A Multiple Model-Based Approach for Gas Turbine Fault Diagnosis

In gas turbines, using instrumental methods, it is impossible to detect the faults that detour the performance and specifications of the main components, such as compressors, combustors, and turbines. These faults usually cause total efficiency to fall with fuel consumption and increased pollution. This paper proposes a novel multiple-model-based fault estimation method to detect, isolate, and estimate such thermodynamic faults. These faults are produced gradually or suddenly as a fall in the gas turbine main components' efficiencies or their air/gas mass flow rates during exploitation. Diagnosing is fulfilled by including some parameters as the components' health indicator variables in the gas turbine nonlinear model and continuously estimating them. To enhance the robustness of the method, by using linear models through the Bayesian theorem, an equivalent adaptive model in the frame of a convex set that could cover plants' dynamics in a vast operating range is established. Then, the existing faults are estimated by decoupling the generated robust residuals from an adaptive filter. Finally, the competency of the proposed approach is evaluated using an actual gas turbine operating datum and specification in a simulation environment.