Content uploaded by Patrick Millot
Author content
All content in this area was uploaded by Patrick Millot on Jun 25, 2018
Content may be subject to copyright.
Designing Intelligent Manufacturing Systems through Human-Machine Cooperation
Principles: A Human-Centered Approach
Marie-Pierre Pacaux-Lemoinea,
1
, Damien Trentesauxa, Gabriel Zambrano Reyb, Patrick Millota
a LAMIH, UMR CNRS 8201, University of Valenciennes et Hainaut-Cambrésis, UVHC
Le Mont Houy, 59313 Valenciennes Cedex, France
{marie-pierre.lemoine, damien.trentesaux, patrick.millot}@univ-valenciennes.fr
b Department of Industrial Engineering, Pontificia Universidad
Javeriana, Bogotá, Colombia
gzambrano@javeriana.edu.co
Abstract. Since the start of industrialization, machine capabilities have increased in such a way that
human control of processes has evolved from simple (with mechanization) to cognitive (with
computerization), and even emotional (with semi/full automation). The processes have also evolved
from simple to complicated, and now complex systems. This is notably the case with Intelligent
Manufacturing Systems in which processes have become so autonomous that humans are unaware of
the processes running, while they may need to intervene to update the manufacturing plan or modify
the process configuration if a machine breaks down, or to assist process-intelligent entities when they
find themselves in a deadlock. This paper highlights the lack of attention paid to the correct
integration of humans in Intelligent Manufacturing Systems and provides solutions based on Human-
Machine Cooperation principles to retain humans in the process control loop with different levels of
involvement identified by the levels of automation. The aim of these principles is to propose a
human-centered approach to design and evaluate systems, processes, and their interactions with
humans. Herein, these principles are detailled and applied to Intelligent Manufacturing Systems
using Artifical Self-Organizing systems (ASO) as an example. An assistance system was designed to
support cooperation between ASO and human operators. Experiments were conducted to evaluate
the system and its utility in improving the performance of Human-Machine Systems, as well as its
acceptability with regard to human factors. The results presented highlight the advantages of the
approach.
Highlights:
Models to design and evaluate the performance of Human-Machine Systems are presented.
A human-centered approach is proposed to design human-aware Intelligent Manufacturing Systems.
The first results of the experiments relating to the usefulness of a human-centered approach are
encouraging.
Keywords: Techno-Centered Design, Human-Machine Cooperation, Human-Centered Design,
Levels Of Automation, Intelligent Manufacturing Systems.
1. Introduction
Human factors have been a focal point when designing transportation system processes. However,
with the rapid rise in the complexity of industrial systems and the organization of the humans
involved, human factors have been considered differently, usually in response to requirements
following negative experiences and industrial disasters (Seveso, Bhopal, AZF, Chernobyl…). In
France, the Ministry of Ecology, Sustainable Development, and Energy drew up an inventory of the
technological accidents that occurred in 2014 (Blanc et al., 2015). The report showed that in the case
1
Corresponding author
of automated process failures, technological accidents occurred in a specific context in which human
operators were directly involved in machine control, and indirectly involved in the design,
maintenance, and supervision of the human-machine systems.
The study also mentions that technological accidents are caused primarily by material failures (37%)
and human errors (63%). Regarding human errors, it highlights that “it is essential to understand the
organizational context that results in these primary causes”. In fact, two main causes are identified
and concern automated process running and design. It highlights several shortcomings in the training
and organization of human activity (26%), the adaptation of workstations to control the process
(15%), the tasks to be performed (23%), and the design of the process itself (30%). These results
underline the inappropriate use of human skills and the underestimation of the importance of training
in performing increasingly difficult and complex tasks. The study identified four types of human
error during accidents: no action (40%), appropriate but late action (31%), aggravating action (24%),
and inefficient action (5%). A recent study highlights that researchers often neglect the strengths and
weaknesses of human operators when designing innovative industrial control systems (Trentesaux &
Millot, 2016). Thus, despite the fact that human operators are systematically “in the control loop” of
industrial systems, researchers adopt a techno-centered approach, favoring the definition and
allocation of tasks to automated systems, only taking human operators into consideration at the end
of the design process once the control system has been defined. Furthermore, it is generally
considered that human operators will be there to handle any unexpected situations efficiently.
Consequently, to deal with a complex process that fails or behaves in a hazardous or even
unpredictable way, humans need to be simultaneously coordinated, efficient, and rapid, i.e. they have
to be “magic humans” (Trentesaux & Millot, 2016).
The objective of this paper is to encourage researchers involved in designing industrial control
systems, and more particularly manufacturing control systems - by far those with which there are the
most accidents (Blanc et al., 2015) - to address the aspect of human operators from the early stages
of the design process and to adopt a more human-centered approach. In this paper, the scope is
narrowed to a specific kind of innovative manufacturing system called “Intelligent Manufacturing
System” (IMS), which is justified:
First, by the fact that IMS are considered a promising paradigm along with current mature
technological solutions (Ambient Intelligence, Internet of Things, embedded systems…) and
concepts (Cyber-Physical Systems, Intelligent Products, Holons…) that potentially provide
more efficient, reactive, and resilient control mechanisms than classical manufacturing
systems (Van Brussel, Wyns, Valckenaers, Bongaerts, & Peeters, 1998).
Second, as IMS foster the use of more distributed decisional mechanisms. IMS control
systems are basically composed of self-organizing, autonomous, intelligent, and interacting
decisional control entities (e.g., modeled as holons). Consequently, there is more new
emerging and unpredicted behavior to be addressed. This justifies more the need to integrate
the human operator in the manufacturing control process in order to handle this behavior.
However, it makes it harder to define how to successfully integrate humans due to the lack of
predictability of such systems.
Continuing with the analysis of the current situation presented by (Trentesaux & Millot, 2016), this
paper deals with a human-centered design approach for IMS based on human-machine cooperation
principles. The approach was evaluated through real experiments on a full-size academic Flexible
Manufacturing System (FMS). Section 2 presents the current design of the industrial process,
underlining some problems in terms of IMS. Some solutions are presented in Section 3, which deals
with Human-Machine Cooperation principles and the design and evaluation of IMS. These principles
are then applied to design more cooperative, efficient interactions between humans and intelligent
manufacturing control processes. In Section 4, some experimental results are presented, along with
the advantages of a human-centered approach.
2. Towards a human-centered design approach in IMS
One of the main domains relevant to IMS concerns distributed scheduling in manufacturing control.
Research activities in this domain foster a paradigm that aims to provide manufacturing control
systems with more autonomy and adaptation capabilities through the functional or geographical
distribution of informational and decisional capabilities among artificial entities (e.g., modeled as
holons). This paradigm creates emerging "bottom-up" behavioral mechanisms complementary to
possible “top-down” ones generated by a centralized, predictive system to limit or to push this
emerging behavior evolving within pre-fixed bounds (Cardin et al., 2015). Holons are designed with
the ability to take into account other holon behavior, as well as the current environmental state. The
aim of such adaptation is to improve reactivity, resilience, and robustness of the global system
composed only of holons that choose their own trajectory. This type of entity is also linked to the
concept of “intelligent product” (McFarlane, Giannikas, Wong, & Harrison, 2013) and can be
integrated as part of a more global “cyber-physical system” (Lee, Bagheri, & Kao, 2015).
Researchers in IMS, and industrial systems in general, often consider the human operator as the
overall supervisor (Gaham, Bouzouia, & Achour, 2015), (Zambrano Rey, Carvalho, & Trentesaux,
2013). According to this approach, the human operator has to determine the objectives, as well as
tune the parameters, constraints, and rules in order to influence holon behavior. Even if industrial
system designers perceive this as a “human-in-the-loop” approach, in terms of human factors it can
be interpreted more as a way of keeping humans in the “decisional/responsibility loop” but not in the
“control loop”. This means that each individual decision in such a holonic system impacts the global
control of the system and assumes the withdrawal of human decisions: the human is excluded from
the decisions made by the holonic control system. Moreover, the human, as an external manager
never involved in the dynamics of the construction of holon decisions, is in fact unaware of the
system running and is unable to decide what would be the right rule or parameter to tune.
Figure 1. Techno-centered IMS design approach.
