Human Digital Twin in Industry 4.0: Concept and Preliminary Model

Yannick Naudet, Alexandre Baudet and Margot Risse

Luxembourg Institute of Science and Technology (LIST), Luxembourg

Keywords:

Digital Twin, Human Model.

Abstract:

Digital Twins originally concern technical systems and do not yet integrate human elements properly. This

limits their quality and usefulness when we consider systems where machines and human workers still cohabit.

This paper presents the concept of Human Digital Twin (HDT), the human equivalent of a Digital Twin (DT),

which aim at being coupled with DTs of technical elements in systems where humans play a role. We detail

the state of the art on the subject, propose a deﬁnition for HDT and a preliminary human model, bringing

foundations for handling the human factor in industry with digital twins.

1 INTRODUCTION

The Digital Twin (DT) concept originates from the

aerospace ﬁeld, with premises around 2003, accord-

ing to (Tao et al., 2019), and a ﬁrst publication in

2010 (Shafto et al., 2010). It is now an important key-

word in the domains of Industry 4.0, Factory of the

Future and Smart Factory. According to (Negri et al.,

2017), research on the DT in manufacturing is an evo-

lution of the research stream about Virtual Factories.

There is still no common deﬁnition of what a Dig-

ital Twin is. (Negri et al., 2017) reports 16 different

deﬁnitions from 2010 to 2016 mainly in the aerospace

and manufacturing ﬁelds. According to the authors,

a Digital Twin “provides a virtual representation of a

system along its life-cycle”, where “optimizations and

decision-making would then rely on the same data

that are updated in real-time with the physical sys-

tem”. This is further reﬁned in: “The Digital Twin is

meant as the virtual and computerized counterpart of

a physical system that can be used to simulate it for

various purposes, exploiting a real-time synchroniza-

tion of the sensed data coming from the ﬁeld”.

But as digital representation of our world are cre-

ated with the hype on digital twins, they focus on vir-

tual representations of human-built systems (e.g. fac-

tories, cities, buildings, machines...) and tend to for-

get the essential element: human-beings. Actually, be

it objects or environments, all have some kind of in-

teraction with humans: objects are used by humans

and environments are spaces where objects and hu-

mans cohabit and can have interactions. With the In-

ternet of Things and the progress in Artiﬁcial Intelli-

gence (AI), among other technologies, we have seen

the emergence of the so-called smart things, or smart

systems if we take a generic perspective, including

smart objects and smart environments. This smart-

ness can take different forms and be implemented in

many ways, but it will never be really smart if the cen-

tral variable is forgotten. Finally smart systems are

designed to support humans in the best possible way,

and reaching this objective without having an under-

standing of humans needs or humans themselves is

probably not optimal for the human-smart system in-

teraction. In this perspective, AI needs models of hu-

mans whatever their form and scope, and digital twins

of smart systems need also models that represent dig-

itally the humans they interact with or that they con-

tain. This is where we introduce the concept of the

Human Digital Twin (HDT), which emerged in our

discussions about digital twin in 2019, and at the same

time in the heads of other researchers as we will show

in the next section.

2 STATE OF THE ART

The oldest reference to the concept of Human Dig-

ital Twin comes from the health sector, where it is

conceived as a sophisticated simulation of the human

body (Blake, 2016). Today, only some small parts

of the human body can be modelled to run simula-

tion of treatments (e.g. arthritis in the knee), and the

complexity of the human body is still too high for our

understanding to build a complete model. However,

technological advances in 3D scanning, wearable IOT

Naudet, Y., Baudet, A. and Risse, M.

Human Digital Twin in Industry 4.0: Concept and Preliminary Model.

DOI: 10.5220/0010709000003062

In Proceedings of the 2nd International Conference on Innovative Intelligent Industrial Production and Logistics (IN4PL 2021), pages 137-144

ISBN: 978-989-758-535-7

137

and AI allow already high precision and near real-

time modelling of more and more complex parts.

Taking a generic perspective applicable to any

smart system, Hafez introduces the HDT as a

“human-speciﬁc smart machine dedicated to align-

ing human objectives with the smart machines sup-

porting her”. His approach focuses on ﬁnding re-

current human-machine interaction patterns to main-

tain this alignment by anticipating human responses

in given contexts (Hafez, 2020). (Zibuschka et al.,

2020) build on this view, and characterise the HDT

according to the C2PS (Cloud-based Cyber-Physical

System) reference architecture for DT (Alam and

El Saddik, 2017). It is then deﬁned as a technical

system comprising virtual sensors gathering observa-

tions about one or several humans, functional units fo-

cusing on behavior analysis which derive knowledge

about users. The authors highlight smart home, build-

ing and ofﬁce as application ﬁeld and emphasize the

interest for identity management and data protection.

