Rough set based gas turbine fault isolation study

Gas path fault isolation is one of the key techniques in Engine Health Management systems. In order to accomplish gas path fault isolation successfully for a gas turbine engine, both an accurate off-design performance model and an effective fault isolation approach are necessary. In this thesis, two original and useful contributions to knowledge are presented: a new gas turbine off-design performance model adaptation approach and a new gas turbine fault isolation approach. This new adaptation approach uses optimal multiple scaling factors obtained by using a Genetic Algorithm to scale inaccurate component characteristic maps in gas turbine performance models to improve their prediction accuracy in different off-design conditions. The major feature of this approach is that it provides non- linear map scaling and therefore is able to provide more effective adaptation. The new fault isolation approach can be used to discover knowledge hidden in engine fault samples, transfers that knowledge into rules, and then uses those rules for fault isolation. In addition, it is also capable of selecting appropriate measurements for fault isolation, dealing with uncertainty caused by measurement noise. Enhanced fault signatures, which are represented by the measurement deviations and their ranking pattern in terms of magnitude, are developed to make gas turbine faults easier to distinguish and hence make this fault isolation approach more effective. The new adaptation approach was applied to the off-design performance model adaptation of a gas turbine, while the new fault isolation approach was employed for fault isolation in a gas turbine. The results show that the new adaptation approach is very effective in improving the prediction accuracy of off- design performance models and the new fault isolation approach is not only effective in fault isolation but also in selecting measurements for isolation and generating fault isolation rules.