This design approach is “techno-centered” and priority is given to solving technical issues instead of
human aspects. A techno-centered approach consists in automating a maximum number of functions
(in nominal or degraded mode) in a pre-defined context and it is assumed that the human operator
will supervise and handle all the unforeseen situations in due course. Several approaches dealing
with the design of Decision Support Systems in manufacturing systems are paradoxically based on
such a “techno-centered” approach (MacCarthy, 2006; Trentesaux, Dindeleux, & Tahon, 1998). With
derision, (Trentesaux & Millot, 2016) call such a human operator the magic human. In this study,
Intelligent
manufacturing
control
command
observations
perfect
decisions
decision
needed
perfect observations
Production
Objectives
& constraints
Key
Performance
Indicators
product
service
performances
perfect
actions
perturbations
Manufacturing
system
Magic Human
“Human-in-the-loop” Intelligent Manufacturing Control Systems
Enterprise Resource Planning (ERP)
several examples, mainly from the domain of nuclear power plants, have been provided that
highlight the overestimation of human operator capabilities. It was also observed that the same risk
will emerge soon, if it has not done so already, in manufacturing systems. Figure 1 presents an
updated version of the initial work presented in (Trentesaux & Millot, 2016). In this figure, the
Enterprise Resource Planning (ERP) level is responsible for providing the intelligent manufacturing
control system with production objectives (production to be achieved) and constraints (energy
thresholds, etc.), while it gathers key performance indicators (e.g., energy consumption, production
lead times, and throughput).
The techno-centered approach overestimates the ability of the human operator who must behave
perfectly when desired and within suitable response times, as well as react perfectly in the face of
unexpected situations. But are human operators ready to face the unexpected with the complexity of
IMS? Conversely, what should the human operator do when an unexpected event occurs for which
no anticipated technical solution is available and the human is the only entity able to come up with
one?
2.1 Literature review on human-centered design in IMS
The literature on human-centered design in IMS is scarce and still emerging. Long-established
activities concerned the design of assistance systems for IMS (Erbe, 1998). Human-centered design
in IMS has been addressed more recently in some studies dealing with levels of automation (Dencker
et al., 2009; Frohm, Lindström, Stahre, & Winroth, 2008; Lindström & Winroth, 2010; Säfsten,
Winroth, & Stahre, 2007). In these studies, a method called Dynamo (Dynamic Levels of
Automation for Robust Manufacturing Systems) was proposed to measure the level of automation in
terms of both mechanical and informational work tasks. Levels of automation were then identified
according to the business and manufacturing strategy. Consequently, their goals were related more to
studying the balance between strategic objectives (global performance) and available levels of
automation, than to adapting levels of automation according to human factors. Other studies have
been conducted that consider how the human operator can be the key component to react quickly to
operational changes or urgent requests (Grosse, Glock, & Neumann, 2015), to handle the unexpected
(Valckenaers et al., 2011) along with a control system that embeds accurate models of reality
(mirroring process) (Valckenaers & Van Brussel, 2016). Finally, (Trentesaux & Millot, 2016)
proposed a human-centered approach based on a set of specifications to be satisfied when designing
“human-in-the-loop” IMS.
A growing number of researchers, especially in ergonomics and human engineering, are addressing
the domain of industrial engineering to ensure more human-centered manufacturing control system
designs. As a prerequisite, they have worked on Human-Machine Cooperation, Levels of
Automation and Situation Awareness, and Human-Automation Symbiosis (May et al., 2014; Millot,
2014; Romero, Noran, Stahre, Bernus, & Fast-Berglund, 2015). Some EU projects have also been
launched to illustrate this nascent field such as SO-PC-PRO (Subject Orientation For People Centred
Production) and MAN-MADE (MANufacturing through ergonoMic and safe Anthropocentric
aDaptive workplacEs for context-aware factories in Europe). Other research focuses on the
interaction between human operators, designers, and machines, and proposes ontological filtering
systems to provide a knowledge base that supports exchanges for design and control (Lepratti, 2006;
Nishitani, 1996). Interactions between human operators and machines, and between several types of
machine are studied in the same way for the purposes of harmonization to achieve more efficient
control of a flexible manufacturing cell (Solvang, Sziebig, & Korondi, 2012). Another approach
deals with studying a human operator model that could be used to improve human-machine
cooperation (Oborski, 2004). (Pires, 2005) highlights semi-autonomous aspects of industrial robotic
cells.
Concerning the industrial state of the art in this domain, numerous studies are underway to correctly
integrate the human operator in manufacturing processes, but they are being conducted mainly at
ergonomic level (Methods-time measurement MTM, etc.) and for low-level operational and
efficiency-oriented decisions (e.g., lean management). Upper levels, dealing with supervision,
scheduling, and control, etc., do not really apply a human-centered approach. This also holds true for
IMS, as there are few full-scale real implementations in this field of research.
2.2 Specifying a more human-centered design in IMS
The authors think that nowadays, particularly in IMS but also in industrial systems in general, it is
crucial to revise the basic research design patterns to adopt a more human-centered approach to limit
the hidden assumption of the magic human. This is obviously a huge, complex challenge. However,
this paper does not intend to provide absolute solutions, but rather aims to raise the alarm and to
provide some initial tentative models. These models are based on a set of facts that were identified in
the paper (Trentesaux & Millot, 2016) and are summed up in the following paragraphs.
The human can be the devil and unfortunately, some design engineers consider him/her to be just
that: his/her rationality is bounded (cf. “Simon’s principle”), he/she may forget, make mistakes,
over-react, be absent or even be the root cause of a disaster. For example, due to poor understanding
of the system behavior the human decides to switch to secure mode (Three Mile Island), or does not
do the right thing due to a lack of trust in the automated system or a lack of knowledge of the
industrial system (Chernobyl) (Schmitt, 2012). The controlled system must be designed accordingly
to manage this risk. A typical example of such a system is the railroad “dead-man's vigilance device”
where the driver has to regularly push a button so that the control system knows that he/she is really
in command. This is a first level of mutual control between the system and the human, which
assumes that the other is of limited reliability. More elaborate levels could address the issue of wrong
or abnormal command signals either from the human or the control system.
The human can be the hero, and we often believe that: he/she may save lives through unexpected,
innovative behavior (e.g., Apollo 13). The whole human-machine IMS must be designed to enable
the human to integrate unforeseen processes and mechanisms as much as possible. The human is also
an opportunist who can perceive and use unexpected information in such a way to improve the
current decision or action (Woods, Tittle, Feil, & Roesler, 2004).
The human can be the powerless witness: he/she may be unable to act despite being sure he/she is
right and the automated system wrong. The human can helplessly watch a disaster unfold due to the
design of an automated but poorly sized plant, for example (Fukushima) (Schmitt, 2012). The
automated part of the IMS control system must be designed to ensure it can be observed and
controlled by the human whenever desired.
The human is legally and socially accountable: the issue of allocating authority and responsibilities
between humans and machines is not so easy to resolve (e.g., automatic cruise control, air traffic
control, automated purchase procurement processes, automatic process scheduling). The designer
must consider this aspect when designing and allocating decisional abilities among manufacturing
control entities (Flemisch et al., 2012). In other words, if the human is accountable, he/she must be
allowed to fully control the IMS.
Based on these facts, some specifications to design a more human-centered IMS can be formulated.
Some of these are described in the paper by (Trentesaux & Millot, 2016), and are detailed in the
following paragraphs.
The human must always be aware of the situation: according to (Endsley, 1995), Situation
Awareness (SA) is composed of three levels: SA1 (perception of the elements), SA2 (comprehension
of the situation), and SA3 (projection of future states). Each SA level must, therefore, be considered
to ensure that humans can make decisions and continuously develop their mental models of the IMS
(e.g., to recover control or just to know what the situation is).
The levels of automation (LoA) must be suitable and adaptive: some tasks must be automated (e.g.,
automated manufacturing operations) but some cannot be. However, the related LoA must not be
predefined and definitively established. It must evolve according to situations, events, and human
and holon competences and capacities. Sometimes it must facilitate the work of the human (in
normal conditions, for example), and other times restore his/her control of critical tasks (when
abnormal situations occur in the IMS, for example). Consequently, the automated part of the IMS
control system must cooperate differently with the human according to the global situation. So, task
allocation must be dynamic, handled in an adaptive way, and justified.
The diversity and repeatability of decisions must be considered: typically, to avoid boring repetitive
actions/decisions as often encountered in manufacturing. This also requires the clarification, as much
as possible, of all the rare decisions for which the human is not prepared (e.g., emergency escape and
secure stopping of production, recovery procedure after major failure or hazardous situations). For
that, a time-based, abstraction-based hierarchy (e.g., strategic, tactical, and operational levels) and a
typology of decisions (e.g., according to skill, rule or knowledge-based behavior (Rasmussen, 1983))
can be defined.
Therefore, the human mental workload must be carefully regulated: according to some of the
previous principles, an “optimal workload” can be found between “nothing to do”, inducing a
potential lack of interest, and “too many things to do”, inducing stress and fatigue. A typical
consequence is that the designer must carefully define different time horizons (from real time to long
term), and balance the reaction times of the human with those of the automated part of the IMS. This
is one of the traditional issues dealt with by researchers in human engineering (Lemoine, M. P.,
Debernard, Crevits, and Millot, 1996), (Trentesaux, Moray, & Tahon, 1998).
