Automated Decision Support Framework for IoT : Towards a Cyber

Physical Recommendation System

Mohammad Choaib

1,2

, Moncef Garouani

, Mourad Bouneffa

, Nicolas Waldhoff

, Adeel Ahmad

and Yasser Mohanna

Univ. Littoral C

ote d’Opale, LISIC, Laboratoire d’Informatique Signal et Image de la C

ote d’Opale, France

Department of Electronics and Physics, Lebanese University, Lebanon

Univ. Littoral C

ote d’Opale, UDSMM, Unit

e de Dynamique et de Structure des Mat

eriaux Mol

eculaires, France

nicolas.waldhoff@eilco-ulco.fr, adeel.ahmad@univ-littoral.fr, yamoha@ul.edu.lb

Keywords:

Cyber Physical System, Machine Learning, Industry 4.0, Decision Support Systems, NLP.

Abstract:

Among the key factors of Industry 4.0 are the intensive use of Cyber Physical Systems (CPSs). These systems

can be implemented at many application levels for various application domains. Each application requires a

personalized CPS architecture concerning the Physical sensors as well as the management of Cyber systems.

Furthermore, it requires also the maintenance of the communications among these different sub-systems. How-

ever, researchers and engineers may lack the essential knowledge to design and build an autonomous CPS. In

this paper, we propose the concept of a novel Cyber Physical Recommendation System (CPRS) in order to

address this open challenge. The proposed approach is subjected to assist the competencies of researchers

and engineers to design and build more efﬁcient CPS according to the given objective, domain, and input

application scenarios. In this regard, CPRS recommend the components of the desired CPS, based on a novel

architecture model of the components, connections, and tasks of the CPSs. The proposed system is eventually

intended to aide the progress in leading factories towards the maturity of fourth industrial revolution.

1 INTRODUCTION

The practices and massive data in Industry 4.0 have

signiﬁcantly proven their utility for the transformation

phase of manufacturing processes. The current litera-

ture in this domain prominently demonstrates the re-

search work towards the tuning of machine learning

algorithms and parameters to transform factories into

smart industrial components (Garouani. et al., 2021).

The main concept is to deal with the major chal-

lenges faced by the companies, especially the evolu-

tion of big data, arbitration of decision-making, real-

time monitoring and control of integration of novel

technologies such as Cyber Physical Systems (CPS)

or enabling Internet of Things (IoT). It is principally

intended to better manage the interconnections be-

tween the Cyber systems and Physical systems.

The CPSs are generally described as physical and

engineered systems whose operations are monitored,

coordinated, controlled, and integrated by a commu-

nication and computing core (Rajkumar et al., 2010).

A CPS is recognized as a networked collection of

cyber systems and physical systems that are moni-

tored and controlled by user-speciﬁed semantic laws.

For this purpose, a network connects the cyber and

physical systems to construct a real-time system on

a large scale (Chung, 2017). In this respect, a CPS

can be viewed as a real-time feedback system where

the cyber sub-systems and physical sub-systems are

inter-connected with multiple means of communica-

tion. The recent structures of CPS are often imple-

mented with the increasing use of smart sensors and

embedded systems in addition to the practices in-

volving the cloud computing, data storage, and arti-

ﬁcial intelligence techniques (Garouani et al., 2022c).

The major objective is to mainly pursue the idea of

transformation of industry towards the rise of more

smarter factories that can further enhance the matu-

rity of the fourth industrial revolution. It can also

help, among others, in the advancement of predictive

maintenance, real-time monitoring systems, and im-

proved self-optimization capabilities (Garouani et al.,

2022a).

The CPSs have been widely used in the real-time

systems. Most of these systems have customised con-

ﬁguration according to the application domains and

Choaib, M., Garouani, M., Bouneffa, M., Waldhoff, N., Ahmad, A. and Mohanna, Y.

Automated Decision Support Framework for IoT: Towards a Cyber Physical Recommendation System.

DOI: 10.5220/0011848900003467

In Proceedings of the 25th International Conference on Enterprise Information Systems (ICEIS 2023) - Volume 1, pages 365-373

ISBN: 978-989-758-648-4; ISSN: 2184-4992

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

365

their architecture and connection system. In this con-

text, each application needs its own physical com-

ponents, communication network, and computational

primacy. These could change according to the objec-

tive, type of physical entities being measured, type of

generated output data and parameters,etc. The system

stakeholders usually lack the essential knowledge to

identify and choose suitable components that match

their requirements (Garouani et al., 2022d). Thereby,

the automated assistance for the required human ex-

pertise can allow the engineers and researchers of

smart factories to rapidly build, validate, and deploy

CPS solutions. It may also improve their quality

of service, productivity, and more importantly, re-

duce the need of the interventions from human ex-

perts (Garouani et al., 2022b).

