Abstract

In todays market, the increasing demand on utilizing gas turbine engines can be quite costly if users rely only on traditional time-based maintenance schedules. Meeting both the safety and the economical aspects of such systems could be realized by using an appropriate maintenance strategy in which the prediction of the engine health condition is employed to ensure that the system is maintained only if necessary. Towards this end, in this thesis the prognosis problem in the gas turbine engines is investigated.
As in every rotational mechanical equipment, gas turbine rotating components also degrade during the engine operation which may deteriorate their performance. The engine
degradation may originate from different sources such as aging, erosion, fouling, corrosion, etc. Hard particles mixed with the air can remove the materials from the flow path components (erosion) and cause aerodynamic changes in the blades, which can consequently reduce the affected components performance. Accumulated particles on the flow path components and annulus surfaces of the gas turbine (fouling) can also reduce the flow rate of the gas and consequently decrease the power and efficiency of the affected components.
Among different degradation sources in the engine, erosion and fouling are considered as two well-known degradation phenomena and their effects on the engine system prognostics
are studied in this thesis.
Towards the above end, a controller is designed to control the thrust level of the engine and a Matlab/Simulink platform is employed to incorporate the effects of the above degradation factors and the engine dynamic model. The engine performance degradation trends are modeled by using three types of time-series based techniques namely, the autoregressive integrated moving average (ARIMA), the vector autoregressive (VAR) and the hybrid fuzzy autoregressive integrated moving average (hybrid fuzzy ARIMA) models. One of
the challenges associated with time-series approaches is selecting a proper model which
represents the structure of the time-series and is employed for prediction and prognosis purposes. Two widely used criteria namely, the Akaike’s information criterion (AIC) and the Bayesian information criterion (BIC) are used in order to select the best model. The challenges of coping with the uncertainties due to variety of sources such as measurement noise, insufficient data and changing operating conditions are inevitable factors. Taking the above facts into account, it may not be practical to obtain or be concerned with an exact prediction information. Therefore, we construct instead confidence bounds that provide a
realistic boundary for the prediction and this is applied to all our proposed approaches in this thesis.
The first method in this thesis deals with modeling a measurable parameter using its historical data which is a fine-tuned version of the ARMA model for non-stationary time
series analysis. The second method, VAR model, models the measurable parameters by fusing historical data with the current and past data of some other engine measurable data
in a vector form so that one can get benefit of more measurement parameters of the engine.
The third method deals with fusing two measurable parameters using a Takagi-Sugeno fuzzy inference engine.
In this thesis we are focused on modeling the engine performance degradations due to the fouling and the erosion which are the two main causes of gas turbine engine deterioration. In order to evaluate the performance of the proposed methods, they are applied to three different scenarios. These scenarios include the compressor fouling, turbine erosion phenomena and their combination with different severities. Our numerical simulation results show that the performance of the hybrid fuzzy ARIMA model is superior to that of the ARIMA and VAR methods.