitoring devices can present a financial challenge. This
includes the cost of acquiring and maintaining servers, storage systems, and networking
equipment needed to handle the volume of data generated in a smart manufacturing plant.
Investment in a robust and scalable infrastructure is essential to ensuring the efficient
operation of the diagnostic framework. Choosing the proper hardware for implementing
IoT and machine learning-based security solutions can vary in cost. For example, using
high-quality sensors and IoT cameras can have a significant upfront cost. Additionally,
if you opt for specialized hardware such as graphics processing units (GPUs) or tensor
processing units (TPUs) for deep learning tasks, this should also be considered in the
budget.
Implementing machine learning algorithms, such as convolutional neural network
(CNN), may require additional resources in terms of development and training costs. This
could include hiring machine learning experts and acquiring software and development
tools. Significantly, although significant costs may be associated with the initial imple-
mentation of a diagnostic framework, the long-term benefits may outweigh these costs.
Improved operational efficiency and increased safety can lead to reduced operating costs
and substantial savings in the future.

5. Conclusions
This paper presents a detailed study of security solution in a smart manufacturing
plant using IoT and machine learning. Throughout the research, security challenges in an
IoT system were addressed, and strategies were developed in order to detect anomalies
effectively. The security solutions implemented in the intelligent manufacturing plant sig-
nificantly impacted the efficiency and security of operations. Improved anomaly detection
made it possible to identify abnormal situations in real time, resulting in a faster and more
effective response to unusual events. This led to a decrease in critical incidents and an
overall improvement in plant safety.
The optimization of machine learning algorithms and IoT infrastructure led to a
noticeable improvement in the operational efficiency of the plant. Tuned and optimized
anomaly detection models enabled accurate detection with reduced false favorable rates,
reducing unscheduled downtime and improving resource utilization. The results obtained
in this study compare favorably with previous research related to IoT security and anomaly
detection in industrial environments. The implementation of security solutions based
on machine learning proved to be effective in the early detection of anomalies and the
improvement of security in IoT systems.
It is essential to recognize the limitations and challenges encountered during the
investigation. The availability and quality of historical data may vary compared to actual
data. Additionally, the lack of labeled data in some anomaly situations could affect the
model’s ability to detect these anomalies. The field of security in IoT systems and machine
learning continues to evolve. Future research could incorporate emerging technologies,
such as natural language and advanced image processing, in order to improve anomaly
Sensors 2023, 23, 8286 20 of 22

detection. Integrating cloud-based, real-time security solutions could also expand detection
and response capabilities.

Author Contributions: Conceptualization, W.V.-C.; methodology, A.J.-A.; software, J.G.; validation,

J.G.; formal analysis, A.J.-A.; investigation, J.G.; data curation, W.V.-C. and A.J.-A.; writing—original
draft preparation, J.G.; writing—review and editing, A.J.-A.; visualization, J.G.; supervision, W.V.-C.
All authors have read and agreed to the published version of the manuscript.
Funding: This research received no external funding.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: The data used in this study were obtained from publicly available
sources and are detailed in the corresponding section. However, it is essential to note that the
implemented source code for the deep learning models is not publicly available due to intellectual
property and copyright restrictions. Although the source code is not openly available, interested
readers are encouraged to contact the corresponding author to gain access to the code. To request
the source code, please send an email to [email protected]. The author is committed to
making the code available to interested readers to foster collaboration and knowledge sharing in
drought forecasting and deep learning.
Acknowledgments: The data used in this study was obtained from participating company sources.
However, it is essential to note that the source code implemented for deep learning models is not
publicly available due to intellectual property and copyright restrictions. Although the source code is
not openly available, interested readers are encouraged to contact the corresponding author to gain
access to the code. To request the source code, please send an email to [email protected].