Related to the previous point is the feeling of the need to cooperate. The need to cooperate may
appear when humans feel overloaded and need assistance. If the workload is below a given
threshold, the human does not perceive any need to cooperate and feels capable of performing the
task alone. If the workload increases or may increase over the next few minutes, the human needs to
cooperate despite the extra load due to the cooperation itself (Millot & Pacaux-Lemoine, 2013).
Lastly, an important human aspect concerns willingness to cooperate. Willingness to cooperate is
related to relations between humans, as well as between humans and assistance systems with regards
to self-confidence and trust (Millot & Pacaux-Lemoine, 2013). Willingness to cooperate can be seen
as an increasing function of trust in the system and a decreasing function of self-confidence. So, the
more the human trusts the assistance system, the more the human will be willing to cooperate. The
more the human is self-confident, the less the human wants to cooperate. This seems consistent with
a study conducted by (J Lee & Moray, 1992) on function allocations between humans and
automated control systems. Moreover, (Rajaonah, Tricot, Anceaux, & Millot, 2008) showed that
trust also plays a role in the interaction.
3. Proposal of a human-centered design framework in IMS
From these specifications, it is, of course, impossible to develop a generic model that complies with
all the previous principles for each possible case. Despite this, we can propose a human-centered
design framework for researchers in IMS, and industrial systems in general, with some ideas to limit
the magic human effect in their control systems. For this purpose, Figure 2 presents a global
framework that is an improved version of an initial partial framework presented in (Trentesaux &
Millot, 2016). This framework features some of the specifications introduced previously and is to be
compared with Figure 1. We detail here some aspects of the suggested framework relevant to these
specifications.
First, as suggested in one of the specifications presented previously, the structure is symmetric and
has been decomposed into 3 levels for each of the actors: operational (in the short term, directly
connected to command), tactical (at a higher hierarchical level to achieve intermediate objectives),
and strategic level (the highest decisional level, to plan intermediate objectives).
Cooperation between the human operator and the intelligent manufacturing control system is
achieved through exchange of information at strategic, tactical, or operational level. In addition,
decisions and requests are exchanged between two levels. For example, a strategic decision for the
human operator may be to increase the importance of energy consumption in the system. A
corresponding tactical decision for the human operator could be to delay some products to even out
energy consumption, while an operational decision would be to immediately turn on or off a robot
given the evolution of the instantaneous total energy consumption. Cooperation with the intelligent
manufacturing control system at each level would facilitate the definition of the new energy cost
(strategic level), the list of products to delay (tactical level), and the robot to be turned on/off. It is
important to note that due to possible fast reaction times at operational level, cooperation must be
limited at this level. However, this does not mean that the system is fully automated. We can
therefore consider system automation in 3 subsets. One subset is fully automated, but the human can
intervene if necessary to modify or correct conveyors and product positions. A second one is
partially automated, but feedback of experience enables procedures to be designed that the human
must follow (a kind of human automation). The last subset is neither automated nor foreseen and
therefore must be achieved using human inventive capabilities. This requires paying particular
attention when designing the whole system so that humans can give their very best, especially when
no technical solution is available.
Figure 2. Human-centered IMS design approach
Second, a key proposal is related to the integration of mutual observation (through the cooperation
process) that undertakes to investigate the limited reliability of either the human or the intelligent
manufacturing control system. For that purpose, each cooperating “agent” (the human operator or the
intelligent manufacturing control system) has a behavioral model of the other. Detailing these
behavioral models in depth is not within the scope of this paper but one can imagine different
solutions and ways to define such models. For example, the human operator’s behavioral model of
the intelligent manufacturing control system can be constructed through programming, learning, and
teaching experiences. The second behavioral model of the human operator integrated into the
intelligent manufacturing control system may contain knowledge about normal/abnormal locations
for the human operator to be in the manufacturing system depending on the actual location and tasks
of intelligent robots and AGV (automated guided vehicle), or about his/her current set of tasks with
their average completion time (including technics inspired from the dead man’s vigilance device),
etc. More elaborate behavioral models of the human operator may include a model of his/her
creativity, emotional behavior, stress, and trust (Martínez-Miranda & Pavón, 2010).
Some other specifications cannot really be represented in Figure 2 but are considered in the
suggested framework. This is typically the case for those dealing with Levels of Automation (LoA).
(Sheridan, 1992) proposed well-known guidelines based on 10 levels. At level 1, control is
completely manual while at level 10, it is fully automated. Intermediate level 4 corresponds to the
decision support system (the control selects a possible action and proposes it to the human operator).
At level 6, the control leaves the human operator a limited amount of time to counterbalance the
decision before its automatic execution. This can be specified for each level (strategic, tactical, and
operational). For example, nowadays it is conceivable that the Intelligent Manufacturing Control
system depicted in Figure 2 changes the operational decision level itself from levels 1 to 4 to level 10
because of the need to react within milliseconds to avoid an accident, while it leaves the tactical
decision level unchanged at an intermediate level. Researchers in aeronautics and driving addressed
Tactical level
Operational level command
observations
Product
Service
Behavioral model of the
intelligent manufacturing
control system
Strategic level
decisions
requests
Tactical level
decisions
Limited observat ions
Human operator
Behavioral model of
the human operator
requests
decisions
requests
Manufacturing
system
Strategic level
perturbations
perturbations
perturbations
Intelligent Manufacturing Control
“Human-in-the-loop” Intelligent Manufacturing Control Systems
Production
Objectives
& con straints
Key
Performance
Indicators
Enterprise Resource Planning (ERP)
Information
Exchange Information
Exchange
Operational level
decisions
requests
possible
actions
Information exchang e
Information e xchange
Information exchange
Information e xchange
Information
Exchange
adaptive automation, which could inspire research in IMS and industrial engineering in general
(Inagaki, 2003; Kaber & Endsley, 2004).
The suggested framework encompasses several aspects relating to cooperation, decision levels,
levels of automation, etc. As introduced in Figure 2, cooperation is at the heart of the framework and
is one of its key elements. Moreover, current technology enables innovative cooperation schemes to
be designed. For these reasons, the remainder of the paper focuses on this aspect.
3.1. Human-machine cooperation principles in the suggested framework
Work on Human-Machine Cooperation is very promising as current technology allows more and
more decisional capabilities to be embedded in both material (e.g., machines, products) and
immaterial (e.g., production orders) holons, transforming them into efficient assistance systems to
help humans enhance performance. In this context, it is suggested that each of these assistance
systems embed individual as well as cooperative capabilities. Such capabilities, and also tools, that
can support cooperation among holons and the human operator are now presented.
The principles of Human-Machine Cooperation (HMC) have been stipulated with a focus on
Dynamic Task Allocation, and generic concepts have been proposed to identify all kinds of
interactions between one or several human operators and one or several assistance systems (Millot &
Hoc, 1997; Millot & Mandiau, 1995). Human-Machine Cooperation principles have been applied to
several domains: air traffic control (Jean-Michel Hoc & Lemoine, 1998), robotics (Pacaux et al.,
2011) , fighter aircraft (M P Pacaux-Lemoine & Loiselet, 2002), and car driving (M.-P. Pacaux-
Lemoine, Simon, & Popieul, 2015). One objective of this contribution is to assess the usefulness of
such principles in the context of IMS. Stemming mainly from the global presentation of the Human-
machine Cooperation approach used in different domains in (Marie Pierre Pacaux-Lemoine, 2014),
this paper focuses on the use and adaptation of the models in IMS to design assistance systems.
Situations of cooperation may arise between at least two “agents”. In this paper, the term “agent” is
used in a general sense to define a decisional entity and denotes, for example, a human operator, an
intelligent control system (e.g., a single holon or a full holonic control architecture), or a Decision
Support System. The basic definition used to study cooperation is: “Two agents in a system are in a
situation of cooperation if (a) each strives to reach goals while interfering with the others’ goals (at
least regarding resources or procedures), and (b) they try to manage such interference to make the
others’ activities easier” (M.-P. Lemoine & Hoc, 1996). So, cooperative agents, interacting and
involved in controlling the same IMS process, perform both individual and cooperative tasks. The
abilities corresponding to each type of task are called “know-how” and “know-how-to-cooperate”,
respectively (Millot & Lemoine, 1998) (cf. Figure 3).
Figure 3. Model of a cooperative agent
Agent know-how (KH) is an agent’s ability to control the process without taking into account
exchanges with other agents, and can be divided into two parts: internal and external know-how.
Internal know-how relates to the agent’s skills and capacity to control the process and deal with
problem solving, i.e. analyzing situations and making decisions. Capabilities are related more to the
agents’ abilities to solve problems with regard to their experience and expertise. So, problem solving
can be based on the agent’s skills if the problem is well known, the agent’s rules if the problem is
clearly identified and predefined rules can be used, or the agent’s knowledge if the problem is new
and the agent has to analyze it to find a new solution (Rasmussen, 1983). Processing abilities relate
to the agents’ capacity to solve the problem. The solution is identified but agents can be overloaded
or have no more resources to implement it. External know-how is associated with the ability to
obtain information from the process and the ability to act on the process (i.e. agent input/output with
the process). This know-how enables tasks or functions to be identified, shared, and allocated to
agents in order to reach a common goal.