Motivated by this goal, in this paper, we discuss

the architecture, concept and application domain of

the CPSs. Later on, we present the architecture and

connection of a personalized CPS that serve as a base

for the CPRS (Cyber Physical Recommendation sys-

tem). The proposed system mainly attempts to pro-

vide necessary recommendations for building a CPS

according to the presented set of objectives and ap-

plication scenarios. In this regard, CPRS recom-

mends the components of the desired CPS, based on

a novel architecture of the components, connections,

and achievable tasks. This system is intended to help

in leading the smart factories toward the fourth in-

dustrial revolution. It might enable the identiﬁcation

of CPS parts and components to save time and effort

avoiding unintended accidents.

The rest of the paper is organized as follows : The

section 2 discusses the architecture, concept and do-

main of application of CPS. The section 2.3 presents

our own CPS architecture and connection to be a ba-

sis for the proposed CPRS. The section 3 introduces

the main components of the proposed recommenda-

tion system and discusses how these components col-

laborate to achieve the pursued goals. We later on

show, how to implement the framework for represen-

tative application scenarios, in the section 4, and we

discuss the prototypical implementation. Finally, the

section 5 concludes the paper and outlines future per-

spectives.

2 RESEARCH BACKGROUND

AND RELATED WORK

Nowadays, CPSs are termed as one of the key fac-

tors of the fourth industrial revolution, as they help

in achieving the goal of intelligent, resilient, and self-

adaptable machines required by the industry 4.0 (Lee

et al., 2015). As a result of their ability to integrate

into a wide range of domains and applications, they

are considered to be one of the fastest growing tech-

nologies in the world (Raisin et al., 2020).

2.1 Application Domains

This section presents an overlapped overview of some

of the advanced CPSs used in different domains and

application levels. We attempt to assess the internal

mechanisms of CPSs for various domains; healthcare,

manufacturing, and transportation are the most signif-

icant but neither limited nor exhausted to be applied

across different contexts in order to better understand

the nature of machine states and anomalies. These are

brieﬂy reviewed in the following sections.

2.1.1 Healthcare Domain

Healthcare domain broadly encompass the applica-

tions used in hospital management systems, inten-

sive care units, assisted living, and/or monitoring the

health condition of elderly patients. Healthcare is a

very delicate and sensitive domain to work with since

it depends on accurate decision-making in real-time

analysis. Among them, the CPS with the implica-

tion of cloud computing and big data can play a major

role in the development and improvement of the med-

ical science applications. These may use the stored

data either structured or unstructured e.g the doctors

input and feedback, in addition to the data collected

from medical instruments, biosensors and wearable

devices for high-performance modeling and accurate

decision making. In (Banerjee et al., 2011) the au-

thors present a framework for modeling and analyzing

Cyber Physical Medical frameworks, with two ma-

jor applications. The ﬁrst, is an infusion pump sys-

tem for future prediction of drug concentration while

the second is a multi-infusion pump for chemother-

apy. Healthcare domain is a diverse ﬁeld of research

that has been performed with the help of numerous

CPSs architectural designs and implemented applica-

tions (Haque et al., 2014).

2.1.2 Manufacturing Domain

For the last decade, most of the research literature

has been focused towards the development of smart

factories for more reliable and efﬁcient results. The

market needs have shown continuous demand of new

requirements in contexts of optimal cost, quality, ﬂex-

ibility, and real-time execution. In this regard, the

recent CPSs are integrated as a technological evolu-

tion that has mainly resulted, in response to achieve

these demands. Furthermore, the CPSs are embedded

ICEIS 2023 - 25th International Conference on Enterprise Information Systems

366

in various manufacturing applications like automa-

tion (Garouani et al., 2022c), shop ﬂoor, etc. For ex-

ample a systematic approach for predictive produc-

tion systems is proposed by (Lee et al., 2017) using

CPSs to inject resilience and interoperability so that

the productivity of manufacturing can be optimized.

We believe that data generated by these systems can

be used to further improve the equipment intelligence

and reduce energy consumption to beneﬁt environ-

mental constraints. It require to take advantage of big

data analytic and data enabled predictive collabora-

tive decision making for the real time control systems,

hence, increase the production efﬁciency (Wan et al.,

2018).

