LEAN BLOWOUT FAULT PREDICTION FOR DRY LOW EMISSION GAS TURBINE USING HYBRID OF SUPPORT VECTOR MACHINE AND BAYESIAN BELIEF NETWORK

A Dry-Low Emission (DLE) gas turbine reduces Carbon Oxide (COx) and Nitrogen
Oxide (NOx) emission during power generation. However, DLE gas turbines
frequently encounter trips due to Lean Blowout (LBO) fault. The state-of-the-art
studies on LBO are performed in a laboratory-scale where gas turbine dynamics are
not well represented. There is a potential of utilizing a dynamic model where DLE gas
turbine model is developed to predict LBO fault. However, the superior prediction
technique such as Support Vector Machine (SVM) is deterministic without the
probability of the impending trip. Therefore, this thesis proposes a DLE gas turbine
model with a hybrid of Support Vector Machine-Bayesian Belief Network
(SVM-BBN) for LBO fault prediction.