Robust Estimation and Fault Diagnostics for Aircraft Engines with Uncertain Model Data

The paper first presents a reasonably thorough tutorial type review of the literature in the area of fault diagnostics of dynamical systems. Then it summarizes the specific research being carried out by the author´s group in aircraft engine parameter estimation and fault diagnostics. In turbine engines, performance parameters such as thrust, turbine inlet temperature and stall margins cannot be measured directly and thus there is a need to estimate them from the measured outputs. The engine dynamics is highly nonlinear and the traditional linear models used in Kalman filter based techniques are subject to large uncertainties. In this research, we develop an adaptive estimator that augments the linear Kalman filter with a neural network to compensate for any nonlinearity that is not handled by the linear filter. The neural network is a radial basis function network that is trained off line using a growing and pruning algorithm. Next, a contribution in fault diagnostics in aircraft engines is presented. Specifically, we present a fault detection algorithm which uses a dynamic/adaptive threshold. The algorithm takes the parameter uncertainties into consideration and proposes a dynamic/adaptive threshold that makes use of the bounds on the parameter uncertainties which can thus distinguish an actual fault from the model uncertainties. In the absence of faults, a predetermined constant threshold would lead to more false alarms and missed detections under modeling uncertainties. However, a dynamic/adaptive threshold can accommodate uncertainties in the model and help in reducing false alarms and missed detections. The proposed methodologies are demonstrated by applying it to the simulation model of an aircraft engine available in the literature. The simulation results clearly show the improved effectiveness of the proposed approaches of this research.