The author is committed to making the code available to interested readers to encourage collaboration
and knowledge sharing in machine learning.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Wu, H.-T. The internet-of-vehicle traffic condition system developed by artificial intelligence of things. J. Supercomput. 2022, 78,
2665–2680. [CrossRef]
2. He, M.; Petering, M.; LaCasse, P.; Otieno, W.; Maturana, F. Learning with supervised data for anomaly detection in smart
manufacturing. Int. J. Comput. Integr. Manuf. 2023, 36, 1331–1344. [CrossRef]
3. Mendia, I.; Gil-Lopez, S.; Grau, I.; Del Ser, J. A novel approach for the detection of anomalous energy consumption patterns in
industrial cyber-physical systems. Expert Syst. 2022, e12959. [CrossRef]
4. Zhang, Q.; Han, R.; Xin, G.; Liu, C.H.; Wang, G.; Chen, L.Y. Lightweight and Accurate DNN-based Anomaly Detection at Edge.
IEEE Trans. Parallel Distrib. Syst. 2022, 33, 2927–2942. [CrossRef]
5. Louk, M.H.L.; Tama, B.A. Revisiting Gradient Boosting-Based Approaches for Learning Imbalanced Data: A Case of Anomaly
Detection on Power Grids. Big Data Cogn. Comput. 2022, 6, 41. [CrossRef]
6. Concetti, L.; Mazzuto, G.; Ciarapica, F.E.; Bevilacqua, M. An Unsupervised Anomaly Detection Based on Self-Organizing Map for
the Oil and Gas Sector. Appl. Sci. 2023, 13, 3725. [CrossRef]
7. Chen, P.-Y.; Yang, S.; McCann, J.A. Distributed Real-Time Anomaly Detection in Networked Industrial Sensing Systems. IEEE
Trans. Ind. Electron. 2015, 62, 3832–3842. [CrossRef]
8. Nizam, H.; Zafar, S.; Lv, Z.; Wang, F.; Hu, X. Real-Time Deep Anomaly Detection Framework for Multivariate Time-Series Data in
Industrial IoT. IEEE Sens. J. 2022, 22, 22836–22849. [CrossRef]
9. Rousopoulou, V.; Vafeiadis, T.; Nizamis, A.; Iakovidis, I.; Samaras, L.; Kirtsoglou, A.; Georgiadis, K.; Ioannidis, D.; Tzovaras, D.
Cognitive analytics platform with AI solutions for anomaly detection. Comput. Ind. 2022, 134, 103555. [CrossRef]
10. Liu, X.; Deng, R.H.; Miao, Y.; Vasilakos, A.V. Guest Editorial: 5G-Enabled Intelligent Application for Distributed Industrial
Internet-of-Thing System. IEEE Trans. Ind. Inform. 2022, 18, 2807–2810. [CrossRef]
11. Buzau, M.M.; Tejedor-Aguilera, J.; Cruz-Romero, P.; Gomez-Exposito, A. Detection of Non-Technical Losses Using Smart Meter
Data and Supervised Learning. IEEE Trans. Smart Grid 2019, 10, 2661–2670. [CrossRef]
12. Ahmad, H.S.; Arshad, M.J.; Akram, M.S. Device Authentication and Data Encryption for IoT Network by Using Improved
Lightweight SAFER Encryption with S-Boxes. Int. J. Embed. Real-Time Commun. Syst. 2021, 12, 1–13. [CrossRef]
13. Kim, S.-G.; Park, D.; Jung, J.-Y. Evaluation of One-Class Classifiers for Fault Detection: Mahalanobis Classifiers and the
Mahalanobis–Taguchi System. Processes 2021, 9, 1450. [CrossRef]
Sensors 2023, 23, 8286 21 of 22

14. Saba, T.; Rehman, A.; Sadad, T.; Kolivand, H.; Bahaj, S.A. Anomaly-based intrusion detection system for IoT networks through
deep learning model. Comput. Electr. Eng. 2022, 99, 107810. [CrossRef]
15. Mian, T.; Choudhary, A.; Fatima, S.; Panigrahi, B.K. Artificial intelligence of things based approach for anomaly detection in
rotating machines. Comput. Electr. Eng. 2023, 109, 108760. [CrossRef]
16. Park, M.; Jeong, J. Design and Implementation of Machine Vision-Based Quality Inspection System in Mask Manufacturing
Process. Sustainability 2022, 14, 6009. [CrossRef]
17. Abdel-Basset, M.; Moustafa, N.; Hawash, H. Privacy-Preserved Generative Network for Trustworthy Anomaly Detection in
Smart Grids: A Federated Semisupervised Approach. IEEE Trans. Ind. Inform. 2022, 19, 995–1005. [CrossRef]
18. Huang, M.; Liu, Z.; Tao, Y. Mechanical fault diagnosis and prediction in IoT based on multi-source sensing data fusion. Simul.
