Environmental Engineering Reference
In-Depth Information
Table 10.1
Main
wind
tur-
Parameter
Value
bine data
Nominal power, P
n
1,250 KW
Nominal speed, v
wn
12.5 m/s
Generator reference speed, X
ref
g
1,116 rpm
The generator speed measurement is represented in the upper subplot of
Fig.
10.2
. The second subplot represents the estimation of the second derivative of
the generator speed, from the measurements. At the appearance time of the
excessive noise (10 s), the decision function g(k) (third subplot in Fig.
10.2
)
increases and exceeds the threshold, set at 10, around the time instant 15 s. The
alarm is then switched to 1 and the decision function g
dis
ð
k
Þ
is activated for
detecting the fault disappearance (see lower subplots in Fig.
10.2
). For the chosen
algorithm parameters, the detection delay of an excessive noise on the incremental
encoder is around 5 s.
10.3.2.1 Note
The algorithm for excessive noise detection is based on the detection of a variance
change under the hypothesis of white Gaussian noise. Nevertheless, even with an
additive uniform white noise, the algorithm performance is not significantly affected
due to the fact that the pdf of r
ð
k
Þ
is close to a Gaussian distribution (see Fig.
10.3
).
10.4 Fault Detection and Isolation Based on Hardware
Redundancy
Hardware redundancy can notably be encountered for wind turbine speed mea-
surements. Indeed, both rotor speed and generator speed are measured, and they
are directly linked through the gear ratio. Besides, redundant measurements of the
generator speed are quite usual. We successively address the issue of residual
generation and residual evaluation in the following sections.
10.4.1 Residual Generation
A given physical quantity, say x (a temperature, a pressure, a flow, a position, a
velocity, …) is measured by a set of sensors, possibly based on different sensing
principles. The mathematical model describing the measurement process can be
written as:
