Turbine Engine Performance Estimation Using Particle Filters
Bong-Jun Yang, Prasenjit Sengupta, Padmanabhan K. Menon · 53rd AIAA Aerospace Sciences Meeting · 2015
A nonlinear, non-Gaussian Particle Filter is considered for engine health parameter estimation. The algorithm employs a high-fidelity full engine model as its central element to overcome the performance limitations imposed by the Gaussian noise-linear dynamics assumptions required in the Kalman filter formulation of the problem. A central feature of the present estimation problem is that the number of engine health parameters to be estimated is often greater than the number of available sensor measurements. This renders the linearized engine dynamics not fully observable. However, using an analysis of the high-fidelity engine model, it is shown that the system may be fully observable in a nonlinear sense. It is also shown that the system observability can be enhanced by using specific inputs. Ensuing particle filter implementation demonstrates that the number of parameters that need to be estimated by the particle filter can be greater than the number of available measurements, implying that nonlinear filters can overcome the nondeterminism imposed by linear Kalman filters.