2.1 Related Approaches

In computer ergonomics, Digital Human Modelling

(DHM) refers to a human model as a 3D representa-

tion of a human including anthropometric and kine-

matic aspects ( (Case et al., 2016)), used especially

in inclusive engineering (3D) design since around

1972. In short, DHM build virtual humans that are

realistic 3D representations in motion, based on data

gathered from extensive surveys on human popula-

tions and completed now by 3D scanned templates

of the human body. DHM can certainly be consid-

ered as a basis for building HDT, from the 3D rep-

resentation perspective. Originally in this ﬁeld, the

knowledge of user behaviour characterising real peo-

ple and linked to their physiological or psychological

characteristics (e.g. mood, fatigue, stress) is left to

the ergonomist, and not yet embedded in the human

model. However, embedding further cognitive, be-

havioural or emotional human characteristics is one of

the research tracks for design tools identiﬁed by (Case

et al., 2016), which shows some convergence with the

needs of a HDT.

The Personal Digital Twin (PDT) is another con-

cept similar to HDT, introduced in 2020 (Saracco

et al., 2020) as a solution to tracking needs and pan-

demics control induced by the Covid-19 crisis and

for personalized healthcare. The PDT is deﬁned as

a “representation of various aspects of a person that

might include the movement of the person, the inter-

actions that person has in physical space with other

people, and her health status”. Through the observa-

tions from the sensors embedded in personal mobile

devices (e.g. smartphones and wearables), the goal

is to digitize physical persons, and thus creating their

DT, to “enable anonymous and secure sensing, collec-

tion and analysis of data to inform strategic decisions

that can disrupt the current ways the healthcare sys-

tem works and manages pandemic situations”. The

authors further emphasize that it can be used to cre-

ate virtual social spaces where PDT are connected to-

gether, able to share data and seek for advice from

others. PDT also can act as personal assistant and

proxy to medical authorities, in a networked environ-

ment where PDT manage personal data, can reason

from local context, interact with other PDT, with the

health system and central authorities. The advantages

for predictive analysis, awareness transmission and

problems (here, crisis) prevention are clear, as well

as the induced privacy concerns related to the anytime

tracking of persons and their consent to share personal

data.

2.2 On the Web

HDT is the subject of an anonymous blog at hu-

mandigitaltwin.com

dated from May 2020. The au-

thor refers to having a human model including any

sort of data related to a person and its experience, that

could be clustered according to six dimensions: phys-

ical, cognitive, emotional, social, occupational and ﬁ-

nancial. Here, the concept relates to data sovereignty

and originates from the idea that humans should own

and operate data they produce in the digital world, and

thus stay in control of their digital self, which is unfor-

tunately not the case today. Finally, HDT is foreseen

technically as a human model, analytic functions al-

lowing to process the data from the model and a user

interface (including a dashboard) to access and con-

trol data and processing analytics results.

Avenga Labs refers to HDT as digital twins of

people, which are digital representations of humans

as complex physical objects

. The link to digital ac-

tivities and related data is also made, referring to the

digital shadow of a person, and the privacy concerns

this induces. Another company, Proglove, advertises

largely the HDT, presented as the digital counterpart

of the human worker and introduced also as the miss-

ing representation of humans in Industry 4.0 Digital

Twins

. They present wearables as the enabling tech-

nology for building HDTs, to gather data about work-

ers in supply chains, as a necessary tool providing ac-

https://www.humandigitaltwin.com/

https://www.avenga.com/magazine/human-digital-

twins/

https://itsupplychain.com/human-digital-twin-the-

digital-counterpart-to-the-human-worker/

IN4PL 2021 - 2nd International Conference on Innovative Intelligent Industrial Production and Logistics

138

tionable insights to deal with unexpected potentially

complex situations

. This is referred also by the Pi-

cavi company in the logistics sector, as a representa-

tion of a human being resulting from the tracking of

its activities

. It is used for analytics, process im-

provement and training.

Finally, several references can be found in the

health sector, comprising dedicated research centers.

