Derivative-driven window-based regression method for gas turbine performance prognostics

The domination of gas turbines in the energy arena is facing many challenges from environmental regulations and the plethora of renewable energy sources. The gas turbine has to operate under demand-driven modes and its components consume their useful life faster than the engines of the base-load operation era. As a result the diagnostics and prognostics tools should be further developed to cope with the above operation modes and improve the condition based maintenance (CBM). In this study, we present a derivative-driven diagnostic pattern analysis method for estimating the performance of gas turbines under dynamic conditions. A real time model-based tuner is implemented through a dynamic engine model built in Matlab/Simulink for diagnostics. The nonlinear diagnostic pattern is then partitioned into data-windows. These are the outcome of a data analysis based on the second order derivative which corresponds to the acceleration of degradation. Linear regression is implemented to locally fit the detected deviations and predict the engine behavior. The accuracy of the proposed method is assessed through comparison between the predicted and actual degradation by the remaining useful life (RUL) metric. The results demonstrate and illustrate an improved accuracy of our proposed methodology for prognostics of gas turbines under dynamic modes. © 2017 Elsevier Ltd