Thesis topic 2023

Title: Artificial intelligence solutions for the reliability and maintenance of smart factories

Supervisor : Eric GASCARD (Maître de Conférence, HDR)

Doctoral school : EEATS Spécialité de thèse : Automatique-Productique

Start date : 2023 September

Funding envisaged : Allocation de recherche EEATS

Domain: Industrial system engineering and Automation, Reliability Engineering, Artificial intelligence,
machine learning

Keywords: Industry 4.0, Cyber Physical Production System, Diagnosis, Maintenance, Dynamic Fault
Trees, Machine Learning

Summary of thesis project

The ability to know the health state of a production equipment is of paramount importance in the industry
of the future in order to implement intelligent maintenance of equipments in smart factories. Their
reliability analysis contribute to the sustainability of the equipment and meet the societal and
environmental responsibilities of production systems. In this thesis project, we want to develop reliability
analysis and decision support tools for diagnosis and maintenance using artificial intelligence tools.
Based on incomplete system modeling and system health monitoring data, we will address two
complementary approaches. First, we will address the problem of building dynamic fault trees using
machine learning approaches. Dynamic fault trees [Dugan&al 1992] represent scenarios with dynamic
behaviors and causal links between different events leading to the undesired system failure. This tool
allows a qualitative and quantitative analysis of reliability [Gascard & Simeu-Abazi 2015, Gascard &
Simeu-Abazi 2018]. Secondly, we will study the problem of diagnosis of dynamic systems for which we
have an incomplete knowledge of the failure modes: we will develop a hybrid diagnosis methodology
based on dynamic fault trees [Gascard&al 2011] and timed automata [Gascard&Simeu-Abazi 2013,
Gascard&Simeu-Abazi 2019]. The timed automata [Alur&Dill 1994] will allow the monitoring of the
system's modes of operation (nominal, degraded, failed) and will be built from the nominal model of the
system and the monitoring data.

Research objectives
The overall goal of this PhD project is to develop model- and data-driven methodologies that can
improve the study of the reliability of cyber-physical production systems and their intelligent maintenance
in the context of the industry of the future.

Profile and skills required

We are looking for a motivated candidate to take part in this PhD project. The candidate must have a
Master 2 degree or an engineering degree in Industrial Engineering or Industrial Systems Engineering
or Maintenance and Reliability Engineering.
The project is part of several disciplines: reliability and maintenance, model-based systems engineering,
machine learning (artificial intelligence), data analysis, programming. A candidate with skills in several
of these disciplines would be highly appreciated.

Context
A major scientific barrier in the support of the digital transformation of the production industry (Industry
4.0) is the lack of decision support tools for the study of reliability, diagnosis and maintenance of complex
dynamic systems such as cyber-physical production systems [Cardin2019]. Indeed, these cyber-
physical systems are heterogeneous technologies by nature: they combine mechanical, electronic,
software and network subsystems. This heterogeneity makes them more difficult to model and to apply
reliability methods [Dhillon 2020] (RAMS - Reliability Availability Maintainability and Safety) and
prognostics and health management [Kim&al 2017] (PHM) to their production equipment.
Different research works in the last decade have focused on the development of methods for the
diagnosis, prognosis and maintenance of complex dynamic systems using knowledge-based, model-
based or data-based approaches. However, these approaches assume a complete knowledge of the
system operation either model-based (in its nominal mode and its failure modes) or data-based (large
number of data from all the system operation modes). The aim of this PhD is to propose new reliability
analysis methods and decision support tools for diagnosis and maintenance using artificial intelligence
tools when we have only an incomplete modeling of the system by exploiting the monitoring data of the
system.
Method
To take into account the problem of incomplete modeling of a production system or insufficient
monitoring data (in quantity and quality) representative of its operation in all its modes (nominal,
degraded, failed) during the study of operational safety and maintenance, we propose to develop a
methodology based on dynamic fault trees and timed automata.
The dynamic fault tree tool will allow to represent the sequence of events leading to degraded or failed
modes of the studied system. The dynamic fault trees will be built by inference and machine learning
from the knowledge of a hierarchical and multi-view modeling of the dynamic system [Haj Kacem 2018]
and the monitoring data of the system. The state of the art proposes mainly static fault tree construction
methods based on models, knowledge or data, but very few works [Pai&Dugan 2002, Dehlinger&Dugan
2008, Baklouti&al 2020] propose dynamic fault tree construction methods, and these are based on a
complete knowledge of the structure or behavior of the system.
A lot of work has been done on the development of diagnostic methodologies for complex dynamic
systems based on models or data, but these presuppose the knowledge of a complete model or a
sufficiently large and exhaustive set of data on the different modes of operation of the system. Very little
work has been done on the development of diagnostic methods with incomplete knowledge, in models
or data, of the system [Kwong&Yonge-Mallo 2011, Yin&al 2019, Bates&al 2022]. In this thesis project,
we propose a hybrid diagnostic method combining the use of timed automata for the temporal monitoring
of the system's operating modes and dynamic fault trees for the detection and localization of defects.
Expected results
The thesis aims to develop methods and tools for the analysis of the reliability and the maintenance of
production systems in Industry 4.0 in a context of incomplete modeling of these systems due to their
complexity or incomplete knowledge of their operation by lack of exhaustive data on all possible
behaviors in the presence of degradation or failures. The expected results are :
- Development of a method for building dynamic fault trees by inference and machine learning
from partial system modeling and system monitoring data.
- Development of a hybrid approach to fault diagnosis based on timed automata and dynamic
fault trees.