We can cite the Semic RF company, proposing a

product called Digital Body Total, which is a digital

replica of a human’s being organs and biological or

molecular systems, embedding AI

. Originally called

Cyber Bio Twin

, it aim at providing a framework

for medical diagnosis, experimentation, predictions

and support for clinical trials and decision-making. In

academy, the Geriatronics initiative of the MSRM re-

search center of the Technical University of Munich

works on a HDT for personalised diagnostics, which

is a humanoid DT able to create a real-time visualiza-

tion of the physiological processes within the body

Last, the OnePlanet research center in Netherlands

refers to the HDT as a data-driven digital platform that

collects data related to an individual’s health and nu-

trition, and analyses it to provide personalised advices

on diet, lifestyle and medicines

2.3 In Industry

Because Industry 4.0 has been focusing a lot on dig-

italisation, human factors are not often considered

within researches on Digital Twin. So far, workers

were seen more as spectators than actors (Peruzzini

et al., 2018, citing (Hermann, Pentek, & Otto, 2017)),

but they should instead be considered as part of the in-

telligent system, where they can generate data for ma-

chine programming and processes optimisation while

beneﬁting from inputs and “collaboration with smart

systems” (Peruzzini et al., 2018). In this work, the au-

thors experiment a human-centred approach to indus-

trial systems, where they include human-related data

in the Digital Twin of a factory. Although efforts are

done to account for human factors in the factory of

the future, see e.g, (Longo et al., 2019), which we

detail in section 3, these are not yet integrated with

the DT. Indeed in a recent review on Digital Twin

for smart manufacturing highlights Digital Twin for

https://www.proglove.com/blog/digital-twin/human-

digital-twins-boost-actionable-insights/

https://picavi.com/en/human-digital-twin-productive-

onboarding-for-new-employees/

https://semic.de/en/ai/semic-health

https://www.cyberbiotwin.com/

https://geriatronics.msrm.tum.de/human-digital-twin/

https://oneplanetresearch.nl/innovatie/digital-twin/

people as a research challenge, considering humans

are not yet considered as integral parts of the smart

manufacturing system (Lu et al., 2020). Modelling

humans is expected to help understanding personal

wellbeing and working conditions, designing human-

centred human-machine collaboration taking into ac-

count physical and physiological factors in the pro-

duction optimisation process, and allow building per-

sonalised virtual training.

(Nikolakis et al., 2019) addresses the digital twin

of manual (human-centred) operations, implemented

as a part of a Cyber-Physical System observing and

controlling the shop ﬂoor for optimising human-based

production thanks to simulations. In this context, the

DT integrates both the human operator and its envi-

ronment, but a proper digital human modelling ap-

proach integrating human-speciﬁc factors is essen-

tial. Based on this, simulations can be used to im-

prove at the same time the production quality and the

ergonomics of human-centred operations. To go a

step further than using kinematics-based models, the

authors gather real-time observations of the human

worker behaviour through motion capture, recording

operations and attaching motion patterns and con-

straints to tasks. This allows then to make more realis-

tic simulations in a 3D environment, where the human

model behaves according to a motion model imple-

mented from in-situ operations, highlighting in partic-

ular what should be adapted in the physical working

space conﬁguration. Implemented in a closed control

loop, this ultimately leads to shopﬂoor reconﬁgura-

tion for optimised production quality and ergonomics,

without the need to interrupt the production process.

(Baskaran et al., 2019) highlight the need to rep-

resent both human and machine elements in DT of

industrial processes. They implement the HDT as

a 3D model of human workers to simulate a manu-

facturing process with human-robot interaction. The

model includes biomechanical, anthropometric and

ergonomics characteristics allowing to simulate hu-

man average behaviours validated and standardised

by ﬁeld studies. This HDT approach is used to study

the ergonomic impact in what-if scenarios using a

digital manufacturing simulation tool. HDT is in-

troduced as a “human digital model implementing a

near-real time digital image of a physical human in a

virtual environment”. Indeed, Human-Robot Collab-

oration (HRC) is one of the main application ﬁelds

where digital representations of humans are useful,

whatever their form. In (Lu et al., 2020), virtual sim-

ulations models are used to dynamically allocate and

coordinate tasks between human and robot, based on

their skills and state, respecting the best workload bal-

ancing. The observation of human behaviour com-

Human Digital Twin in Industry 4.0: Concept and Preliminary Model

139

pared to a reference allows the robot to adapt its be-

haviour to human factors affecting the work quality or

the task duration (e.g. skill level, motivation, failure

sensitivity for complex processes).

