Model-based Sensor Fault Detection and Isolation in Gas Turbine

This paper studies an Extended Kalman Filter (EKF) and Multi-Model Hypothesis Testing (MMHT) based sensor fault detection and isolation (FDI) scheme. The discussion is focused on fault signature generation and MMHT rather than EKF design. The proposed FDI logic is designed in the model parameter space that works for both input sensors, (i.e. sensing instruments of ambient and actuators), and output sensors from a model standpoint. A Filter bank is designed for robustness to separate the disturbances from sensor faults by utilizing their differences in dynamics. The proposed algorithm is verified throughout the entire gas turbine operation envelope with Monte Carlo simulation including measurement noise and bias, transients, heat soak dynamic inaccuracy and parameter variations. Numerical simulation results show that the technique can produce acceptable performance in terms of fault detection, false alarm and isolation.