| 102D6 |
|---|
| Performance Evaluation of Model-Based Fault Detection Method for Black-Box Systems |
| AUTHORS |
| Kousuke Kumamaru, Katsuhiro Inoue, Yasuhiro Hosoyamada, Tomomi Iwamura (Kyushu Institute of Technology) |
| ABSTRACT |
| This paper is concerned with a design scheme to model-based fault detection for non-linear black-box systems. Such a fault detection problem has two crucial issues. The first issue is to build an appropriate model of the object system. The second one is to construct a fault detection index which is sensitive to faults occurred in the system. As to the modeling, Quasi-ARMAX model which has features of linear structure and flexibility to express many kinds of non-linearity. On the other hand, as to the detection index, Kullback discrimination information (KDI) is introduced. It can be used as an index to evaluate difference between two models identified in different intervals. Furthermore, a multi-modeling approach to fault detection is considered for the system with high non-linearity, in which the object system is expressed by several local linear ARMAX models. In this work, the fault detection method is applied to a ship propulsion system, which was constructed by R.Izadi-Zamanabadi, and M.Blanke as realistic plant model for benchmark test of fault diagnosis. So far the effectiveness of this method for one fault-mode has been confirmed through simulation studies on the ship propulsion system. Therefore in this paper, simulations for more variety of fault-modes are executed in order to examine its performance. |
Back to Program
Time Table
Top of This Site
Annual Conference 2002