3 THE HDT CONCEPT

3.1 Deﬁnition

We introduce here a generic deﬁnition of the Human

Digital Twin, as a speciﬁc DT dedicated to humans:

Deﬁnition (Human Digital Twin). A Human Digital

Twin (HDT) is a subclass of the Digital Twin whose

particularity lies in the human nature of the twinned

entity. It is a real-time mirroring computerized sys-

tem of a human agent, able to simulate or emulate his

characteristics and behavior in context.

To some extent, an HDT would be based on the same

general principles than a classical DT, except the sub-

stantial difference that the twined physical system

consists of a human agent. Therefore, the HDT can be

considered as a subclass of DT, inheriting from all its

properties, but which also implies to take some spec-

iﬁcations into account during all conceptualization,

modeling, implementation and maintenance phases.

The HDT should be familiar with tasks’ short and

long term objectives of its physical twin, and inte-

grates some AI-based tools to analyse incoming phys-

ical and physiological data, and to provide predictive

operations. As synchronized states mirroring and up-

dating remain primary stakes, the HDT implementa-

tion must be articulated around real-time data gath-

ering from the human entity. Where this task do

not seems to show major difﬁculties with non-human

agents, continuously measuring human physical and

physiological data and transforming them into signif-

icant and actionable knowledge reveals much more

impediments. But before all, the HDT needs to rely

on a proper human model, enough close to reality to

allow simulation and possibly emulation.

3.2 Virtual Human

A starting point for conceiving a (structural and be-

havioural, in contract to a visual representation like in

DHM) human model for a HDT can be to look at the

concept of Virtual Human, to which a book is dedi-

cated (Burden and Savin-Baden, 2019). Summarising

from the multiple views and deﬁnitions from the liter-

ature, authors deﬁne virtual humans as “Software pro-

grams which present as human and which may have

behavior, emotion, thinking, autonomy and interac-

tion modelled on physical human capabilities”. The

concept is the averaged view representing systems

ranging from simple virtual humans that enact only

partially, in a simplistic way the listed properties, to

Virtual Sapiens: “Sophisticated virtual humans which

achieve similar levels of presentation, behavior, emo-

tion, thinking, autonomy, interaction, self-awareness

and internal narrative to a physical human”.

(Burden and Savin-Baden, 2019) propose a model

for virtual human, based on one mandatory charac-

teristic -it is virtual-, and ten traits that characterise

also human-beings: (1) embodiment (physical or vir-

tual); (2) humanity (humanoid or not); (3) natural

language communication; (4) autonomy; (5) emotion

(demonstrating and responding to); (6) personality

(own speciﬁc behaviour); (7) reasoning; (8) learning;

(9) imagination; and (10) self-awareness (linked to

sentience at the extreme). Virtual humans can be de-

ﬁned according to the proportion of this traits they

have, with at the boundaries the virtual humanoid

(lower bound, which has only very basic character-

istics) and the virtual sapiens, who has all the traits at

full level.

3.3 Human Factors in Industry 4.0

In 2019, an extensive work has been done on deﬁn-

ing a taxonomy of human factors, encompassing all

the capabilities of industrial workers inﬂuencing their

work (Longo et al., 2019). Factors are classiﬁed ac-

cording to three spheres of capabilities, namely cog-

nitive, physical and psychological, which are divided

in traits, themselves divided in facets. In total, there

are 11 traits and 50 facets. This taxonomy provide

a big set of characteristics that can be included in a

human model. However this is only a basis. Indeed

to build a model allowing to simulate/emulate human

behaviour, the way each of these characteristics work

and inﬂuence each other should be sought in relevant

theories. In the following we detail the main factors

and highlight the theories we think the most relevant.