2.1.3 Transportation Domain

The transportation domains often involve the np-

complex combinatorial problems. Many of the prob-

lems, such as trafﬁc congestion, have arisen due to the

increase in the number of transportation vehicles and

actors. Trafﬁc bottle-necks, air pollution, and safety

issues require to address a carefully focused atten-

tion on each individual elements of a transportation

system. The transportation systems can reach a new

smart solution for different applications with the in-

tegration of smart sensors, embedded systems, and

CPSs computing capabilities. The cooperative vehi-

cle safety systems (Fallah et al., 2010), intelligent in-

tersection management systems (Zheng et al., 2017),

intelligent charging systems for electric vehicles (Ge

et al., 2012), and different application in other trans-

portation ﬁelds (aviation, automotive aerospace, intel-

ligent transportation, trafﬁc control, etc.) are some of

the examples in this domain.

2.2 CPS Architectures

Although the main concept of the CPS remains the

same, the state of the art literature shows a variety

of different architectures. For example, Jamwal et

al. present a CPS architecture for Intelligence Manu-

facturing Systems (IMS) spanned over three layers, as

given below (Jamwal et al., 2020) :

Physical Connection Layer: The concerned indus-

trial components at this layer embed the equip-

ment such as sensors and various types of mea-

surement devices in the manufacturing resources.

Different communication means are used to con-

nect the involved set of machines in order to es-

tablish a proper, robust, and uniform connection

among the actuators, manufacturing resources,

measurement instruments, and sensors.

Middle Layer: This layer is aimed at data transfer,

data management, and device management. It is

also used to allow the interconnection between the

external applications, the elements of the physical

layer and the computation layer.

Computation Layer: This layer is of utmost impor-

tance to carry the decision-making with the help

of analysis and processing of stored data. The ma-

jor objective is to render the autonomous capabil-

ities for the machine functions in order to make

them self-aware and self-dependent.

Some authors, in (Yan et al., 2019) present a four-

level framework for CPS that is partitioned into sen-

sors and control, connection, cognition, and appli-

cation & services. Although we can also witness

in (Lee et al., 2015) a ﬁve-layer architecture where

the authors present the CPS system as composed of

a smart connection level, data-to-information conver-

sion level, cyber level, cognition level, and conﬁgura-

tion level.

(Djouad et al., 2018),also present an architecture en-

compassed at ﬁve major parts, as discussed below :

1. The Physical sphere contains physical process

(e.g. Temperature) and Physical entities (cars, hu-

mans, animals, etc.)

2. The Cyber-Physical sphere contains sensors, ac-

tuators, and a user interface for control.

3. The Cybernetic sphere contains computational el-

ements, storage elements, and a CPS applica-

tion (which consists of computational services de-

signed to assist the user by interacting with other

cyber components).

4. The Data and Information sphere contains the data

collected by sensors and the output information

generated by the computational elements.

5. The Deployment sphere contains hardware and

software components that can offer information

and data; it may provide actuating capabilities.

2.3 Proposed CPS Architectures and

Connections

The ﬂexibility and multi-dimensional capabilities of

using the CPSs allow them to be incorporated across

a vast range of application domains. These are may

include predictive maintenance, monitoring and con-

trol for the applications in agriculture, electric smart

grid, instrumentation, infrastructure and communica-

tions systems, etc. Inspired by the CPS design con-

cept given by (Djouad et al., 2018), we propose a

novel CPS concept which is intended to be more sim-

ple and more convenient to be used as a base concept

Automated Decision Support Framework for IoT: Towards a Cyber Physical Recommendation System

367

for the proposed recommendation system in our ap-

proach. This architecture is intended to be used as

a conceptualization of the Cyber Physical Systems.

This means that it will be used as a sphere of CPS

ontology. We eventually intend to develop a recom-

mendation system concerning the design of CPS. The

proposed architecture play a central role in the deﬁ-

nition of the knowledge-base that may constitute the

core of the recommendation system.

The proposed architecture is depicted in the ﬁg-

ure 1 which divides the CPS into three main parts :

Physical Sphere: The physical sphere contains com-

ponents responsible on interacting with the real

world systems to measure and collect information

or to perform a given task.

1. Sensing Sphere is responsible to collect data

and information from the physical entities by

using smart components and smart sensors.

2. Execution Sphere is responsible to perform

and execute commands given by the output data

after being analyzed and processed. For exam-

ple, these orders can be performed by actuators,

machines or even using alarms.

3. Human in The Loop use the HCI (Human

Computer Interaction) strategies such as the

user can trigger an actuator by giving command

whenever it is needed.

4. Control Unit controls the module by receiving

and modifying data to be processed on cyber

level, followed by the execution of produced

commands.

