LiU Homepage
Abstract |
Model Based Diagnosis with Application to Automotive Engines
Model based diagnosis has shown to be a promising approach to increase
the performance of diagnosis systems. One important application is
automotive engines for which legislative regulations specify hard
requirements. Here a model based approach is suggested for the
diagnosis of the air intake system of the SI-engine. The diagnosis
system is experimentally validated and it is shown that it
performs well. In the residual structure, i.e. the information of what residuals that
are sensitive to what faults, a \emph{don't care} option is introduced.
It is shown that \emph{don't care} improves the
robustness of the diagnosis system. Generally there exists a large number of possible residuals that can
be included in the diagnosis system. Therefore a problem in the
design of diagnosis systems is the selection of residuals. Another
problem is how to optimally introduce \emph{don't care} in the
residual structure. Both described problems and also the selection of
thresholds, are solved by a systematic procedure. The procedure is
based on the optimization of a new probability based performance
index. The optimization is performed by using real process measurement
data that are provided as an input to the procedure. The procedure is successfully applied to the previously described
problem of diagnosing the air-intake system of the SI-engine. Compared
to the first approach, which involved many ad-hoc solutions, better
performance is obtained in the sense that faults of less than half the
size can be detected. Central for model based diagnosis is the generation of residuals. Here
a method for linear residual generator design based on parity
equations is developed. It is an extension to the Chow-Willsky scheme
and includes both the continuous and discrete case. Also, it is
universal in the sense that all decoupling linear residual
generators can be obtained. It is shown that this is not the case for
previous extensions to the Chow-Willsky scheme. Fault detectability and strong fault detectability have earlier been
defined as properties of a residual. Here it is shown that both
detectability and strong detectability are properties of the system.
Two conditions on the system for detectability and
strong detectability analysis are provided.
Mattias Nyberg
1997
Informationsansvarig: webmaster
Senast uppdaterad: 2021-11-10
