Aller au menu Aller au contenu
Laboratory of Grenoble for sciences of conception, optimisation and production
< >

> GSCOP-Recruiting > GSCOP-Theses

Diagnosis based on hybrid models and machine learning tools - Application to Internet of Things

Thesis Director   : Jean-Marie Flaus, Olivier Adrot
PhD School EEATS (Electronique, Electrotechnique, Automatique, Traitement du Signal)
Start date 1 octobre 2018
Proposed funding
: allocation de recherche Ecole Doctorale

Brief Description :

A production system of goods or services is an organized system of activities, information, human and material resources providing a service or a manufactured product from a supplier to a customer. It can be described by dynamic models disturbed by the customers’ demands which are difficult to anticipate, by the variable costs of products, by the constraints on the production line, etc. This variability, which may be associated with a lack of knowledge of the manufacturing process for example, leads to take many uncertainties into account. A production system is also subject to hazards such as malfunctions or breakdowns of equipment, lost resources (unavailable operators, out of stocks, ...), which have to be quickly diagnosed to ensure the availability of the production line. These hazards generate different operating modes (such as degraded mode, maintenance operation, ...) that can be precisely apprehended by hybrid models. The fields of application of the diagnosis are varied since they can concern the detection of an hazardous situation in order to preserve the health of the operators, the localization of breakdowns to guide the maintenance operations to limit the unavailability of the production line, the analysis of potential cyberattacks in case of detected abnormal behaviors, ...

This work focuses more specifically on the rapidly expanding field of production systems controlled and monitored by Industrial Internet of Things (IIoT) devices. This concerns more specifically decentralized systems whose structure is variable according to the number of active elements (equipment or sensors) at a given moment; and for which a hybrid model is interesting. It can also concern systems with embedded intelligence, remote and distant measurement systems ... The IIoT makes it possible to generate a lot of very useful data for the diagnosis; however the challenge is to be able to use this data to build a dynamic model of the observed system.

     To do this, the objective consists in using machine learning tools. This area of expertise belonging to artificial intelligence concerns the design, the analysis, the development and the implementation of methods allowing a computer system to evolve through a systematic process to make decisions (concerning the diagnosis here) difficult or problematic for more conventional algorithmics.

The aim of this thesis is to couple machine learning approaches to methods of diagnosis based on dynamic systems to develop a new diagnostic approach in an uncertain and variable context [1], [2]. The developed approaches will have to be adapted to data collected by IIoT devices [3] and will be tested on an experimental platform under development.

[1] Diagnostic de systèmes hybrides incertains par génération automatique de Relations de Redondance Analytique Symboliques évaluées par approche ensembliste, thèse soutenue par Ngo Q. D. et encadrée par Flaus J.-M. et Adrot O., Doctorat de l’Université de Grenoble, Spécialité Automatique Productique, 31 aout, 2012.

[2] Uncertainty quantification in dynamic system risk assessment: a new approach with randomness and fuzzy theory. Abdo H. and Flaus J.-M., International Journal of Production Research 54, 5862-5885 (2016).

[3] Advantage and contribution of Internet of Things (IoT) for Occupational Health and Safety, Adrot O., Flaus J.-M., Technological innovation and organisational changes : the potential impacts on prevention INNOVORG 2017

Contact(s) :

Jean-Marie Flaus : email  :

Olivier Adrot : email :


Date of update March 22, 2018

Univ. Grenoble Alpes