Agent know-how-to-cooperate (KHC) is also divided into two parts: internal and external. Internal
know-how-to-cooperate allows an agent to build a model of other agents (their know-how and know-
how-to-cooperate) in order to facilitate cooperation with them. External know-how-to-cooperate is
the ability to exchange information with others (i.e. input/output with others). Know-how-to-
cooperate enables interactions between agents to be managed and helps mutual understanding and
control.
KH and KHC can be seen as two functions used by agents in turn or at the same time, with varying
priority as regards the state of the system, type of activity, and agent’s workload for instance. Figure
4 focuses on a human and an assistance system working in cooperation on one level (whereas Figure
2 deals with three levels). KH functions are described according to the models proposed by
(Parasuraman, Sheridan, & Wickens, 2000; Rasmussen, 1983). KHC functions are based mainly on
the definition of cooperation (Jean-Michel Hoc & Lemoine, 1998). “Assistance system” is a generic
term used to deal with any kind of non-human agent; it can be a decision support system, a robot, a
holon, etc. In Figure 4, between the KH and KHC of the human and the assistance system, the light
blue rectangle represents the Common Work Space (CWS), which can propose a representation of
past, current, and future states of the process, as well as agents’ individual and cooperative activities
Know-how (KH)
Internal ability to solve problems (regarding the process)
capabilities: knowledge, rules, skills / experience, expertise
processing abilities: workload, fatigue, distraction…
External ability to:
get information (from the process and the environment)
act (on the process)
Know-how-to-cooperate (KHC)
Internal ability to:
build a model of other agents (KH and KHC)
deduce the other agents’ intentions
analyze the task and identify the cooperative organization
produce a common plan regarding tasks and coordination
External ability to communicate:
understanding other agents
providing other agents with information
(Marie-Pierre Pacaux-Lemoine & Debernard, 2002). It also facilitates the exchange of information
between agents and helps improve team situation awareness (Millot & Pacaux-Lemoine, 2013).
Figure 4. Human-Machine cooperation model based on (Marie Pierre Pacaux-Lemoine, 2014)
According to the capabilities of the humans and the assistance systems, they can cooperate to “gather
information”, “analyze information”, “make decisions”, and “implement actions”. Such sub-
functions can be shared, traded, and combined by the KHC function control via the slider on each
scale displayed in the CWS. The closer the slider is to the human side, the greater the authority of the
human and the greater the responsibility for the sub-function. The closer the slider is to the assistance
system side, the greater the authority of the assistance system to control the sub-function. However,
it is still currently difficult to deal with assistance system responsibilities. Most of the time,
assistance system designers are not concerned with these responsibilities. The position of the slider is
managed as a result of KHC functions, meaning that the human as well as the assistance system can
accept, refuse or negotiate the proposal of the other (M.-P. Pacaux-Lemoine & Debernard, 2002).
Figure 4 presents cooperation between a human and an assistance system with the same objective.
However, this representation can also be used to study or design cooperation among several humans,
among several humans and assistance systems (as in a holonic manufacturing control architecture) or
even among several assistance systems (usually decision support systems at tactical and strategic
levels and holons at operational level). This representation can be used to study and design
cooperation between levels of activity or hierarchical levels (cf. Figure 5).
Performance
Assistant system
Objectives
Constraints
Human
Interface
Manufacturing system
Decision
making
Information
analysis
Information
gathering Action
implementation
Decision
making
Information
analysis
Information
gathering Action
implementation
Human Know-How
Assistance system Know-How
Info. gathering on
the other agent Interference
detection Interference
management Cooperative
action
Info. gathering on
the other agent Interference
detection Interference
management Cooperative
action
Assistance system Know-How-to-Cooperate
Human Know-How-to-Cooperate
CWS
Raw data Command
Figure 5. Multi-level cooperation based on (M. P. Lemoine et al., 1996)
Humans and assistance systems can cooperate with regard to each sub-function and the
complementary objectives of each level of activity. These can be called the horizontal extension and
the vertical extension, respectively, of the shared control concept in (M.-P. Pacaux-Lemoine & Itoh,
2015).
3.2 Implementation of the suggested Human-Machine cooperation approach in the proposed
framework
Human-machine cooperation principles considering multi-level aspects can be extremely helpful to
identify current human and system capabilities in IMS and to propose improvements with regard to
those capabilities. The following example stems from a study conducted subsequent to the
publication of a set of initial ideas suggested by (Zambrano Rey et al., 2013). Herein, the idea is to
evaluate the impact of a human-centered design on manufacturing system performance (e.g., energy
used, production completed in a fixed time…) compared to a techno-centered design. Other
performance indicators related to human aspects (workload, ease of use…) will also be assessed.
This study was based on the specific artificial self-organizing intelligent control of a flexible
manufacturing system (FMS). Artificial self-organizing (ASO) systems are a possible way of
implementing IMS principles and are considered in the remainder of this paper. They have been
selected for their potential reactivity in the face of human interventions. The ASO manufacturing
control system (ASO-MCS) proposed is based on holonic and bio-inspired paradigms (Barbosa,
Leitão, Adam, & Trentesaux, 2015). In a techno-centered design approach, the system interprets
human intervention as an external disturbance. However, since the behavior of the ASO is highly
reactive and hard to predict by the human, the human-centered approach must address the following
questions: How can the human supervise the ASO-MCS? How can the ASO-MCS perceive the
intervention of the human as help and not a disturbance?
We tried to answer these questions by implementing a technical solution of the proposed human-
centered design framework with the Human-Machine cooperation principles presented previously.
Instead of focusing on ASO and operational level, three levels of activity were analyzed to assist the
humans and manage interactions between levels. Figure 6 details the strategic, tactical, and
operational levels. Each level is described according to the cooperation model presented in the
previous section. In this example, we consider that the humans present at each level are different
people, but it could be the same person depending on the size of the ASO and the organization of the
manufacturing system.
Level
Strategic level definition of the plan
Tactical level application of the plan by triggering tasks
Operational level control of the process
Agent x1Agent xn
Agent y1Agent yn
Agent z1Agent zn
Agent z2
Agent y2
Cooperation inside level
Cooperation
between levels
Number of agents
Figure 6. Multi-level cooperation in IMS
At strategic level, the human knows or defines the global objectives of the manufacturing system in
terms of performance to be attained using information provided by the ERP. In the example, we try
to control the number of products and energy consumption, i.e. the human’s strategic KH. Based on
these objectives, a strategic decision assistance system proposes a manufacturing plan to the human;
this is the system’s KH. The human and the assistance system exchange information via the
Common Work Space (CWS), a visual interface for exchanging information between each agent
involved in the manufacturing management system. The CWS is global at strategic level, but focuses
in more detail on smaller parts of the manufacturing system at tactical level and even more so at
operational level. At strategic level, the CWS allows the human to give the decision assistance
system objectives. However, to be able to do this the human needs a model of the assistance system;
this is his/her KHC. Through this ability to cooperate, he/she can provide information that the
assistance system can compute correctly. Conversely, the assistance system provides its analysis and
decision making, and displays a set of decisions via the CWS. The assistance system also needs a
KHC in order to display decisions that are visible, understandable, and acceptable by the human.
At tactical level, the human and the assistance system know the manufacturing plan provided at
strategic level. The local objectives of the human are to supervise the progress of the plan controlled
by the ASO-MCS, and so the human is expected to redress or update the manufacturing plan after a
negative trend. This is the human’s tactical KH. Nevertheless, as explained above, ASO systems are
complex and highly reactive systems that humans may understand little and manage without
stopping the process. In order to avoid such an extreme situation, an ASO-MCS with a tactical
assistance system able to use the current ASO state and provide decisions is presented in Figure 7. In
the ASO-MCS proposed, the ASO is comprised of products and machines. Products are modeled as
holons (PH1,…PHn) that can decide for themselves on which machine holon (MH1,…,MHm) their
Manufacturing system
KH: Modification/correction of conveyors or products
KH: Definition of commands for each product
CWS: Subparts of Self-Organizing System
Operational level
Functional
Objectives Conveyors and products
Output Input /
Command
KH: Supervision of plan progress
KHC: Providing orders
KH: Analysis of Self-Org. System and proposal of potential fields
KHC: Sharing and explanation of decisions
CWS: Self-Organizing System
Tactical level
Local objectives
Constraints Potential fields
modification
Orders / Decisions
Requests
Global objectives
Constraints
Strategic level
KH: Definition of plan
KHC: Providing objectives and understanding system decisions
KH: Proposal of plan
KHC: Providing set of decisions
CWS: Global manufacturing system
Number of products
Energy consumption
Orders /
Manufacturing plan
Requests
Human A
Strategic Decision Assistance System
Human B
Human C
Tactical Decision Assistance System
Self-Organizing System: Holons
next operation will be processed. In addition, PHs can dynamically decide which routing path,
composed of successive conveyor segments (and nodes), will be used to reach the selected MH.