3.4 Relevant Factors and Models

If several ways are possible to model a virtual human

or a HDT, cognitive architectures constitute probably

a highly relevant one. As explained in (Lieto et al.,

2018), using cognitive architectures indicates “both

abstract models of cognition, in natural and artiﬁcial

agents, and the software instantiations of such mod-

els”. Psychology and computer sciences were the pi-

oneering sciences to consider human user in a cogni-

tive architecture context. With more than estimated

IN4PL 2021 - 2nd International Conference on Innovative Intelligent Industrial Production and Logistics

140

300 cognitive architectures, the application domains

and variables of each architecture are varied and het-

erogeneous. Brieﬂy, most of the cognitive architec-

tures are focused on one or several of the follow-

ing core cognitive abilities or capabilities: percep-

tion, attention, action selection, memory, learning,

reasoning and metacognition (Kotseruba and Tsot-

sos, 2020). Most of those capabilities are present in

virtual humans. Learning and memory are consid-

ered together in the virtual human model, perception-

attention-action is linked to autonomy, and metacog-

nition can be mapped to imagination.

One of the most known model implementing part

of these cognitive abilities is certainly BDI (Belief,

Desire, Intention), which is used since years by the

Multi-Agent-Systems community. BDI agents are

generally characterized by affective states such as

emotions, mood or personality but sometimes also

by affective capacities such as empathy or emotional

regulation (S

anchez-L

opez and Cerezo, 2019). Dis-

cussing on how to model human social behaviour

in agent-based systems, (Kennedy, 2012) list a set

of basic principles to implement: human ability to

process sensory information; personality; motivations

and needs; rationality and the ability to represent

knowledge, learn, memorize and act according to this

knowledge ); emotions, leading to intuitive and un-

conscious decision-making; and in social behaviour

the imagination about others behaviour (theory of

Mind) and its inﬂuence on self-behaviour. The author

drives to BDI, but also PECS (Physical, Emotional,

Cognitive and Social factors) and ”fast and frugal de-

cision hierarchies” conceptual frameworks for agents

on one side, and on the other side to cognitive archi-

tectures like Soar and ACT-R (see the book chapter

for references).

Another example of human behaviour simulation

by a software agent can be found in (Kamara-Esteban

et al., 2017), where the human agent implements a be-

haviour model following the Behaviour, Activity, Ac-

tion principle, where behaviour is deﬁned by an ac-

tivity that consists in a set of consecutive actions. As

a core human ability, action selection is the process

of “what” has to be decided and “how” to decide it.

Several factors – relevance, utility and internal fac-

tors - have an impact on the “next” action, but the

main variance of a behaviour is mostly determined

by internal factors, which are key in cognitive archi-

tectures (Kotseruba and Tsotsos, 2020), but also in

human-computer interaction, social robotics and vir-

tual agents. Whatever the model or architecture and

the set of abilities considered, these factors are the

most important.

Internal factors of a human are intangible forces

that may cause, moderate or increase a response to an

event. When they are considered, there are at least

three main interdependent determinants to get a faith-

ful model, which we detail here: personality, desire,

and emotions.

• Personality, as a relatively stable variable (Costa

and McCrae, 1988) in most cultures in the world,

can help to predict and model patterns of be-

havior. With the Five Factor Model and Big 5

theory, personality is broken down into 5 main

dimensions (Openness, Conscientiousness, Ex-

traversion, Agreeability, Nevrosism) and depend-

ing on the assessment tool into 30 traits. Follow-

ing several combinations of personality dimen-

sions and emotions, typical reactions can be pre-

dicted (Shvo et al., 2019). Since decades, person-

ality, in addition to intelligence (also called cog-

nitive ability or g factor), are the two biggest pre-

dictors of job performance and training efﬁciency,

whatever the job, as shown in thousand of studies,

e.g., (Schmidt and Hunter, 1998).

• Desire or Motivation is also a key determinant of

a human behavior. In several cognitive architec-

tures, preservation drives, curiosity or interaction

drives are examples of modelled drives. To eas-

ily draw the inter-dependencies with emotions and

personality, we consider that the Reiss 16 desires

model could be relevant as suggested by (Shvo

et al., 2019). Any behavior is usually executed to

satiate one or more basic desires (e.g. power, cu-

riosity, social contact, saving, etc) and the human

user has always to choose which drive is priority.

• Emotions, unlike personality, have a transient na-

ture. Like motivation, several emotion models ex-

ist. Russell’s circumﬂex model of affect (Russell,

1980), ﬁrstly proposed in 1980, instigated a lot of

progress and innovation for emotion understand-

ing and comprehension. This model allows to ref-

erence 28 emotion-denoting adjectives, spatially

represented in a two-dimensional circle, depend-

ing on the arousal and valence levels to which they

are associated. Today, the OCC theory (Ortony

et al., 1990) is one of the most used. It is com-

posed of 21 emotions (joy, distress, fear, anger,

love, hate, etc.) with several levels (primary, sec-

ondary, tertiary). For an innate or primary emo-

tion (e.g. love), a human may feel a secondary

emotion (e.g. affection) and then as consequence

of the secondary emotion, a tertiary emotion (e.g.

tenderness).