Model. Pract. Theory 2020, 102, 101981. [CrossRef]
19. Bin Mofidul, R.; Alam, M.; Rahman, H.; Jang, Y.M. Real-Time Energy Data Acquisition, Anomaly Detection, and Monitoring
System: Implementation of a Secured, Robust, and Integrated Global IIoT Infrastructure with Edge and Cloud AI. Sensors 2022,
22, 8980. [CrossRef]
20. Lu, S.; Lu, J.; An, K.; Wang, X.; He, Q. Edge Computing on IoT for Machine Signal Processing and Fault Diagnosis: A Review.
IEEE Internet Things J. 2023, 10, 11093–11116. [CrossRef]
21. Serradilla, O.; Zugasti, E.; de Okariz, J.R.; Rodriguez, J.; Zurutuza, U. Adaptable and Explainable Predictive Maintenance:
Semi-Supervised Deep Learning for Anomaly Detection and Diagnosis in Press Machine Data. Appl. Sci. 2021, 11, 7376.
[CrossRef]
22. Hlávka, J.P. Security, Privacy, and Information-Sharing Aspects of Healthcare Artificial Intelligence. In Artificial Intelligence in
Healthcare; Academic Press: Cambridge, MA, USA, 2020.
23. Tange, K.; De Donno, M.; Fafoutis, X.; Dragoni, N. A Systematic Survey of Industrial Internet of Things Security: Requirements
and Fog Computing Opportunities. IEEE Commun. Surv. Tutor. 2020, 22, 2489–2520. [CrossRef]
24. Bourdon, M.; Gimenez, P.F.; Alata, E.; Kaaniche, M.; Migliore, V.; Nicomette, V.; Laarouchi, Y. Hardware-Performance-Counters-
Based Anomaly Detection in Massively Deployed Smart Industrial Devices. In Proceedings of the 2020 IEEE 19th International
Symposium on Network Computing and Applications, NCA 2020, Cambridge, MA, USA, 24–27 November 2021.
25. Susto, G.A.; Terzi, M.; Beghi, A. Anomaly Detection Approaches for Semiconductor Manufacturing. Procedia Manuf. 2017, 11,
2018–2024. [CrossRef]
26. Alfeo, A.L.; Cimino, M.G.; Manco, G.; Ritacco, E.; Vaglini, G. Using an autoencoder in the design of an anomaly detector for
smart manufacturing. Pattern Recognit. Lett. 2020, 136, 272–278. [CrossRef]
27. Gültekin, Ö.; Cinar, E.; Özkan, K.; Yazıcı, A. Real-Time Fault Detection and Condition Monitoring for Industrial Autonomous
Transfer Vehicles Utilizing Edge Artificial Intelligence. Sensors 2022, 22, 3208. [CrossRef]
28. Tran, D.H.; Nguyen, V.L.; Nguyen, H.; Jang, Y.M. Self-Supervised Learning for Time-Series Anomaly Detection in Industrial
Internet of Things. Electronics 2022, 11, 2146. [CrossRef]
29. Yazdinejad, A.; Dehghantanha, A.; Parizi, R.M.; Hammoudeh, M.; Karimipour, H.; Srivastava, G. Block Hunter: Federated
Learning for Cyber Threat Hunting in Blockchain-Based IIoT Networks. IEEE Trans. Ind. Inform. 2022, 18, 8356–8366. [CrossRef]
30. Gamec, J.; Basan, E.; Basan, A.; Nekrasov, A.; Fidge, C.; Sushkin, N. An Adaptive Protection System for Sensor Networks Based