These decisions are based on the evolution of a specific set of variables propagated through the
conveying system that represent machine attractiveness, and indicate the routes from each node to
any machine. The design of these variables is based on the potential fields (PF) emitted by machines:
the greater the potential field, the more attractive the machine or the conveying segment propagating
the field. The value of the potential field is sensed by the products and helps them to choose their
next machine/direction. The PF increases or decreases dynamically, depending on the status of the
machine for example (available, unavailable, busy…). Potential fields are propagated through the
routing network and mitigated depending on the distance (the farther the machine, the less interesting
it is for a product; the busier the routing path, the greater the mitigation). The reader can find more
information in (Pach, Berger, Sallez, & Trentesaux, 2015).
Figure 7 describes the interaction between the ASO, the assistance system, and the human operator at
tactical level. The assistance system contains an emulation of the manufacturing system to be
controlled. The emulation is a “mirror” of reality and so it can help the human supervise the progress
of the plan (tactical level, KH in Figure 6). This will help the human make the right decisions and
issue the right orders that will be applied to the real FMS through the real MCS (tactical level, KHC
in Figure 6). For this purpose, the assistance system enables possible decisions to be tested using the
emulator. Hence, by updating the product state (PSn) and machine state (MSm) in the emulator, the
human operator can use the assistance to improve his/her understanding and to explain the behavior
of the ASO (tactical level, KHC in Figure 6), before actually making a decision in the real MCS. The
possible decision of the operator is an input parameter of the assistance system and an estimation of
the Objective Functions (OF) provides the human with valuable information via the CWS. The
human’s KHC is to provide the assistance system with the right information. The assistance system’s
KHC is to provide decisions that meet the human’s requests and to justify them in a way that the
human can understand and accept. Therefore, as presented in Figure 6, the main purpose of the ASO-
MCS with a tactical assistance system is to enable the analysis and comprehension of the self-
organizing system and, in this case, the possibility of using potential fields to help the system recover
after a perturbation. Consequently, the human’s interaction with the tactical assistance system allows
him/her to make more informed decisions before actually making them in the real system.
Figure 7. The ASO-MCS with tactical assistance
At operational level, the human shares the physical area with the ASO-MCS components. Hence, the
consequences of any decision or action made by the human will be reflected immediately. For
instance, the human’s functional objectives are to intervene if a problem occurs that is likely to cause
a deadlock such as a mechanical or electrical problem with conveyors, machines, or products. The
human, therefore, must act because the ASO-MCS cannot manage such an issue by itself. The ability
of the human to address such specific issues in the right way is his/her KH. The human, therefore,
focuses on one or a few elements of the ASO-MCS (e.g., a machine holon under maintenance), and
is unaware of the global functioning of the ASO-MCS. At this level, without assistance, the human
has no KHC to help him/her intervene on part of the ASO-MCS without disturbing it because he/she
is unable to build a representation of such a reactive system (which means that in Figure 2 there is no
“Behavioral model of the intelligent manufacturing control system” box in this situation). From the
ASO-MCS point of view, the same situation may occur. Given the distributed decision-making in the
ASO-MCS, the system does not have a global view of itself and does not give the human a global
representation of itself. In addition, the ASO-MCS may consider the human’s actions as a
disturbance, and it cannot take him/her into account in a positive way (which means that in Figure 2
there is no “Behavioral model of the human operator” box in this situation). The absence of
cooperation at this level may be resolved by an assistance system and cooperation at tactical level.
The details of the assistance systems used during the short experiments are provided in the following
section.
4. Initial steps to test the approach
The experiments were conducted on the flexible manufacturing system at the University of
Valenciennes and Hainaut-Cambrésis (named AIP-PRIMECA FMS). This system is an assembly cell
with conveyors on which products, fixed to moving shuttles, reach machines in order to obtain
manufacturing services (cf. Figure 8).
PRODUCTS CONVEYOR MACHINES
FMS
MH1
MH2
MHm
PH2
PH1
PHn
Potential
fields
FMS Emulator
Input module
Output module
(CWS)
FMS structural
model
Product and
machine holons
(Behaviors)
Global Objectives
(eOF, aPF(MHm))
FMS status ASO
Human Operator
MCS
FMS actual
performance (CWS)
Machine
control
rPF(MHm)
FMS real-time data
Assistance System
(rOF1,rOF2)
(PSn,MSm)
FMS actual
status
Off-line
H-M data flow (online)
Data flow (online)
PF controller Human choicevalidation (online)
eOF : estimated obj ective function
aPF(MHm): Potentialfield
Recommendation for the mth
machine holon
rPF(MHm): Potentialfield set to
the mth machine holon
PS :Product state
MS: Machine state
rOF1: Actual value of objective
función 1
rOF2: Actual value of objective
función 2
PH :Productholon
MH: Machine holon
Acronyms
Figure 8. The flexible manufacturing system (FMS)
As introduced in the previous section, products are physically self-organizing holons composed of a
physical and an embedded informational part. As autonomous entities, they choose their own
trajectory according to potential fields. In the experiments, the human modified the potential fields
emitted by the resources to control product trajectories and targets. The tactical level was tested in
this paper. The following section describes the tactical decision assistance system implemented.
4.1. Description of the tactical decision assistance system implemented
As seen in Figure 7, the tactical assistance system here is composed of an instantiated emulator of
the manufacturing system to be controlled, an input module, a potential field controller, and an
output module (CWS). The emulator comprises three parts: the FMS structural model, the decisional
entities’ behavior, and the current status of the FMS. The FMS structural model corresponds to that
of the AIP-PRIMECA FMS (Figure 8 and bottom-right in Figure 7), and models physical
components, physical constraints, and managerial constraints. The decisional entities are created to
reflect the decision-making algorithms and relationships between products and machine holons
(ASO) as they are implemented on the AIP-PRIMECA FMS. Lastly, the current status of the FMS is
composed of a set of FMS parameters that may change according to the FMS conditions, e.g.,
machine availability, conveyor segment availability, and length of waiting lines. Hence, the first two
modules are not connected to the real FMS but are offline so the emulator can be updated if the AIP-
PRIMECA FMS changes. The third module requires the ASO-MCS to be connected so the emulator
can provide valid information. Nevertheless, an error of about 15% between the AIP-PRIMECA
FMS status (hardware) and the emulated FMS was accepted because of the stochastic behavior of
products and the lack of certain technical constraints (e.g. friction) considered in the emulator.
Through the input module, the human provides goals (system throughput and energy consumption as
introduced earlier). Such goals are transferred to the potential field controller in which a
proportional-integral controller was implemented to calculate the best recommended potential field
for each machine (aPF(MHm) in Figure 7) and to estimate the objectives (cf. eOF in Figure 7)
according to the requested goals. The tactical assistance system proposes such values to the human
by means of the output module (cf. the left side of Figure 9). The output module is a frame of
reference common to the human and the tactical decision assistance system, and it is used to design
n1
M4
M2
M3
n2n3n4n5
n6
n7
n8
n9
n10
n11
Legend Machine node Routing node
M5
M6
M7
KUKA
Robot 1
KUKA
Robot 2
KUKA
Robot 3
Automated
Inspection unit
Manual
Recovery unit
Optional
Workstation
Supply
Storage
areas
Job input storage area
M1
Loading/
Unloading unit
Shuttle
Storage
area
Plates
Products
the tactical level CWS. In the emulator, the human can also choose not to use the PF controller and
validate its own choices (cf. sliders in Figure 9). This can help the human to observe the
consequences of his/her decisions before interacting with the real FMS. Following this test, the KH
of the tactical decision assistance system was composed of the emulator and the evaluation of the
entire ASO-MCS according to the machines’ potential field values.
Figure 9. Common work space (CWS) in the tactical assistance system (the output module is highlighted by the red rectangle)
The human calculated the potential fields using the emulator, with or without assistance, then
cooperated with the real AIP PRIMECA by assigning the potential fields to the real machines
exploited by the machine holons via the ASO-MCS (rPF(MHm) in Figure 7). The consequences of
such decisions were then displayed by the actual values of the objective functions (rOF in Figure 7).
With this organization, the human is aware of the status, performance, and processing of the ASO-
MCS.
4.2. Presentation of the experimental protocol
For this experimental protocol, the real FMS was not used, so the human operator was only
confronted with the emulator. The tactical decision assistance system was evaluated through short
experiments conducted with one participant. It was then, of course, impossible to conduct a statistical
analysis. However, our primary objective was to validate a complete feasible Human-Centered
Design approach based on the suggested framework and focusing on the issue of cooperation. For
this purpose, three experiments were designed:
Experiment 1 was performed with a techno-centered approach in which no human operator
was involved.