In addition to the choice of models for every

internal force, the literature showed that the inter-

Human Digital Twin in Industry 4.0: Concept and Preliminary Model

141

dependencies between each force is also a challenge

in terms of modelling (S

anchez-L

opez and Cerezo,

2019). To implement a HDT, the ﬁrst challenge is to

rely on a proper model of human, modelling each in-

ternal state together with their dependencies. A good

candidate can be found in (Shvo et al., 2019), where

Emotion, Motivation, Personality and Mood are taken

together with the State of an entity as inputs for deﬁn-

ing an action scheduler. In this model, each variable

is deﬁned from known theories and combined with at-

tention on regulation and feedback loops.

4 OUR HDT MODEL

Industrial 4.0 settings can be digitized according to

one single DT, or a set of different DT connected

together, representing different departments, shop

ﬂoors, machines or other systems. The ﬁrst case does

not ﬁt to HDT, because humans are considered like

other entities, from the inﬂuence they have on the

overall system that is twinned. We consider the sec-

ond approach, each time it make sense to have DTs of

individual elements, or entities, contributing to a same

objective. In a shop ﬂoor, this would be machines,

robots and humans involved in production tasks.

Following our deﬁnition, the HDT can be de-

signed as a subclass of DT, where the twinned physi-

cal entity is a human being. We have formalised an ar-

chitectural model, taking an information-centric per-

spective based on the interactions between the phys-

ical system - that can also be referred to as ”Physi-

cal Twin” (PT)- and the DT. In this model, the DT

is itself lying on interconnections between synchro-

nization, data management, and services layers, and

a mediator component manages the information ex-

changes between these layers. We show in Figure 1

only the parts that are relevant for the purpose of this

paper. Physical Entity refers to a physical entity that

is virtually mirrored and is the target of the twinning

objectives. It can be a Human Agent or Non-Human

Agent, and performs some Tasks in an Environment

where other entities evolve and with which it is in re-

lation. The physical entity is coupled with a Digital

Twin, which as a mirror of the physical space can also

be a Human Digital Twin or a Non-Human Digital

Twin. In this paper, our focus is the Internal Model

concept, representing the set of models embedded in

a DT, and more speciﬁcally the Entity Model, which

in the case of a HDT is the model of a twinned human.

The HDT’s entity model constitutes one of the

most challenging aspects, and it will often be limited

to the intended HDT’s scope. However, it remains

important to have a good awareness of all the human

Figure 1: Conceptual Model of DT and HDT concepts.

characteristics, to be sure neglected variables can ac-

tually be neglected in each use-case. In industry, one

of the objectives supported by the HDT would be to

allow a certain level of production (quality, quantity)

combining well-being of workers and performance.

This implies in particular taking into account compo-

nents of emotion through the detection of physiologi-

cal signals. Fatigue anticipation is one of the common

example. Another one is the detection of a detrimen-

tal level of stress, or any other factor leading to an in-

crease in error occurrences probability. But the other

dimensions can not be neglected: motivation varia-

tions, associated to the personality have also an im-

pact on productivity as well as the physical state of

the human worker.

We propose a preliminary (meta-)model of human

in Figure 2, relevant for simulating humans at work

with a HDT. It presents the main human character-

istics, abilities and states of a human agent in con-

text, performing a task according to a given demand,

and the inﬂuential links between the different vari-

ables impacting the worker behaviour. The main el-

ements from the works retained from the state of the

art are present, especially the three spheres of (Longo

et al., 2019), Belief, Desire and Intention elements

and the psychological characteristics driving the in-

ternal forces that will impact behaviour: Personality,

Emotion, Motivation, completed by Mood to account

for the work of (Shvo et al., 2019). In its current state,

the model focuses on variables inﬂuencing the emo-

tional state, and highlights the relevant theories and

models for each of the main ones. Variables driving

the physical state are also modelled, but not detailed.

IN4PL 2021 - 2nd International Conference on Innovative Intelligent Industrial Production and Logistics

142

Figure 2: Preliminary human model for HDT.