5. Micro-Controllers contains the devices which

are used to read and store measured data that

shall be scheduled later on for further analysis

and processing.

Network Sphere is subjected to the communication

means and establishment of connection among the

devices of physical sphere. This part must con-

sider the following aspects :

1. Security: The development of CPS involves

the various interconnected systems. It also

involves the progress of networking, comput-

ing, and integration of novel smart technologies

that most domains nowadays requires. Among

other, these include the quality factors like the

efﬁciency, accuracy, and authenticity. So due to

vast interconnection and the use of IoT, CPSs

can be widely contingent on various malicious

uninvited access through cyber-attacks, pollut-

ing input data, or the manipulation of physical

entities. So, security measures regarding the

system limitations and background must be pri-

oritized during the building of a CPS.

2. Data Transport Sphere: This part is responsi-

ble for the interconnection between the physi-

cal and the cyber layer where data transmission

is done using communication technologies such

as BLT, 4G, 5G, LANs, IR, RFID, Wi-Fi, and

Zigbee along with other technologies. Due to

the increasing number of devices which are in-

terconnected through the communication pro-

tocols over the internet (TCP/IP, IPv4, IPv6).

These must ensure data routing and transmis-

sion either through cloud computing, or inter-

net gateways. The communicated data can also

be subjected to the encryption in order to ensure

enhanced security to prevent malicious attacks.

Cyber Sphere is ﬁnally responsible for analyzing,

processing, and storage of data and information

which have been subsequently measured in the

physical layer that may produce more data. It

might be subjected to control instructions for

the computational elements or to execute speciﬁc

tasks. Some of the sub-system modules in this

part are as follows :

1. Front-End Module: This module implements

the interactive platform which are used to vi-

sualize data (Input or output) to present the de-

sired information on usable user interfaces.

2. Back-End Module: This module is a part of

the middle layer between the data and the front-

end module. Since the back-end processes ma-

nipulate the stored data by some deﬁned algo-

rithms to generate an output decision, that can

be used to control or exploit the database to

eventually develop an application subject to the

user groups.

3. Storage Module: This module gives an easy

access to store and manage data that may be

used to analyze the physical entity conditions.

Examples of storage systems are DB, Cloud,

Excel, MongoDB, oracle, MySQL, etc.

4. Server Module: The servers offer numerous

services to varied network objectives. Even

though they come in different forms (Computer

software programs, or hard drives) they mostly

carry out the same function of receiving, stor-

ing, and sharing data. The servers also play a

crucial role in a company’s ability to collab-

orate, since these provide the ability to com-

municate the essential data components for the

clients and the dependency services that the

other applications may need.

(Oks et al., 2019) propose a reference architec-

ture for the design of the demonstrators for industrial

cyber-physical systems. It displays the components

ICEIS 2023 - 25th International Conference on Enterprise Information Systems

368

CPS

Network

WIFI/RF/IR/BLE

ZIgbee/Cable/....

ProtocolsSecurity

Physical SphereCyber Sphere

Sensing Part Execution Part

Human In the

loop

Sensors

Actuators and

alarms

Control Unit

Micro-

Controllers

HMI

RPi/Arduino

FrontendBackendStorageServerDB/Cloud/etc..

MySql/MongoDB

Figure 1: General overview of the proposed CPS architecture.

of the CPS from the physical sphere to communica-

tion and cyber spheres. It also include the objectives,

scenarios, conﬁgurations for better demonstration of

the connections among these components.

We eventually intend to develop a general connec-

tion concept that can be applied in most of the appli-

cations domains and that can serve as a base for the

recommendation system. The illustrated concept in

ﬁgure 2 serves as a framework to build a CPS accord-

ing to the user’s objective. This architecture is com-

posed of two major parts; the ﬁrst part is the physical

sphere which contains the HMI for human interfer-

ence, along with the sensing and actuating compo-

nents. Additionally, it has the control unit and the

micro-controllers to store data and carry out com-

mands. While the second part, which is the cyber

sphere consists of the front-end and back-end for the

data processing and presentation. It also involves the

database and the server components to communicate

and transfer data.

Control UnitSensor

Actuator C

HMI

Physical Sphere

Data

Base

Front End

Back End

Cyber Sphere

Server

Figure 2: Proposed CPS Connection.

The communication network mainly establishes

link between the physical sphere and cyber sphere.

It also deals with the communication links among the

different sub-components of the physical and cyber

spheres. According to the given scenario, these com-

ponents (sensors, actuators, DB, etc.) shall be recom-

mended as deemed ﬁt to the user needs to achieve ul-

timate objective.