Experiment 2 was performed with a techno-centered approach. A human operator was
involved but no recommended potential fields were provided; the human made his own
choices based on his own perception of the environment.
Experiment 3 was performed with a human-centered approach. Recommended potential
fields from the PF controller were provided by the tactical decision assistance system and the
human was able to use these values or make his own choices.
The scenarios were performed by the participant. Subjective and objective results were obtained
from the data analysis.
Subjective data deal with workload evaluation and were collected using online and offline
questionnaires. The online questionnaire consisted in asking the participant to evaluate his workload
every 3 min on a scale from 0 and 7. The offline questionnaire consisted in using the NASA-TLX
method proposed by (Hart S. G. & Staveland L. E., 1988). The participant had to evaluate his
workload according to 6 different sources. The participant had to place a mark on each scale that
represented the magnitude of each factor of the task already executed (cf. Figure 10).
Figure 10. NASA-TLX rating scales
The next step of this method was to select the member of each pair that provided the most significant
source of task workload variation (cf. Figure 11).
Figure 11. NASA-TLX pair-wise comparison of factors
Objective data were calculated according to the expected performances of the manufacturing system,
namely throughput and energy consumption.
The experimental protocol consisted in presenting the project objectives to the participant
(objectives, ASO-MCS, potential fields, FMS emulator) (10 min), and then training the participant to
Mental demand
Physical demand
Temporal demand
Own performance
Effort
Frustration level
Low High
Low High
Low High
Good Poor
Low High
Low High
Physical demand / Mental demand Temporal demand / Physical demand Temporal demand / Frustration level
Temporal demand / Mental demand Own performance / Physical demand Temporal demand / Effort
Own performance / Mental demand Frustration level / Physical demand Own performance / Frustration level
Frustration level / Mental demand Effort / Physical demand Own performance / Effort
Effort / Mental demand Temporal demand / Own performance Effort / Frustration level
use the system (15 min). The first scenario (Experiment 2) was conducted without the assistance
system (10 min), then the NASA-TLX questionnaire was completed (5 min). The second scenario
(Experiment 3) was conducted with the assistance system (10 min), and then the NASA-TLX
questionnaire was completed (5 min). Two scenarios with similar levels of difficulty were
developed. In each scenario, 6 machines could be used to successively perform operations on
products, and the machines could perform one or several types of operation. Three machines were
active at the beginning of the experiment; three others were inactive but available in case of machine
breakdowns. Unexpected events were triggered in order to make the ASO-MCS unstable and thus
necessitate human intervention. The unexpected event consisted in one machine breaking down in
each scenario. Therefore, the human had to select another machine according to its capabilities (type
of operation) and adjust the potential fields of all the machines, including the backup machine, to
reach production goals. In terms of productivity, herein, the purpose was not to maximize the
throughput but to keep it bounded between a lower limit (4 products per 300s) and an upper limit (6
products per 300s). Keeping the throughput bounded helps to keep energy consumption under
control, for which a maximum level was set to 20 kWh.
4.3. Results
A 2nd year Master’s student from the University of Valenciennes and Hainaut-Cambrésis conducted
the experiment. This participant had no previous training on the flexible cell. The results presented in
this section were obtained following completion of the scenario and filling out the questionnaires.
Figure 12. Objective results: throughput
The results obtained from the number of products manufactured (throughput) underline the
usefulness of the assistance system and the human-centered approach (Experiment 3). Figure 12
underlines that when the ASO-MCS is completely autonomous (Experiment 1) and a machine breaks
down, the ASO-MCS can reorganize itself but reaches its limit because idle machines are not used
since the PF is set to zero. Hence, the remaining machines available cannot maintain the throughput
within the limits. When the human is in the loop, the human is aware of the situation and starts a new
machine and/or modifies the machines’ potential fields. Nevertheless, without the suggested tactical
decision assistance (Experiment 2), the human is unable to maintain the indicator within the limits,
Production time (Sec.)
Throughput
Throughput without human (Experiment 1)
Throughput with human and techno-centered approach (Experiment 2)
Throughput with human and human-centered approach (Experiment 3)
as the number of products and thus the energy consumption threshold are exceeded. The best results
were obtained when the human was present and assisted (Experiment 3).
Figure 13. Objective results: energy consumption
Since the throughput and the energy consumption can be two opposite indicators, particular attention
was paid to energy consumption. Energy consumption is mainly considered according to machine
usage. Hence, when a machine breaks down, the first reaction is to start one or several idle machines.
However, a better approach might be to balance machine usage by modifying the potential fields of
currently available machines instead of turning more machines on. This second choice is not obvious
to the human, and the results in Figure 13 highlight the help provided by the assistance system in the
human-centered approach. Most of the time, the desired consumption limit (red line in Figure 13) is
exceeded without the assistance system (Experiment 2) but respected with the assistance system in
the loop (Experiment 3).
These results derived from the objective data highlight that the manufacturing system is much more
stable and the objectives are attained when the human is in the loop and helped by the cooperative
assistance system. However, even if performance is improved with the human-centered approach,
human factors such as workload must show that this good performance is due to good interactions
with the process, assistance system, and holons. In these experiments, the quality of the interaction is
given by two sets of subjective data evaluated as follows.
The first set of subjective data was obtained from the analysis of the completed NASA-TLX
questionnaire. The calculation of the global workload is based on equation (1). Ri stems from the
“NASA-TLX rating scales” questionnaire (Figure 10). Ri represents the position of the mark on the
100 millimeter line of the question linked to the workload factor “i”. αi stems from the “NASA-TLX
pair-wise comparisons of factors” questionnaire (Figure 11). αi is the number of occurrences of the
workload factor “i” proposed by the questionnaires.
(1)
The global workload calculated using the NASA-TLX method is 42% without the assistance system
(Experiment 2) and 24.67% with the assistance system (Experiment 3). This drop by almost half is
mainly due to a decrease in mental demand as a result of the help provided by the assistance system
to anticipate the future behavior of the ASO, an increase in performance due to an increase in self-
Energy consumption with techno-centered approach (Experiment 2)
Energy consumption with human-centered approach (Experiment 3)
Production time (Sec.)
Watt per hour
confidence and trust in the assistance system, and a decrease in effort because the assistance system
performs part of the task.
The second set of subjective results addresses the participant’s online evaluation of the workload.
Figure 14 again underlines a quicker decrease in the workload using the assistance system. The high
value for the “with assistance system” condition (Experiment 3) at 300 and 600 seconds could be due
to the amount of training the participant received at the beginning of the experiment, which was
probably insufficient.
Figure 14. Subjective results: online evaluation of workload
4.4. Discussion
The objective of these experiments was to compare the benefits of the human-centered design
approach with the Techno-centered design approach to control an ASO-MCS-based intelligent
manufacturing system. The objective and subjective results underline the trend that a human-
centered design approach improves the global performance in the face of complex and conflicting
production objectives described in terms of energy consumption and throughput, as well as human
operator workload. Human-Machine Cooperation principles help to identify the elements that can
assist the human operator in controlling the IMS and more particularly, in the experiments, the ASO-
MCS. Indeed, the activity of the human operator today was analyzed and compared with the activity
he/she would have in the future with the ASO-MCS. This comparison highlights the necessity to
provide an assistance system that can assist ASO-MCS control when there are deadlocks or machine
breakdowns. This tactical assistance system was equipped with KH to analyze and make decisions
regarding the ASO-MCS, and KHC in order to transmit its analysis to the human operator. The
detailed model presented in Figure 4, and its adaptation to IMS and the multi-level aspects presented
in Figure 6, show the elements we needed to take into account as well as their interactions. So, the
trend towards an improvement in performance is due to a good balance between what the ASO-MCS
is able to do at operational level and what the human operator is able to do when assisted at tactical
level. Human operators and ASO take advantage of each other if the Human-ASO system is well
defined and cooperative. In addition, the trend towards a decrease in the human operator workload is
due to the ASO model the human operator was able to build owing to the ASO KHC and the
information it provides via the CWS. Nevertheless, ASO KHC is not currently able to model human
operator KH and KHC, which is a challenging perspective among others presented in the next
section.
0
1
2
3
4
5
6
0 300 600 900 1200 1500
Online workload
Seconde
Techno-centered approach (Experiment 2)
Human-centered approach (Experiment 3)
5. Conclusion and future research
The aim of this paper was to raise awareness of the risks linked to the greater complexity of
intelligent manufacturing systems, as well as removing humans from the control loop. Current
research has highlighted the impact of such risks through examples, and specific human factors have
been pointed out to control such risks. Nevertheless, some advanced research has already tackled
these aspects and several studies have been presented. Among these studies, one deals with
principles to develop human-system cooperation based on a human-centered design framework. This
approach has been detailed and applied to Intelligent Manufacturing systems. An assistance system
has been designed to help humans when they have to supervise and control an artificial self-
organizing manufacturing control system. The results of the experiments conducted on the flexible
manufacturing system at the University of Valenciennes and Hainaut-Cambrésis have been presented
and underline the positive contribution of human-machine cooperation in the design and supervision
of self-organizing manufacturing control systems.
