1.2 Roust Filtering and Fault Detection
5
ogy based on software redundancy, which was developed in United States in the
early 1870s. The key points to promote the rapid development of fault detection and
diagnosis technology is the urgent need of the market.
A fault occurs when some or all components of the dynamical system fails, which
leads to the deterioration of the whole system function. Therefore, when the dynamical system fails, all or part of the parameters show different characteristics with the
normal state. In general, the fault information will also show this difference. Fault
detection and diagnosis is to analyze and deal with the characteristics of faults, and
then to detect and isolate them. A complete fault detection and diagnosis process
includes fault feature extraction, fault evaluation and fault decision.
With the development of fault detection and diagnosis technology, many schemes
have appeared. Professor Frank believed that all fault detection and diagnosis methods can be divided into three categories [109], namely knowledge-based method
[110], model-based analytical method [111] and signal processing based method
[112]. Among these, the model-based fault detection methods have drown much
attention and many achieved rich results are available.
In 1971, Dr. Bear first proposed the concept of fault diagnosis detection filter
[113]. In his doctoral dissertation, he proposed the method of replacing hardware
redundancy with analytical redundancy, and made the closed-loop of the system be
stable by self-organization, so as to obtain the fault information of the system through
the output of comparator. In [114], Mehra and Peschon studied the model-based
fault detection. These two results were widely promoted as the origin of analytical
model-based fault diagnosis technology. Subsequently, a large number of researchers
discussed this model-based fault detection method based on state estimation [115–
119]. In general, the model-based fault detection schemes are mainly divided into
two categories, i.e., observer-based methods and filter-based methods. In fact, the
essence of these two methods is similar. Firstly, a state observer or filter is constructed
to estimate the system state and obtain the estimated value of the system output.
Then, the residual signal, which is the deviation signal between the estimated output
value and the actual measured value, is analyzed to determine whether faults occur.
In general, the residual signal of the system is very small or tends to zero. Once
the actuator, sensor or other components of the system fail, the residual signal will
change, which contains rich fault information. We only need to identify the fault
direction of the residual signal to accurately locate faults.
In 2004 and 2005, Zhong et al. respectively designed the fault detection problems
for linear multi-model jumping system in continuous-time case [120] and in discretetime case [121]. Subsequently, many scholars studied the fault detection and diagnosis of nonlinear multi-model jumping system [122–131], discrete-time multi-model
jumping system [132–134], uncertain multi-model jumping system [135, 136], generalized multi-model jumping system [137, 138] and networked multi-model jumping system [40, 139, 141].
In this book, we also studied the model-based fault detection. By the given multimodel jumping system model, the optimal fault detector is designed by using observer
and filter method. By reconstructing the system, the residual dynamic characteristics
of the external disturbance signal and fault signals, including nonlinearities and
5
ogy based on software redundancy, which was developed in United States in the
early 1870s. The key points to promote the rapid development of fault detection and
diagnosis technology is the urgent need of the market.
A fault occurs when some or all components of the dynamical system fails, which
leads to the deterioration of the whole system function. Therefore, when the dynamical system fails, all or part of the parameters show different characteristics with the
normal state. In general, the fault information will also show this difference. Fault
detection and diagnosis is to analyze and deal with the characteristics of faults, and
then to detect and isolate them. A complete fault detection and diagnosis process
includes fault feature extraction, fault evaluation and fault decision.
With the development of fault detection and diagnosis technology, many schemes
have appeared. Professor Frank believed that all fault detection and diagnosis methods can be divided into three categories [109], namely knowledge-based method
[110], model-based analytical method [111] and signal processing based method
[112]. Among these, the model-based fault detection methods have drown much
attention and many achieved rich results are available.
In 1971, Dr. Bear first proposed the concept of fault diagnosis detection filter
[113]. In his doctoral dissertation, he proposed the method of replacing hardware
redundancy with analytical redundancy, and made the closed-loop of the system be
stable by self-organization, so as to obtain the fault information of the system through
the output of comparator. In [114], Mehra and Peschon studied the model-based
fault detection. These two results were widely promoted as the origin of analytical
model-based fault diagnosis technology. Subsequently, a large number of researchers
discussed this model-based fault detection method based on state estimation [115–
119]. In general, the model-based fault detection schemes are mainly divided into
two categories, i.e., observer-based methods and filter-based methods. In fact, the
essence of these two methods is similar. Firstly, a state observer or filter is constructed
to estimate the system state and obtain the estimated value of the system output.
Then, the residual signal, which is the deviation signal between the estimated output
value and the actual measured value, is analyzed to determine whether faults occur.
In general, the residual signal of the system is very small or tends to zero. Once
the actuator, sensor or other components of the system fail, the residual signal will
change, which contains rich fault information. We only need to identify the fault
direction of the residual signal to accurately locate faults.
In 2004 and 2005, Zhong et al. respectively designed the fault detection problems
for linear multi-model jumping system in continuous-time case [120] and in discretetime case [121]. Subsequently, many scholars studied the fault detection and diagnosis of nonlinear multi-model jumping system [122–131], discrete-time multi-model
jumping system [132–134], uncertain multi-model jumping system [135, 136], generalized multi-model jumping system [137, 138] and networked multi-model jumping system [40, 139, 141].
In this book, we also studied the model-based fault detection. By the given multimodel jumping system model, the optimal fault detector is designed by using observer
and filter method. By reconstructing the system, the residual dynamic characteristics
of the external disturbance signal and fault signals, including nonlinearities and