3 THE RECOMMENDATION

SYSTEM

Every application requires its own physical compo-

nents, communication network, and computational

part. The stakeholders of the system usually lack the

required knowledge to identify the suitable compo-

nents that are susceptible to match their requirements.

Thus, it may require a decision support system that

can help stakeholders to identify the CPS parts and

components, which can save time and effort along

with avoiding the mistakes and accidents. The gener-

ated output is further maintained with the help of the

suggestion engine through the use of a Knowledge-

Base (KB). The knowledge-base contains the collec-

tion of meta-features that describe the sensors, their

speciﬁcations, concerned domains and any descrip-

tions. The global architecture of the proposed frame-

work is illustrated in Figure 3.

The system receives the user input (such as prod-

uct features, domain, case, price range, etc.) through

the interactions with the chatbot. In case of the intro-

duction of a fresh problem the system retrieves the

information regarding the user’s objective, applica-

tion scenario completed by the properties, details and

any other information regarding the new input project.

This input is mainly achieved by the frontend. As a

Automated Decision Support Framework for IoT: Towards a Cyber Physical Recommendation System

369

User

Chat Bot

front -end

user input

(Product features,

domain,case,price,

range,etc).

New Problem

Specifications,

product

features,domain,

case,price,

range,etc

Retrieve

(Similarity, grouping, logical query)

back-end

Revision

(ranking, confidence)

Output

(Recommendation,

user feedback,

etc.)

Figure 3: The global architecture of the Recommendation

System.

result, a new problem is generated and stored in the

KB which shall be analysed later on. In this respect,

when the system receives a new unseen problem, it

ﬁrst extracts the meta-features that describes the prob-

lem then these are analyzed and processed using Nat-

ural Language Processing (NLP) techniques in order

to extract the keywords that describe the domain of

the application. The steps in this order are executed to

make use of the stored data in the KB (speciﬁcation,

description, and domain) to produce a sorted list, af-

ter a fetch and return. It contains the candidate which

refers to particular sensors in the KB. Retrieval of

data is made according to similarity, grouping, and

logical queries. Then, the output is revised accord-

ing to the obtained ranking and conﬁdence, before its

recommendation to the user. Feedback questions are

presented to take the opinion of users according to

the compatibility of the result to his/her objective and

needs. In this respect, the feedback allows the knowl-

edge base to be upgraded with the results of the each

new problem.

The Meta-knowledge base provides the means to

store and search the knowledge issued from ofﬂine

phase (the training of the recommendation system).

Broadly, it can be divided into the following two ma-

jor categories :

• Storing the knowledge (description, speciﬁca-

tions, meta-features) in a structured way.

• Searching and acquiring information while driv-

ing explanations acquisition process.

The proposed system intends to make consider-

able use of ontologies in form of knowledge graphs,

both for improved information gathering and query

understanding by the suggestion engine as well as for

facilitating the acquisition of explanations.

Below is an example of the data stored in the KB

where sensors are presented by means of name, type,

speciﬁcation, description, and domain.

10 M. Choaib et al.

The proposed system intends to make considerable use of ontologies in form

of knowledge graphs, both for improved information gathering and query under-

standing by the suggestion engine as well as for facilitating the acquisition of

explanations.

Below is an example of the data stored in the KB where sensors are presented

by means of name, type, speciﬁcation, description, and domain (FHI humidity

sensor) :

Listing 1.1. xml ﬁle that describe the humidity sensor

1 < Se ns or >

2 < Name > FH I </ Na me >

3 < Type > Hum id it y se ns or </Typ e >

4 < Speci fi catio n >

5 < Po wermax > 5 </ Po wermax >

6 < RangeminC > -60 </ Ra ngeminC >

7 < RangemaxC > 140 </ Ra ngemaxC >

8 < Rangemi nR H > 0 </ Ra ng eminRH >

9 < Rangema xR H > 10 0 < / Ran ge ma xRH >

10 < Accur ac ym in > -1 < / Acc ur acymin >

11 < Accur ac ym ax > 1 </ Accurac ym ax %>

12 < Price > 5.8 $ </ Price >

13 </ Spe ci ficat io n >

14 < Descr ip ti on >

15 <Use > Designed for high volume , c ost s en si tive

appli ca tions </U se >

16 < Manuf ac turer > TE C onnec ti vity </ Ma nufac tu rer >

17 < Data T ype > A na lo g </Dat a Type>

18 < Progr am mable > Tr ue < / Pro gr ammab le >

19 < Commu ni catio n > Cable </ Commu ni catio n >

20 </ Des cr iption >

21 < Do ma in >

22 < Histori ca l App >

23 1 -au to motiv e c abin air control

24 2 -hom e ap pliance s

25 3 -in du stria l p ro cess control systems

26 4 -Me te orolo gy

27 </ Histo ri cal App >

28 < Do ma in of A pp > I nd us tr y </ Do ma in of App >

29 </ Domain >

30 </Sensor >

5 Conclusion and prospective

The Cyber Physical Systems are usual complex amalgamation of hardware and

software components. These are applied on various application levels for diﬀerent