The first aspect is to improve the experimental conditions. These initial results obviously need to be
reinforced through sound statistical studies by conducting more experiments with more participants.
The type of participant could also be an experimental variable to be investigated. Indeed,
experiments could be conducted with novice or experienced human operators, trained or not in the
use of ASO-MCS and assistance systems. Another experimental aspect to be improved is the realism.
In these experiments, the real FMS was not used and hence the human operator was only confronted
with the emulator. The next step would be to conduct experiments on the real FMS with several new
dimensions to control such as the size of the platform and the fact that the human operator does not
have an overview of the complete FMS (except with the assistance system), the real-time constraints
(controlled with the emulator), and technical problems with conveyors, shuttles or machines that
were avoided with the emulator but which are real aspects that need to be taken into account with a
real FMS. The last improvement concerns the evaluation method. A cooperative agent model can
also be used to evaluate interactions between a human operator and assistance systems by coding the
individual and cooperative activities of each one. KH and KHC models can code the activity during
experiments but also after experiments using a self-confrontation method (Jean-Michel Hoc &
Lemoine, 1998). Comparison of coding from several experiments may highlight interesting
information such as human operator misunderstandings or difficulties in using a new system.
A second aspect concerns improvements to the assistance system. In these experiments, assistance
system KHC, which provides the human operator with information through the common work space,
was limited. This KHC could be improved if the assistance system is able to build a representation of
the human operator. Of course, this aspect must be achieved using a multidisciplinary approach. The
assistance system could learn from human decision making, have information regarding his/her
workload, his/her position in the FMS… information that it could use to adapt its behavior to the
human operator’s needs or errors. Such a human operator representation may be used to anticipate
risks and to manage them through better human-machine task allocation. This must be done from an
organizational point of view but also in real time in order to avoid accidents.
In conclusion, in the context of sustainable development, this paper has highlighted the importance
of human-centered designs in IMS and the main ambition of the authors is to foster research in this
context.
Acknowledgement
The authors would like to acknowledge T. Bonte, T. Laurain, and S. Boulnois for their contributions
to the system design and evaluation.
References
Barbosa, J., Leitão, P., Adam, E., & Trentesaux, D. (2015). Dynamic self-organization in holonic
multi-agent manufacturing systems: The ADACOR evolution. Computers in Industry, 66, 99–
111. http://doi.org/10.1016/j.compind.2014.10.011
Blanc, P., Normand, A., Robert, C., Gauthier, A., Michel, J.-F., Pitrat, D., … Truchot, M.-J. (2015).
Inventaire 2015 des accidents technologiques. Ministère de l’Écologie, du Développement
durable et de l’Énergie, Direction générale de la prévention des risques.
Cardin, O., Trentesaux, D., Thomas, A., Castagna, P., Berger, T., & Bril El-Haouzi, H. (2015).
Coupling predictive scheduling and reactive control in manufacturing hybrid control
architectures: state of the art and future challenges. Journal of Intelligent Manufacturing, 1–15.
http://doi.org/10.1007/s10845-015-1139-0
Dencker, K., Fasth, Å., Stahre, J., Mårtensson, L., Lundholm, T., & Akillioglu, H. (2009). Proactive
assembly systems-realising the potential of human collaboration with automation. Annual
Reviews in Control, 33(2), 230–237. http://doi.org/10.1016/j.arcontrol.2009.05.004
Endsley, M. R. (1995). Toward a theory of situation awareness in dynamic systems: Situation
awareness. Human Factors, 37(1), 32–64.
Erbe, H.-H. (1998). Assistance systems supporting operations in manufacturing. In Proceedings of
the BASYS - 3rd IEEE/IFIP International Conference on Information Technology for Balanced
Automation Systems in Manufacturing (pp. 447–456).
Flemisch, F., Heesen, M., Hesse, T., Kelsch, J., Schieben, A., & Beller, J. (2012). Towards a
dynamic balance between humans and automation: Authority, ability, responsibility and control
in shared and cooperative control situations. Cognition, Technology and Work, 14(1), 3–18.
http://doi.org/10.1007/s10111-011-0191-6
Frohm, J., Lindström, V., Stahre, J., & Winroth, M. (2008). Levels of Automation in Manufacturing.
Ergonomia - an International Journal of Ergonomics and Human Factors, 30(3), 1–28.
Gaham, M., Bouzouia, B., & Achour, N. (2015). Human-in-the-Loop Cyber-Physical Production
Systems Control (HiLCP2sC): A Multi-objective Interactive Framework Proposal. In T.
Borangiu, A. Thomas, & D. Trentesaux (Eds.), Service Orientation in Holonic and Multi-agent
Manufacturing (pp. 315–325). Cham: Springer International Publishing.
http://doi.org/10.1007/978-3-319-15159-5_29
Grosse, E. H., Glock, C. H., & Neumann, W. P. (2015). Human factors in order picking system
design: A content analysis. IFAC Proceedings Volumes (IFAC-PapersOnline), 48(3), 320–325.
http://doi.org/10.1016/j.ifacol.2015.06.101
Hart S. G., & Staveland L. E. (1988). Development of NASA-TLX (Task Load Index): Results of
empirical and theoretical research. In P. A. Hancock and N. Meshkati (Eds.) Human Mental
Workload. Amsterdam: North Holland Press.
Hoc, J.-M., & Lemoine, M.-P. (1998). Cognitive Evaluation of Human-Human and Human-Machine
Cooperation Modes in Air Traffic Control. International Journal of Aviation Psychology, 8(1),
1–32. http://doi.org/10.1207/s15327108ijap0801_1
Inagaki, T. (2003). Adaptive Automation: Sharing and Trading of Control. In E. Hollnagel (Ed.),
Handbook of Cognitive Task Design (pp. 147–169). Lawrence Erlbaum Associates.
http://doi.org/10.1201/9781410607775
Kaber, D. B., & Endsley, M. R. (2004). The effects of level of automation and adaptive automation
on human performance, situation awareness and workload in a dynamic control task.
Theoretical Issues in Ergonomics Science (Vol. 5).
http://doi.org/10.1080/1463922021000054335
Lee, J., Bagheri, B., & Kao, H. A. (2015). A Cyber-Physical Systems architecture for Industry 4.0-
based manufacturing systems. Manufacturing Letters, 3, 18–23.
http://doi.org/10.1016/j.mfglet.2014.12.001
Lee, J., & Moray, N. (1992). Trust, control strategies and allocation of function in human-machine
systems. Ergonomics, 35(10), 1243–1270.
Lemoine, M. P., Debernard, S., Crevits, I., & Millot, P. (1996). Cooperation Between Humans and
Machines : First Results of an Experiment with a Multi-Level Cooperative Organisation in Air
Traffic Control. Computer-Supported Cooperative Work, 5(2), 299–321.
Lemoine, M.-P., & Hoc, J.-M. (1996). Multi-level Human-Machine Cooperation in Air Traffic
Control : an experimental evaluation. In T. R. G. Green, J. J. Cañas, & C. Waren (Eds.), ECCE-
8: Proceedings of the Eighth European Conference on Cognitive Ergonomics : Granada, Spain,
10-13 September 1996. Granada, Spain: European Association of Cognitive Ergonomics.
Lepratti, R. (2006). Advanced human-machine system for intelligent manufacturing : Some issues in
employing ontologies for natural language processing. Journal of Intelligent Manufacturing,
17(6), 653–666. http://doi.org/10.1007/s10845-006-0035-z
Lindström, V., & Winroth, M. (2010). Aligning manufacturing strategy and levels of automation: A
case study. Journal of Engineering and Technology Management - JET-M, 27(3-4), 148–159.
http://doi.org/10.1016/j.jengtecman.2010.06.002
MacCarthy, B. (2006). Organizational, Systems and Human Issues in Production Planning,
Scheduling and Control. In J. W. Herrmann (Ed.), Handbook of Production Scheduling (pp. 59–
90). Boston, MA: Springer US. http://doi.org/10.1007/0-387-33117-4_3
Martínez-Miranda, J., & Pavón, J. (2010). Human Attributes in the Modelling of Work Teams BT -
Balanced Automation Systems for Future Manufacturing Networks: 9th IFIP WG 5.5
International Conference, BASYS 2010, Valencia, Spain, July 21-23, 2010. Proceedings. In Á.
