OATAO - Open Archive Toulouse Archive Ouverte Open Access Week

An unsupervised approach for health index building and for similarity-based remaining useful life estimation

Schwartz, Sébastien and Montero Jiménez, Juan José and Vingerhoeds, Rob A. and Salaün, Michel An unsupervised approach for health index building and for similarity-based remaining useful life estimation. (2022) Computers in Industry, 141. 103716. ISSN 0166-3615

[img] (Document in English)

PDF (Author's version) - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader
2MB

Official URL: https://doi.org/10.1016/j.compind.2022.103716

Abstract

Predictive maintenance techniques attempt to trigger a maintenance intervention at the right moment by estimating the life expectation. Predictive maintenance is increasingly implemented by automated approaches able to perform diagnostics and prognostics. The main part of recent research in these approaches is focused in machine learning structures whose reasoning is implicit and cannot be easily explained. This poses a problem for their implementation in highly constrained area such as aeronautics. To overcome this constraint, explicit reasoning approaches such as the Similarity-Based Model (SBM) can be implemented. The SBM has been widely used for fault diagnostics and the remaining useful life (RUL) estimation, but the development of SBM includes tasks that often rely on high skilled experts. For instance, data reduction techniques required for SBM are often performed by experts judgment whose outcomes are not always consistent. The produced features from these techniques are used to build the Health Index that can be used to create the degradation trends that serve as a reference for the SBM. To overcome these difficulties, an automatic and unsupervised approach based on the Kernel Principal Component Analysis is proposed to enhance the Health Index creation. It preserves as much of the sensor information as possible improving the similarity-based RUL estimation. Additionally, when estimating the RUL of a system, the most similar degradation trends stored in the SBM library are used to compute individual RULs, the final RUL is obtained by a fusion rule technique that combines all these individual RULs into a consolidated value. For the fusion rule techniques, a self-adaptive method that does not rely on human expertize is proposed. This fusion rule can benefit of the accumulated knowledge over the SBM operation. This unsupervised approach to develop a SBM is validated with promising results against an equivalent and supervised algorithm that came out best in the 2008 prognostic health management challenge.

Item Type:Article
HAL Id:hal-03937709
Audience (journal):International peer-reviewed journal
Uncontrolled Keywords:
Institution:French research institutions > Centre National de la Recherche Scientifique - CNRS (FRANCE)
Université de Toulouse > Ecole nationale supérieure des Mines d'Albi-Carmaux - IMT Mines Albi (FRANCE)
Université de Toulouse > Institut National des Sciences Appliquées de Toulouse - INSA (FRANCE)
Université de Toulouse > Institut Supérieur de l'Aéronautique et de l'Espace - ISAE-SUPAERO (FRANCE)
Université de Toulouse > Université Toulouse III - Paul Sabatier - UT3 (FRANCE)
Laboratory name:
Statistics:download
Deposited On:13 Jan 2023 10:37

Repository Staff Only: item control page