domains. Hence, the CPS conﬁgurations diﬀer from domain to domain and at

multiple application levels. Furthermore, with technological evolution, the CPS

Figure 4: Xml ﬁle that describe a humidity sensor.

4 CASE STUDY : REAL TIME

SURVEILLANCE SYSTEM

To illustrate our approach, we consider the construc-

tion of a Patient Surveillance System (PSS) for elderly

people. In this context, the main purpose is to help

the users in achieving their objective by analyzing

the needs. For instance, concerning the PSS prob-

lem, the objective is to build a system that support

in the day-to-day living tasks and assure the patient

safety. For this purpose, the recommendation system

provides the information on sensors/actuators related

to the different application levels and tasks. The even-

tual goal is to choose the most convenient set of sen-

sor(s)/actuator(s) among the recommendations made

by the system.

In the proposed system, the eventual recommen-

dation is chosen interactively, based on the details

provided by the user with the help of the chat-

bot. These are presented in form of the textual key-

words (such as reason, objective, budget, etc.), as il-

lustrated in Figure 5. The details may be clear or

somehow vague according to the user’s knowledge of

the domain. The user’s description can range from ex-

ICEIS 2023 - 25th International Conference on Enterprise Information Systems

370

Objective

I need a home surveillance

system for elderly, to monitor

their movement, and to detect

gas leak, fire, and humidity

rising.

1-Objective?

2-Need?

3-Budget?

4-Range?

etc.

Chat Box

Home Survellance

Monitor Movement

Detect Gas Leak ,fire,

and humidity

400$

200 m^2

User's Input

Figure 5: Example:Home Surveillance for Elder Patients (Data Input).

plaining the exact need with details (such as explain-

ing the need for monitoring the patient’s movement,

pill-taking times, or other stuff. It may also include

conﬁguration of material regarding his safety e.g a

gas leakage, ﬁre detection, extreme humidity, etc.) or

otherwise it can include the general parameters (such

as describing the need for patients monitoring system

or home safety surveillance system). Regardless the

issue, the system must be able to analyze and rec-

ommend either a precise recommendation (a speciﬁc

type of sensors like pressure sensor) or a general one

(a set of multiple sensors concerning the same appli-

cation such that gas, ﬁre detection or humidity sen-

sors). Still, in both cases, it must respond according

to the user’s needs.

For the sake of simplicity, we take the PSS appli-

cation as a case study as it can correspond to a large

number of public such as the elderly people. In this

case, the system can be considered at two consecutive

levels, as explained in the following :

• Data Input: In this part, the users can interact with

the system using the chatbot through a series of

questions where the user can express his objec-

tive, need, use case, budget, range, etc. According

to the users preferences the input data can be gen-

erated as shown in ﬁgure 5. This data can be used

by the system, later on for the analysis in order to

generate more accurate recommendation.

• Data Analysis: In this part, the input text can be

pre-processed using e.g segmentation technique

which is used to split the text into meaningful seg-

ments that can be composed into words, sentences

or topics, or the tokenization technique that can be

used as a way for separating a piece of text into

smaller units called tokens. These can be broadly

classiﬁed into three types: word, character, and

sub-word. For the sake of clarity, let us consider

the word monitoring, it can be tokenized as char-

acters: m-o-n-i-t-o-r-i-n-g. Similarly, we may use

Stemming technique, which in fact is a process of

removing a part of a word, or reducing a word to

its stem or root for example monitor is the root of

the word monitoring.

As a result of pre-processing keywords we formu-

late the data preparations that can be generated, as

shown in ﬁgure 6. The processed data can be ana-

lyzed using semantic analysis based on RNN/LSTM.