Ortiz, R. D. Franco, & P. G. Gasquet (Eds.), (pp. 276–284). Berlin, Heidelberg: Springer Berlin
Heidelberg. http://doi.org/10.1007/978-3-642-14341-0_32
May, G., Maghazei, O., Taisch, M., Bettoni, A., Cinus, M., & Matarazzo, A. (2014). Toward
Human-Centric Factories: Requirements and Design Aspects of a Worker-Centric Job
Allocator. In B. Grabot, B. Vallespir, S. Gomes, A. Bouras, & D. Kiritsis (Eds.), Advances in
Production Management Systems. Innovative and Knowledge-Based Production Management
in a Global-Local World: IFIP WG 5.7 International Conference, APMS 2014, Ajaccio, France,
September 20-24, 2014, Proceedings, Part III (pp. 417–424). Berlin, Heidelberg: Springer
Berlin Heidelberg. http://doi.org/10.1007/978-3-662-44733-8_52
McFarlane, D., Giannikas, V., Wong, A. C. Y., & Harrison, M. (2013). Product intelligence in
industrial control: Theory and practice. Annual Reviews in Control, 37(1), 69–88.
http://doi.org/10.1016/j.arcontrol.2013.03.003
Millot, P. (2014). Risk Management in Life-Critical Systems. (P. Millot, Ed.). Wiley Blackwell.
Millot, P., & Hoc, J.-M. (1997). Human-Machine Cooperation: Metaphor or possible reality? In 2nd
European Conference on Cognitive Science ECCS’97 (pp. 165–174). Manchester, U.K.
Millot, P., & Lemoine, M.-P. (1998). An Attempt for generic concepts toward Human-Machine
Cooperation. In IEEE International Conference on Systems, Man and Cybernetics. San Diego,
USA.
Millot, P., & Mandiau, R. (1995). Men-Machine Cooperative Organizations: Formal and Pragmatic
implementation methods. In J.-M. Hoc, P.-C. Cacciabue, & E. Hollnagel (Eds.), Expertise and
Technology: Cognition Computer Cooperation (pp. 213–228, 199). New Jersey: Lawrence
Erlbraum Associates.
Millot, P., & Pacaux-Lemoine, M. P. (2013). A common work space for a mutual enrichment of
human-machine cooperation and team-situation awareness. IFAC Proceedings Volumes (IFAC-
PapersOnline), 12(PART 1), 387–394. http://doi.org/10.3182/20130811-5-US-2037.00061
Nishitani, H. (1996). Human-computer interaction in the new process technology. Journal of Process
Control, 6(2), 111–117.
Oborski, P. (2004). Man-machine interactions in advanced manufacturing systems. The International
Journal of Advanced Manufacturing Technology, 23(3-4), 227–232.
http://doi.org/10.1007/s00170-003-1574-5
Pacaux, M. P., Debernard, S., Godin, A., Rajaonah, B., Anceaux, F., & Vanderhaegen, F. (2011).
Levels of automation and human-machine cooperation: Application to human-robot interaction.
IFAC Proceedings Volumes (IFAC-PapersOnline), 18(PART 1), 6484–6492.
http://doi.org/10.3182/20110828-6-IT-1002.00312
Pacaux-Lemoine, M. P. (2014). Human-Machine Cooperation Principles to Support Life-Critical
Systems Management. In Risk Management in Life-Critical Systems (pp. 253–277). Wiley
Blackwell. http://doi.org/10.1002/9781118639351.ch12
Pacaux-Lemoine, M. P., & Loiselet, A. (2002). A common work space to support the cooperation
into the cockpit of a two-seater fighter aircraft. Cooperative Systems Design: A Challenge of
Mobility Age, 157–172.
Pacaux-Lemoine, M.-P., & Itoh, M. (2015). Towards vertical and horizontal extension of shared
control concept. In IEEE International Conference on Systems, Man, and Cybernetics (pp.
3086–3091). Hong Kong, China. http://doi.org/10.1109/SMC.2015.536
Pacaux-Lemoine, M.-P., Simon, P., & Popieul, J. (2015). Human-Machine Cooperation principles to
support driving automation systems design. In 3rd International Symposium on Future Active
Safety Technology Toward zero traffic accidents (FAST-zero’15). Gothenburg, Sweden.
Pach, C., Berger, T., Sallez, Y., & Trentesaux, D. (2015). Reactive control of overall power
consumption in flexible manufacturing systems scheduling: A Potential Fields model. Control
Engineering Practice, 44, 193–208. http://doi.org/10.1016/j.conengprac.2015.08.003
Pires, J. N. (2005). Semi-autonomous manufacturing systems: The role of the human-machine
interface software and of the manufacturing tracking software. Mechatronics, 15(10), 1191–
1205. http://doi.org/10.1016/j.mechatronics.2005.06.008
Rajaonah, B., Tricot, N., Anceaux, F., & Millot, P. (2008). The role of intervening variables in
driver-ACC cooperation. International Journal of Human Computer Studies, 66(3), 185–197.
http://doi.org/10.1016/j.ijhcs.2007.09.002
Rasmussen, J. (1983). Skills, Rules, and Knowledge; signals, Signs, and Symbols, and Other
Distinctions in Human Performance Models. IEEE Transactions on Systems, Man, and
Cybernetics, 13(3), 257–266.
Romero, D., Noran, O., Stahre, J., Bernus, P., & Fast-Berglund, Å. (2015). Towards a Human-
Centred Reference Architecture for Next Generation Balanced Automation Systems: Human-
Automation Symbiosis. In S. Umeda, M. Nakano, H. Mizuyama, H. Hibino, D. Kiritsis, & G.
von Cieminski (Eds.), Advances in Production Management Systems: Innovative Production
Management Towards Sustainable Growth: IFIP WG 5.7 International Conference, APMS
2015, Tokyo, Japan, September 7-9, 2015, Proceedings, Part II (pp. 556–566). Cham: Springer
International Publishing. http://doi.org/10.1007/978-3-319-22759-7_64
Säfsten, K., Winroth, M., & Stahre, J. (2007). The content and process of automation strategies.
International Journal of Production Economics, 110(1-2), 25–38.
http://doi.org/10.1016/j.ijpe.2007.02.027
Schmitt, K. (2012). Automations influence on nuclear power plants: a look at three accidents and
how automation played a role. In International Ergonomics Assocociation World Conference.
Recife, Brazil.
Sheridan, T. B. (1992). Telerobotics, automation, and human supervisory control. MIT Press.
Solvang, B., Sziebig, G., & Korondi, P. (2012). Shop-floor architecture for effective human-machine
and inter-machine interaction. Acta Polytechnica Hungarica, 9(1), 183–201.
Trentesaux, D., Dindeleux, R., & Tahon, C. (1998). A multicriteria decision support system for
dynamic task allocation in a distributed production activity control structure. International
Journal of Computer Integrated Manufacturing, 11(1), 3–17.
http://doi.org/10.1080/095119298130930
Trentesaux, D., & Millot, P. (2016). A Human-Centred Design to Break the Myth of the ``Magic
Human’' in Intelligent Manufacturing Systems. In T. Borangiu, D. Trentesaux, A. Thomas, &
D. McFarlane (Eds.), Service Orientation in Holonic and Multi-Agent Manufacturing (pp. 103–
113). Cham: Springer International Publishing. http://doi.org/10.1007/978-3-319-30337-6_10
Trentesaux, D., Moray, N., & Tahon, C. (1998). Integration of the human operator into responsive
discrete production management systems. European Journal of Operational Research, 109(2),
342–361.
Valckenaers, P., & Van Brussel, H. (2016). Design for the Unexpected: From Holonic
Manufacturing Systems towards a Humane Mechatronics Society (1st Editio). Butterworth-
Heinemann.
Valckenaers, P., Van Brussel, H., Bruyninckx, H., Saint Germain, B., Van Belle, J., & Philips, J.
(2011). Predicting the unexpected. Computers in Industry, 62(6), 623–637.
http://doi.org/10.1016/j.compind.2011.04.011
Van Brussel, H., Wyns, J., Valckenaers, P., Bongaerts, L., & Peeters, P. (1998). Reference
architecture for holonic manufacturing systems: PROSA. Computers in Industry, 37(3), 255–
274. http://doi.org/10.1016/S0166-3615(98)00102-X
Woods, D. D., Tittle, J., Feil, M., & Roesler, A. (2004). Envisioning Human – Robot Coordination in
Future Operations. IEEE Transactions on Systems, Man and Cybernetics, 34(2), 210–218.
http://doi.org/10.1109/TSMCC.2004.826272
Zambrano Rey, G., Carvalho, M., & Trentesaux, D. (2013). Cooperation models between humans
and artificial self-organizing systems: Motivations, issues and perspectives. Proceedings - 2013
6th International Symposium on Resilient Control Systems, ISRCS 2013, 156–161.
http://doi.org/10.1109/ISRCS.2013.6623769