Ph. D. Project
Human factors and dynamic production scheduling
2022/09/16 - 2025/09/15
The changing nature of industrial requirements brings with it extensive new studies in manufacturing systems
where reactivity in the short term and adaptability to market changes in the long term become increasingly
demanded. To cover these new requirements, some efforts in manufacturing control have been paid to
propose hybrid control architecture and methodologies, where distributed artificial intelligence plays an
essential role (Cardin et al., 2017). For this purpose, the industry for the future paradigm brings technological
advancement due to Information and Communication Technologies (ICT), as well as to advancement in
distributed computer sciences. In order to make these architectures viable for industrial applications, the
human aspects must be considered and still remain a major and complex challenge. In the literature, the
human aspects have a big interest on different research axis, such as human/machine cooperation or
collaboration, with means to ease these interactions (Pacaux-Lemoine 2016, Boy 2017), on ethic (Trentesaux,
2021), or on the anthropocentric approaches to design hybrid control architectures (Bril El-Haouzi, 2017).
These approaches aim to make the smart technological entities of the system understandable by the humans,
while reproducing human behaviors in the decision-making process (Mezgebe et al., 2019) and designing social
relationships between them (Valette et al, 2020).

The objective of the thesis is to focus on human factors in the manufacturing control decision-making process.
At first, a focus will be done on the scheduling problem, well known in the operational research field. The
integration of human factors, such as fatigue or cognitive charge, were seldom investigated, especially due to
the variability and the unpredictability of humans (El Mouayni et al., 2019). The variability issue is induced by
heterogeneity, since each human has specific individual characteristics, physically (age, physical condition) or
mentally (emotional posture). This variability is also time depended due to human physiology that fluctuates
from day to day. Thus, this time dependency of the variability may lead to unpredictability and will cause some
human errors in the system, which remains a key scientific challenge on scheduling to tackle. Indeed, the time
dependency can be seen as a temporal loop, where scheduling and human factors both depend on each other.
For example, one task scheduling for human may be painful and the caused fatigue may affect next tasks
efficiency, until a deterioration that creates human errors and consequently, a decreasing performance of the
given schedule.

From these purposes, one of different issues of the thesis is the integration of this temporal loop in scheduling
problem, in order to propose a viable hybrid control architecture that tackles human factors. The originality of
the proposition lies on the design and use (instantiation) of an agent model (state, dynamic, decisions), which
will integrate both human (e.g. physical and mental conditions) and social (interaction with others)
dimensions. These dimensions can be modelled by specific states related to human factors, evolving in short
(fatigue), mid (cognitive charge) and long (mood) time horizon. These "human factor" states will be integrated
in scheduling mathematical model and evolve according to the decision made, i.e. the calculated sequence of
actions. Thus, each action will affect differently the human factors and some of these actions, although leading
to optimality when human is not considered, will instead become poor since they lead to a significant
deterioration of human condition and imply human errors.

However, the way to a obtain human factor state-based model will lead to a second major issue on the
estimation of these human factor states, but also on the effect of each action on them. The idea is to design
predictive-like methods, which may be based on artificial intelligence (e.g. reinforcement learning), In order to
rapidly observe these human factor states, by means of existing measurement, such as the efficiency or non-
efficiency of actions realized by the agent. In order to observe the human factors' states, predictive methods
(which can be from artificial intelligence) will be used to estimate the factors, by using other measured states
such as the efficiency (or lack of efficiency) of the actions performed by the agent. Similarly, the effect of
actions can be modelled with several methods, but the choice is dependent on the way that the agent
evolution will be modelled (e.g. with discrete-event model)

In order to show the applicability of proposed methods, the cyber-physical production cell of "TRACILOGIS"
testbed platform will be used. Some agent models with human factors will be implemented/instantiated
Human Factor, Scheduling, AI, Model Based Agent
Modeling and Control of Industrial Systems