The choice of RNN/LSTM is justiﬁed by the fact that

RNN are an important variant of neural networks that

are heavily used in Natural Language Processing. It

can be further observed in the literature that it pro-

vides ﬂexibility in the network to work with vary-

ing lengths of sentences. The Recurrent Neural Net-

work (RNN) has a major advantage, in comparison

to the standard neural networks, as it gives the back

propagation technique that may enlarge the training

data set for the recurrent learning. In addition, it

also manifest the advantage of sharing features that

are learned across different positions of text which in

both cases can’t be achieved with standard neural net-

work. The Long Short-Term Memory (LSTM) algo-

rithm is a type of a popular RNN for learning sequen-

tial data prediction problems. Like any other neural

network, LSTM comprises a few layers that aid in

pattern recognition and learning for improved perfor-

mance. The fundamental function of an LSTM can

be viewed as holding the necessary data while dis-

carding the data that is neither necessary nor help-

ful for subsequent prediction. All this analysis can

be stored in the Knowledge-Base that is built using

Neo4j (V

agner, 2018). Neo4j is an open-source graph

oriented database management system. It allows data

to be represented as nodes connected by a set of arcs

whereas each object (either node or an arc) can have

their own properties.

Automated Decision Support Framework for IoT: Towards a Cyber Physical Recommendation System

371

NLP Module

Semantic

analysis

based on

RNN/LSTM

Processing

segmentation

tokenization

stemming

etc.

Extracted Key Words

Data

Processing

Data

Analysis

Monitoring

Elder

Patients

Real Time

Humidity

400$

Movement

Home Safety

Fire

Gas

Knowledge Base

Neo4J

MongoDB

Recommendation classification criteria

Gas sensor : X32ofg, 10$, 87%

efficiency.

Humidity sensor : FHI, 5.8$, 88%

efficiency.

Fire Detection sensor :2HU, 3$, 90%

efficiency.

Output example:

Explainabilty of

Recommendation

User's Input

New

Problem

Figure 6: Example:Home Surveillance for Elder Patients (Data Analysis).

The Output recommendation can be classiﬁed on

the basis of some criteria, as chosen by the user

with the help of the communication via the chat-

bot. These criteria can include the classiﬁcation on

the basis of budget, efﬁciency, range, etc. The output

can be presented to the user on the chatbot includ-

ing the explainability factors of the generated recom-

mendation. The explainability part render the capa-

bility to justify the recommendation in a transparent

manner, as it provides the explanations of the recom-

mendation system on multiple levels. For instance,

let us consider the recommendation of a temperature

sensor. This sensor is recommended because it ﬁts

the objective of the user, since a similar application

was found that has used this sensor in the historical

traces (like Room Temperature Measurement applica-

tions). The knowledge-base provides the means to

query the trances in the Historical App component,

as shown in the listing 4. We can also observe that

the speciﬁcation ﬁts well the criteria from the budget

constraints (5 euros), temperature range (0°C,100°C)

and accuracy (-1,1) as were given by the user. Sim-

ilarly, the speciﬁcations from the knowledge-base

can give the supplementary information (power range,

data type, ability to be programmable or not, and

communication types, etc.) related to the sensors.

5 CONCLUSION AND

PROSPECTIVE

The Cyber Physical Systems are usually complex

amalgamation of hardware and software components.

These are applied on various application levels for

the different domains. Hence, the CPS conﬁgura-

tions may differ from domain to domain and at mul-

tiple application levels. Furthermore, according to

the technological evolution, the CPS are required to

be maintained at both the layers of hardware com-

ponents and their articulating software components;

which presents a major research challenge. In this pa-

per, we attempt to brieﬂy encompass the architecture,

concept and domains of different applications of CPS.

In this respect, we present a novel CPS design and

modeling approach that may allow the maintenance

ICEIS 2023 - 25th International Conference on Enterprise Information Systems

372

and management of its conﬁgurations to be stored in

a knowledge-base. The facts in the knowledge-base

are further exploited with the help of a Cyber Physi-

cal Recommendation System (CPRS). In this context,

the CPRS presents a model architecture that may pro-

vide necessary recommendations for building a CPS

according to the presented objective, details, and ap-

plication scenario. In the future work, we plan the

to improve the knowledge-base using the training and

testing phase with the help of natural language pro-

cessing algorithms and techniques. While this must

also provide means to explore the explainabilty as-

pects of such a supporting system.

ACKNOWLEDGEMENTS

The authors would like to thank the University of the

Littoral Cote d’Opale, and the Lebanese University

for the ﬁnancial support.

REFERENCES

Banerjee, A., Gupta, S. K., Fainekos, G., and Varsamopou-

los, G. (2011). Towards modeling and analysis of

cyber-physical medical systems. In Proceedings of the

4th International Symposium on Applied Sciences in

Biomedical and Communication Technologies, pages

1–5.

Chung, D. S. (2017). STUDENT NUMBER: 2708062

COURSE NAME: CIS 601 DEPARTMENT: Depart-

ment of electrical engineering and computer science.

page 22.