5 DISCUSSION

The ﬁrst interest of the HDT in industry is probably

to emulate human worker behaviors and interactions

with its surrounding, knowing this behaviour. From

the production perspective, it would ensure humans

can be integrated in predictive maintenance processes

like it is done now with machines and undesirable

events caused by human behaviour can be anticipated

for dynamic adaptation of the shop ﬂoor. From the

human perspective, it would ensure well-being and

quality of experience, with tasks tailored to workers.

In this context, the HDT is a step towards resilience

of industrial settings. Of course, this needs to be un-

derstood as a coarse illustration of the fact that HDT

can allow to maintain human’s well-being over long-

term and to anticipate any kind of deterioration of his

conditions - physical as well as psychological.

But the most interesting comes when consider-

ing the perspective it opens for control and supervi-

sion with collaborative decision-making among DTs.

HDTs could share their knowledge of the human

model and internal state with the non-human DTs, so

that they can integrate it in their own reasoning. This

leads to environments where collaborative decision-

making can be implemented, all DTs interacting to-

gether in the back to dynamically optimise the overall

objective by adapting machine commands and send-

ing warnings, instructions or recommendations to hu-

man workers. DTs and HDTs would then act together

as a system of autonomous agents, at any time re-

specting the constraints and preferences of their re-

spective twins, and trying to reach their own individ-

ual objective(s) while acting together to fulﬁll a com-

mon objective.

We have shown in this paper that the HDT is

a very young concept that can have a number of

functions in different domains, including supporting

health and medicine, being the keeper of personal

data, the embodiment of our digital shadow, or the

digital agent representing a human worker in indus-

try. It can be even much more. If the focus here

was industry, the necessity to have a proper human

model is domain-independent. Here we have high-

lighted the important factors to investigate and the

relevant theories and existing models, constituting a

basis for building a future model formalising not only

the structure but also the behaviour. For each of the

variables, the dedicated theories and models need now

to be carefully integrated. The model can then be pro-

gressively reﬁned and assessed on ﬁeld, until it be-

Human Digital Twin in Industry 4.0: Concept and Preliminary Model

143

comes precise enough. The challenge will be in par-

ticular to model properly the complex inﬂuences driv-

ing the internal forces.

When the model has reached an acceptable level

of accuracy, it can be further implemented as a soft-

ware agent integrated in a DT structure, to simulate a

human worker, learning and adapting from the worker

behaviour and synchronising with the ﬁeld, to reach a

state where emulation is possible. Of course the HDT

by nature carries all the concerns linked to data pro-

tection, acceptance and ethics, which we did not ad-

dress here. This is another story...

REFERENCES

Alam, K. M. and El Saddik, A. (2017). C2ps: A digital

twin architecture reference model for the cloud-based

cyber-physical systems. IEEE Access, 5:2050–2062.

Baskaran, S., Niaki, F. A., Tomaszewski, M., Gill, J. S.,

Chen, Y., Jia, Y., Mears, L., and Krovi, V. (2019).

Digital human and robot simulation in automotive as-

sembly using siemens process simulate: A feasibility

study. Procedia Manufacturing, 34:986–994. 47th

SME North American Manufacturing Research Con-

ference, NAMRC 47, Pennsylvania, USA.

Blake, T. (2016). The human’s digital twin. Professional

Engineering, 29(12):30 – 33.

Burden, D. and Savin-Baden, M. (2019). Virtual Humans:

Today and Tomorrow. Chapman & Hall/CRC Artiﬁ-

cial Intelligence and Robotics Series. CRC Press, Tay-

lor & Francis Group.

Case, K., Marshall, R., and Summerskill, S. (2016). Digi-

tal human modelling over four decades. International

Journal of the Digital Human, 1(2):pp. 112–131.

Costa, P. T. and McCrae, R. R. (1988). Personality in adult-

hood: a six-year longitudinal study of self-reports and

spouse ratings on the neo personality inventory. Jour-

nal of personality and social psychology, 54(5):853.

Hafez, W. (2020). Human digital twin: Enabling human-

multi smart machines collaboration. In Bi Y., Bha-

tia R., K. S. e., editor, Intelligent Systems and Applica-

tions. IntelliSys 2019. Advances in Intelligent Systems

and Computing, volume 1038. Springer, Cham.

