Synthetic data generation for hybrid prognosis models

Projekt:

JVTC

Sammanfattning:
In the context of condition based maintenance, diagnosis and prognosis are fundamental
tools in order to determine the state of a monitored system and estimate its remaining
useful life. The importance of both processes lies in their usefulness to assure appropriate
reliability and safety levels. It has to be highlighted that the former is a broadly used
process whereas still no much work has been done regarding the latter.
Three different kinds of modelling of systems are used for diagnosis and prognosis:
physical, data-driven and symbolic modelling. Physical models are based on the knowledge
behind the studied system and they are expressed by means of mathematical equations.
Data-driven models take advantage of data acquired from sensors placed in the monitored
system to create algorithms that predict the value of an outcome given some inputs.
Symbolic models are formed by an amount of information ranging from work orders, stock
information, maintenance reports and planning, to laws, standards and good practices,
among others.
In this research work, an emphasis has been placed in physical modelling, applied
to the field of rolling element bearings. However, this kind of modelling has not been
considered as exclusive for the other two options. In fact, this thesis proposes to make
the most of physical modelling with the aim of generating synthetic data that complement
the data and information that can be obtained from the data-driven and the symbolic
approaches. This data and information fusion is known as hybrid modelling.
In order to implement the physical modelling, a multi-body model for rolling element
bearings has been developed in such a way that the dynamic response of these machine
elements can be obtained, whatever the kind of rolling element and the configuration of
the bearings. Besides that, the capability to reproduce non-stationary operating conditions
has been considered in the model.
Synthetic data have been generated using this model taking into account the healthy
or faulty state of the rolling element bearing. Thus, the effect of its state on the dynamics
is analysed and a classification is done using vibration data. Moreover, this model has
also been used to estimate the fatigue life of bearings in non-stationary conditions in
combination with a finite element model.

Författare: Urko Leturiondo
Utgivare: Luleå tekniska universitet
Utgivningsdatum: 2014-11
Diarienummer: TRV 2011/58769
ISBN: 978-91-7583-035-3
Språk: Engelska
Kontaktperson: Per Olof Larsson Kråik, UHjbs