Djouad, A., Atil, F., and Beddiar, C. (2018). Domain model

for cyber-physical systems. page 8.

Fallah, Y. P., Huang, C., Sengupta, R., and Krishnan, H.

(2010). Design of cooperative vehicle safety systems

based on tight coupling of communication, comput-

ing and physical vehicle dynamics. In Proceedings of

the 1st ACM/IEEE international conference on cyber-

physical systems, pages 159–167.

Garouani, M., Ahmad, A., Bouneffa, M., and Hamlich, M.

(2022a). AMLBID: An auto-explained automated ma-

chine learning tool for big industrial data. SoftwareX,

17:100919.

Garouani, M., Ahmad, A., Bouneffa, M., Hamlich, M.,

Bourguin, G., and Lewandowski, A. (2022b). Towards

big industrial data mining through explainable auto-

mated machine learning. The International Journal

of Advanced Manufacturing Technology, 120:1169–

1188.

Garouani, M., Ahmad, A., Bouneffa, M., Hamlich, M.,

Bourguin, G., and Lewandowski, A. (2022c). Using

meta-learning for automated algorithms selection and

conﬁguration: an experimental framework for indus-

trial big data. Journal of Big Data, 9:1169–1188.

Garouani., M., Ahmad., A., Bouneffa., M., Lewandowski.,

A., Bourguin., G., and Hamlich., M. (2021). Towards

the automation of industrial data science: A meta-

learning based approach. In Proceedings of the 23rd

International Conference on Enterprise Information

Systems - Volume 1: ICEIS, pages 709–716.

Garouani, M., Ahmad, A., Bouneffa, M., Lewandowski,

A., Bourguin, G., and Hamlich, M. (2022d). Towards

meta-learning based data analytics to better assist the

domain experts in industry 4.0. In Artiﬁcial Intelli-

gence in Data and Big Data Processing, pages 59–66,

Cham. Springer International Publishing.

Ge, Y., Dong, Y., and Zhao, H. (2012). A cyber-physical en-

ergy system architecture for electric vehicles charging

application. In 2012 12th International Conference on

Quality Software, pages 246–250. IEEE.

Haque, S. A., Aziz, S. M., and Rahman, M. (2014). Review

of cyber-physical system in healthcare. 10(4):217415.

Publisher: SAGE Publications Sage UK: London,

England.

Jamwal, A., Agrawal, R., Manupati, V. K., Sharma, M.,

Varela, L., and Machado, J. (2020). Development

of cyber physical system based manufacturing system

design for process optimization. 997(1):012048.

Lee, J., Bagheri, B., and Kao, H.-A. (2015). A cyber-

physical systems architecture for industry 4.0-based

manufacturing systems. 3:18–23.

Lee, J., Jin, C., and Bagheri, B. (2017). Cyber physical

systems for predictive production systems. 11.

Oks, S. J., Jalowski, M., Fritzsche, A., and M

oslein, K. M.

(2019). Cyber-physical modeling and simulation: A

reference architecture for designing demonstrators for

industrial cyber-physical systems. 84:257–264. Pub-

lisher: Elsevier.

Raisin, S. N., Jamaludin, J., Rahalim, F. M., Mohamad, F.

A. J., and Naeem, B. (2020). Cyber-physical system

(CPS) application- a REVIEW. 1(2):52–65. Number:

Rajkumar, R. R., Lee, I., Sha, L., and Stankovic, J. (2010).

Cyber-physical systems: the next computing revolu-

tion. In Proceedings of the 47th Design Automation

Conference on - DAC ’10, page 731. ACM Press.

agner, A. (2018). Store and visualize EER in neo4j. In

Proceedings of the 2nd International Symposium on

Computer Science and Intelligent Control. ACM.

Wan, J., Yin, B., Li, D., Celesti, A., Tao, F., and Hua,

Q. (2018). An ontology-based resource reconﬁgu-

ration method for manufacturing cyber-physical sys-

tems. 23(6):2537–2546. Publisher: IEEE.

Yan, J., Zhang, M., and Fu, Z. (2019). An intralogistics-

oriented cyber-physical system for workshop in the

context of industry 4.0. 35:1178–1183.

Zheng, B., Lin, C.-W., Liang, H., Shiraishi, S., Li, W., and

Zhu, Q. (2017). Delay-aware design, analysis and ver-

iﬁcation of intelligent intersection management. In

2017 IEEE International Conference on Smart Com-

puting (SMARTCOMP), pages 1–8. IEEE.

Automated Decision Support Framework for IoT: Towards a Cyber Physical Recommendation System

373