Kamara-Esteban, O., Azkune, G., Pijoan, A., Borges,

C. E., Alonso-Vicario, A., and de Ipi

na, D. L. (2017).

Massha: An agent-based approach for human activity

simulation in intelligent environments. Pervasive and

Mobile Computing, 40:279–300.

Kennedy, W. G. (2012). Modelling Human Behaviour

in Agent-Based Models, pages 167–179. Springer

Netherlands, Dordrecht.

Kotseruba, I. and Tsotsos, J. K. (2020). 40 years of

cognitive architectures: core cognitive abilities and

practical applications. Artiﬁcial Intelligence Review,

53(1):17–94.

Lieto, A., Bhatt, M., Oltramari, A., and Vernon, D. (2018).

The role of cognitive architectures in general artiﬁ-

cial intelligence. Cognitive Systems Research, 48:1–3.

Cognitive Architectures for Artiﬁcial Minds.

Longo, F., Nicoletti, L., and Padovano, A. (2019). Modeling

workers’ behavior: A human factors taxonomy and a

fuzzy analysis in the case of industrial accidents. Int.

Journal of Industrial Ergonomics, 69:29–47.

Lu, Y., Liu, C., Wang, K. I.-K., Huang, H., and Xu,

X. (2020). Digital twin-driven smart manufactur-

ing: Connotation, reference model, applications and

research issues. Robotics and Computer-Integrated

Manufacturing, 61:101837.

Negri, E., Fumagalli, L., and Macchi, M. (2017). A re-

view of the roles of digital twin in cps-based produc-

tion systems. Procedia Manufacturing, 11:939 – 948.

27th International Conference on Flexible Automa-

tion and Intelligent Manufacturing, FAIM2017, 27-30

June 2017, Modena, Italy.

Nikolakis, N., Alexopoulos, K., Xanthakis, E., and Chrys-

solouris, G. (2019). The digital twin implementation

for linking the virtual representation of human-based

production tasks to their physical counterpart in the

factory-ﬂoor. International Journal of Computer Inte-

grated Manufacturing, 32(1):1–12.

Ortony, A., Clore, G. L., and Collins, A. (1990). The cog-

nitive structure of emotions. Cambridge university

press.

Peruzzini, M., Grandi, F., and Pellicciari, M. (2018). Ex-

ploring the potential of operator 4.0 interface and

monitoring. Computers & Industrial Engineering.

Russell, J. A. (1980). A circumplex model of af-

fect. Journal of Personality and Social Psychology,

6(39):1161–1178.

Saracco, R., Autiosalo, J., Kerckhove, D., Flammini, F.,

and Nisiotis, L. (2020). Personal digital twins and

their role in epidemics control: An IEEE digital re-

ality white paper.

Schmidt, F. L. and Hunter, J. E. (1998). The validity and

utility of selection methods in personnel psychology:

Practical and theoretical implications of 85 years of

research ﬁndings. Psychological bulletin, 124(2):262.

Shafto, M., Conroy, M., Doyle, R., Glaessgen, E., Kemp,

C., LeMoigne, J., and Wang, L. (2010). Draft model-

ing, simulation, information technology & processing

roadmap, technology area 11. Technical report, Na-

tional Aeronautics and Space Administration, NASA

Headquarters, Washington, DC 20546.

Shvo, M., Buhmann, J., and Kapadia, M. (2019). An inter-

dependent model of personality, motivation, emotion,

and mood for intelligent virtual agents. In Proceed-

ings of the 19th ACM International Conference on In-

telligent Virtual Agents, pages 65–72.

anchez-L

opez, Y. and Cerezo, E. (2019). Designing emo-

tional bdi agents: good practices and open questions.

The Knowledge Engineering Review, 34:e26.

Tao, F., Zhang, H., Liu, A., and Nee, A. Y. C. (2019). Dig-

ital twin in industry: State-of-the-art. IEEE Transac-

tions on Industrial Informatics, 15(4):2405–2415.

Zibuschka, J., Ruff, C., Horch, A., and Roßnagel, H.

(2020). A human digital twin as building block of

open identity management for the internet of things.

In Roßnagel, H., Schunck, C. H., M

odersheim, S.,

and H

uhnlein, D., editors, Open Identity Summit 2020,

pages 133–142, Bonn. Gesellschaft f

ur Informatik

e.V.

IN4PL 2021 - 2nd International Conference on Innovative Intelligent Industrial Production and Logistics

144