Sciences des données et diagnostic prédictif dessystèmes de mesure dans l'industrie 4.0
Loading...
Date
2026-07-04
Journal Title
Journal ISSN
Volume Title
Publisher
Université 20 Août 1955 - Skikda Faculté de Technologie Département de génie électrique
Abstract
In the context of Industry 4.0, the increasing number of sensors and the inherent uncertainty
of industrial measurements make predictive fault diagnosis increasingly challenging. This thesis
addresses the diagnosis of industrial systems using Principal Component Analysis (PCA), a
well-established statistical method for modeling highly correlated processes. After reviewing the
theoretical foundations of classical PCA and its limitations when dealing with imprecise data,
this work proposes an extension to interval-valued data through the SPCA (Symbolic Principal
Component Analysis) approach, allowing for a more realistic representation of measurement
uncertainty. This approach is then applied to fault detection and isolation, using respectively the
SPE index and the variable reconstruction principle, both extended to the symbolic framework.
The performance of the proposed method is validated through an application to a real industrial
case : a benchmark dataset from a gas turbine. The results show that the SPCA approach
enables reliable fault detection and accurate fault isolation, even under measurement uncertainty,
confirming its relevance for the monitoring and predictive diagnosis of modern industrial systems.