References
[Alur&Dill 1994] Alur, R., & Dill, D. L. (1994). A Theory of Timed Automata. Theoretical Computer
Science, (126), 183-235.
[Baklouti&a l2020] Baklouti, A., Nguyen, N., Mhenni, F., Choley, J. Y., & Mlika, A. (2020). Dynamic fault
tree generation for safety-critical systems within a systems engineering approach. IEEE Systems
Journal, 14(1), 1512-1522.
[Bates&al 2022] Bates, I. W., Karimoddini, A., & Karimadini, M. (2022). A Learning-Based Approach for
Diagnosis and Diagnosability of Unknown Discrete Event Systems. IEEE Transactions on Neural
Networks and Learning Systems.
[Cardin 2019] Cardin, O. (2019). Classification of cyber-physical production systems applications:
Proposition of an analysis framework. Computers in Industry, 104, 11-21.
[Dehlinger&Dugan 2008] Dehlinger, J., & Dugan, J. B. (2008). Analyzing dynamic fault trees derived
from model-based system architectures. Nuclear Engineering and Technology, 40(5), 365-374.
[Dhillon 2020] Dhillon, B. S. (2020). Reliability, Maintainability, and Safety for Engineers. CRC Press.
[Dugan&al 1992] Dugan, J. B., Bavuso, S. J., & Boyd, M. A. (1992). Dynamic fault-tree models for fault-
tolerant computer systems. IEEE Transactions on reliability, 41(3), 363-377.
[Gascard&al 2011] Gascard, E., Simeu-Abazi, Z., & Younes, J. (2011). Exploitation of Built in test for
diagnosis by using Dynamic Fault Trees: Implementation in Matlab Simulink. Proceedings of the 20th
European Safety and Reliability Conference, ESREL.
[Gascard&Simeu-Abazi 2013] Gascard, E., & Simeu-Abazi, Z. (2013). Modular modeling for the
diagnostic of complex discrete-event systems. IEEE Transactions on Automation Science and
Engineering, 10(4), 1101-1123.
[Gascard&Simeu-Abazi 2015] Gascard, E., & Simeu-Abazi, Z. (2015). Failure root causes analysis of
complex systems—Dynamic Fault Tree approach. Proceedings of the 25th European Safety and
Reliability Conference, ESREL.
[Gascard&Simeu-Abazi 2018] Gascard, E., & Simeu-Abazi, Z. (2018). Quantitative analysis of dynamic
fault trees by means of Monte Carlo simulations: Event-driven simulation approach. Reliability
Engineering & System Safety, 180, 487-504.
[Gascard&Simeu-Abazi 2019] Simeu-Abazi, Z., & Gascard, E. (2019). Fault diagnosis method for timed
discrete-event systems: Application to autonomous electric vehicle. Proceedings of the 3rd International
Conference on Control, Automation and Diagnosis, ICCAD.
[Haj Kacem 2018] Haj Kacem, M.A. (2018). Contribution au développement d'une méthodologie de
diagnostic des systèmes Cyber-Physique. Thèse de doctorat, Université Grenoble Alpes.
[Kim&al 2017] Kim, N. H., An, D., & Choi, J. H. (2017). Prognostics and health management of
engineering systems. Switzerland: Springer International Publishing.
[Kwong&Yonge-Mallo 2011] Kwong, R. H., & Yonge-Mallo, D. L. (2011). Fault diagnosis in discrete-
event systems: Incomplete models and learning. IEEE Transactions on Systems, Man, and Cybernetics,
Part B (Cybernetics), 41(1), 118-130.
[Pai&Dugan 2002] Pai, G. J., & Dugan, J. B. (2002, November). Automatic synthesis of dynamic fault
trees from UML system models. In 13th International Symposium on Software Reliability Engineering,
2002. Proceedings. (pp. 243-254). IEEE.
[Yin&al 2019] Yin, X., Chen, J., Li, Z., & Li, S. (2019). Robust fault diagnosis of stochastic discrete event
systems. IEEE Transactions on Automatic Control, 64(10), 4237-4244.

Contact: eric.gascard@g-scop.grenoble-inp.fr