Thesis defense of Javier Fernandez Anakabe

Back

Thesis defense of Javier Fernandez Anakabe

THESIS

Thesis defense of Javier Fernandez Anakabe

Title of the thesis: “An Attribute Oriented Induction based methodology to aid in Predictive Maintenance: Anomaly Detection, Root Cause Analysis and Remaining Useful Life”. Obtained the SOBRESALIENTE qualification .

2020·02·28

$titulo.getData()


  • Title of the thesis: “An Attribute Oriented Induction based methodology to aid in Predictive Maintenance: Anomaly Detection, Root Cause Analysis and Remaining Useful Life”.
  • PhD programme: DOCTORATE PROGRAMME IN MECHANICAL ENGINEERING AND ELECTRICAL ENERGY
  • Thesis directors: Urko Zurutuza Ortega, Ekhi Zugasti.
  • Court:
    • President: Olatz Arbelaitz Gallego (UPV-EHU)
    • Vocal: Magda Ruiz Ordoñez (Universidad Politécnica de Cataluña)
    • Vocal: Rosario Maria Basagoiti Astigarraga (Mondragon Unibertsitatea)
    • Vocal: Urko Leturiondo Zubizarreta (IKERLAN, S. Coop.)
    • Secretary: Carlos Cernuda García (Mondragon Unibertsitatea)

Abstract

Predictive Maintenance is the maintenance methodology that provides the best performance to industrial organisations in terms of time, equipment effectiveness and economic savings. Thanks to the recent advances in technology, capturing process data from machines and sensors attached to them is no longer a challenging task, and can be used to perform complex analyses to help with maintenance requirements. On the other hand, knowledge of domain experts can be combined with information extracted from the machines’ assets to provide a better understanding of the underlying phenomena. This thesis proposes a methodology to assess the different requirements in relation to Predictive Maintenance. These are (i) Anomaly Detection (AD), (ii) Root Cause Analysis (RCA) and (iii) estimation of Remaining Useful Life (RUL).

Multiple machine learning techniques and algorithms can be found in the literature to carry out the calculation of these requirements. In this thesis, the Attribute Oriented Induction (AOI) algorithm has been adopted and adapted to the Predictive Maintenance methodology needs. AOI has the capability of performing RCA, but also possibility to be used as an AD system. With the purpose of performing Predictive Maintenance, a variant, Repetitive Weighted Attribute Oriented Induction (ReWAOI ), has been proposed. ReWAOI has the ability to combine information extracted from the machine with the knowledge of experts in the field to describe its behaviour, and derive the Predictive Maintenance requirements.

Through the use of ReWAOI, one-dimensional quantification function from multidimensional data can be obtained. This function is correlated with the evolution of the machine’s wear over time, and thus, the estimation of AD and RUL has been accomplished. In addition, the ReWAOI helps in the description of failure root causes.

The proposed contributions of the thesis have been validated in different scenarios, both emulated but also real industrial case studies.