on Analysis of Neighboring Nodes. Sensors 2021, 21, 6116. [CrossRef]
31. Chakravarthi, S.S.; Kannan, R.J.; Natarajan, V.A.; Gao, X.-Z. Deep Learning Based Intrusion Detection in Cloud Services for
Resilience Management. Comput. Mater. Contin. 2022, 71, 5117–5133. [CrossRef]
32. Rosa, L.; Cruz, T.; de Freitas, M.B.; Quitério, P.; Henriques, J.; Caldeira, F.; Monteiro, E.; Simões, P. Intrusion and anomaly detection
for the next-generation of industrial automation and control systems. Futur. Gener. Comput. Syst. 2021, 119, 50–67. [CrossRef]
33. Fahim, M.; Sillitti, A. Anomaly Detection, Analysis and Prediction Techniques in IoT Environment: A Systematic Literature
Review. IEEE Access 2019, 7, 81664–81681. [CrossRef]
34. Benaddi, H.; Jouhari, M.; Ibrahimi, K.; Ben Othman, J.; Amhoud, E.M. Anomaly Detection in Industrial IoT Using Distributional
Reinforcement Learning and Generative Adversarial Networks. Sensors 2022, 22, 8085. [CrossRef]
35. Tanuska, P.; Spendla, L.; Kebisek, M.; Duris, R.; Stremy, M. Smart Anomaly Detection and Prediction for Assembly Process
Maintenance in Compliance with Industry 4.0. Sensors 2021, 21, 2376. [CrossRef]
36. Guan, S.; Zhao, B.; Dong, Z.; Gao, M.; He, Z. GTAD: Graph and Temporal Neural Network for Multivariate Time Series Anomaly
Detection. Entropy 2022, 24, 759. [CrossRef]
37. Kong, F.; Li, J.; Jiang, B.; Wang, H.; Song, H. Integrated Generative Model for Industrial Anomaly Detection via Bidirectional
LSTM and Attention Mechanism. IEEE Trans. Ind. Inform. 2021, 19, 541–550. [CrossRef]
38. Mansouri, A.; Majidi, B.; Shamisa, A. Metaheuristic neural networks for anomaly recognition in industrial sensor networks with
packet latency and jitter for smart infrastructures. Int. J. Comput. Appl. 2021, 43, 257–266. [CrossRef]
39. Truong, H.T.; Ta, B.P.; Le, Q.A.; Nguyen, D.M.; Le, C.T.; Nguyen, H.X.; Do, H.T.; Nguyen, H.T.; Tran, K.P. Light-weight federated
learning-based anomaly detection for time-series data in industrial control systems. Comput. Ind. 2022, 140, 103692. [CrossRef]
40. Ibrahim, M.; Alsheikh, A.; Awaysheh, F.M.; Alshehri, M.D. Machine Learning Schemes for Anomaly Detection in Solar Power
Plants. Energies 2022, 15, 1082. [CrossRef]
Sensors 2023, 23, 8286 22 of 22

41. Li, Z.; Fei, F.; Zhang, G. Edge-to-Cloud IIoT for Condition Monitoring in Manufacturing Systems with Ubiquitous Smart Sensors.
Sensors 2022, 22, 5901. [CrossRef]
42. Savic, M.; Lukic, M.; Danilovic, D.; Bodroski, Z.; Bajovic, D.; Mezei, I.; Vukobratovic, D.; Skrbic, S.; Jakovetic, D. Deep Learning
Anomaly Detection for Cellular IoT with Applications in Smart Logistics. IEEE Access 2021, 9, 59406–59419. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.