Environmental Engineering Reference
In-Depth Information
10.1 Introduction
Fault diagnosis systems aim at detecting and locating degradations in the operation
of wind turbine components as early as possible. This way, maintenance opera-
tions can be performed in due time, and during time periods with low wind speed.
Therefore, maintenance costs are reduced as the number of costly corrective
maintenance actions decreases. Besides, the loss of production due to maintenance
operations is minimized.
Fault detection and isolation (FDI) in wind turbines has been the subject of a
large number of publications notably stimulated by different benchmark problems
[
1
,
2
]. In those benchmarks, the authors mostly consider various sensor and
actuator faults like sensor faults on pitch position and generator speed measure-
ments, pitch actuator faults, and converter faults. Structural health monitoring has
also been considered for wind turbines specifically. The aim is to detect changes
such as delamination and cracks in the structural parts of the turbine, namely the
tower and the blades. Presently, visual inspection through regular maintenance
operations is the usual approach. Yet specific robots are being developed to make
this task easier, and the permanent monitoring, thanks to appropriately placed
sensor networks for strain and displacement measurement, is also under investi-
gation [
3
,
4
]. Last but not least, vibration monitoring techniques have been thor-
oughly studied for the monitoring of the gearbox, the bearings, and the generator.
Commercial products have been developed for these components, specially since
gearbox is responsible for the largest downtime [
5
,
6
].
A complete monitoring system should have a modular structure with appro-
priate methods for each component, as depicted in Fig.
10.1
. The lower layer
within this architecture consists of measurement validation modules. The latter
aim at detecting and locating (or isolating) sensor faults. These different mea-
surement validation modules will be the focus of this chapter. A short section will
address faults that can be detected by analyzing the measurement samples issued
by a single sensor. This includes excessive noise on the measurements, flat signals,
and outliers. However, the focus of this chapter will be on incipient faults like
small sensor bias and drifts, which typically require the use of several sensor
signals to achieve FDI. Two approaches can be distinguished depending on the
instrumentation: hardware redundancy and analytical redundancy. The first
approach applies for redundant sensors, namely groups of sensors that are mea-
suring the same physical quantity. A sensor exhibiting a significant discrepancy
with respect to the others will be discarded. A way to handle this issue within a
proper statistical framework will be discussed. On the other hand, analytical
redundancy amounts to detecting incoherencies between a model of the supervised
system and data recorded online on this system [
7
]. A systematic approach to
handle sensor faults in the voltage and current sensors that equip a wind-driven
doubly fed induction generator (DFIG) will be used to illustrate this class of
methods.
