IEEE Systems, Man and Cybernetics Magazine - January 2023 - 6

control systems and that the quantitative prediction method
is the most suitable. For microfaults, Quan et al. [13] pointed
out that detection and estimation methods, such as
observers, can be used to deal with microfaults.
In engineering applications, the literature [14], [15] has
established different types of observers for closed-loop control
systems to generate a fault residual. This method can
accurately identify the mutation gain fault and slow change
bias fault in the fault sensor in real time, and the residual
generated by the designed security prediction system is not
affected by external disturbance. The structure, theoretical
derivation, and stability proof of the observer are simple.
Ruder [16] used real-time simulation to establish a hardin-the-loop
simulation model and studied the modulation
effect of a bearing fault on the stator current of a directdriven
wind turbine equipped with a variable-shaft-frequency
permanent magnet synchronous generator to verify the
correctness of the proposed security prediction method.
All of these methods can model and process the hidden
danger of the industrial control system to a certain extent so
as to achieve the function of detecting and analyzing the
fault, but they cannot accurately predict the fault signal of
the pitch angle sensor and the actuator of the variable propeller
system in real time. In the variable blade system, the
pitch angle sensor and the variable blade actuator are important
devices to ensure the normal operation of the variable
blade system, so this article mainly focuses on the real-time
prediction method of the safe operation trend of the two.
Aiming at the problems described previously, combined
with the relevant data of wind turbine security prediction,
the main contributions of this article are as follows:
◆ A data-driven CNN was designed to monitor and predict
the health condition of the wind turbine system. The
exact system model is not needed to establish the CNN.
◆ A fault decision index combining both the current and
history information is proposed to improve the ability
of fault decision.
Dynamics Modeling of
a Wind Turbine Rotor System
The general dynamics model of the wind turbine rotor system
used in this article is shown in Figure 1, where vg
resents the wind speed before air enters the wind wheel,
vg
Wind
Turbines
Mr
Mr represents the torque of the main drive shaft, ~ represents
the rotational speed of the motor rotor, m
g
a
represents
the reference value of the pitch angle, and a represents
the pitch angle.
The linearized rotor system model of the wind turbine
after failure is shown [17]:
xt Ax tBLu tFft Dd taa a
^^ ^^ ^
=+ ++
=+
where [] ; ~ gm represents the measured
^h = am ^^=hhdt vtg
^h represent the
xt
^h = a~ ~
yt Mrg gm
T
ggm
value of the generator speed; ut
T
() [];
and Ls
,
= ~~ a fta^h and fts
additive fault values of generalized actuator and sensor,
respectively; and La
F ,a and Fs
matrix and fault matrix, respectively.
Security Prediction Based on a CNN
The goal of security prediction is to estimate the immeasurable
system state value. In essence, it is a mathematical
model of the system, which receives the input value and
output value of the system and compares it with the state
estimate value of the estimator itself to get the estimation
error. The simplest form of an estimator is a Luenberger
IEEE Systems, Man and Cybernetics Magazine - January 2023

Table of Contents for the Digital Edition of IEEE Systems, Man and Cybernetics Magazine - January 2023
