OATAO - Open Archive Toulouse Archive Ouverte Open Access Week

Contribution to intelligent monitoring and failure prognostics of industrial systems.

Soualhi, Moncef. Contribution to intelligent monitoring and failure prognostics of industrial systems. PhD, Génie industriel, Institut National Polytechnique de Toulouse, 2021, 161 p.

(Document in English)

PDF (Author's version) - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader


This thesis was conducted within the framework of SMART project funded by a European program, Interreg POCTEFA. The project aims to support small and medium-sized companies to increase their competitiveness in the context of Industry 4.0 by developing intelligent monitoring tools for autonomous system health management. To do so, in this work, we propose efficient data-driven algorithms for prognostics and health management of industrial systems. The first contribution consists of the construction of a new robust health indicator that allows clearly separating different fault states of a wide range of systems’ critical components. This health indicator is also efficient when considering multiples monitoring parameters under various operating conditions. Next, the second contribution addresses the challenges posed by online diagnostics of unknown fault types in dynamic systems, particularly the detection, localization, and identification of the robot axes drifts origin when these drifts have not been learned before. For this purpose, a new online diagnostics methodology based on information fusion from direct and indirect monitoring techniques is proposed. It uses the direct monitoring way to instantaneously update the indirect monitoring model and diagnose online the origin of new faults. Finally, the last contribution deals with the prognostics issue of systems failure in a controlled industrial process that can lead to negative impacts in long-term predictions. To remedy this problem, we developed a new adaptive prognostics approach based on the combination of multiple machine learning predictions in different time horizons. The proposed approach allows capturing the degradation trend in long-term while considering the state changes in short-term caused by the controller activities, which allows improving the accuracy of prognostics results. The performances of the approaches proposed in this thesis were investigated on different real case studies representing the demonstrators of the thesis partners.

Item Type:PhD Thesis
Uncontrolled Keywords:
Institution:Université de Toulouse > Institut National Polytechnique de Toulouse - Toulouse INP (FRANCE)
Laboratory name:
Research Director:
Medjaher, Kamal and Nguyen, Thi Phuong Khanh
Deposited On:13 Oct 2021 14:21

Repository Staff Only: item control page