On-line detection of aircraft icing: An application of optimal fault detection and isolation

This dissertation describes an innovative method utilizing gain-scheduled optimal linear observers (Fault Detection and Isolation Filters) for on-line fault detection and isolation of failures in aircraft utilizing existing sensors and avionics. Aircraft icing, and tail-plane icing in particular, is a serious safety concern and current methods of detecting aircraft icing are generally inaccurate and unreliable. Ice detection is accomplished by modeling aircraft icing as an input failure to a dynamic system. The method described here will allow the retrofit of older planes because it only utilizes existing sensors. This icing detection method motivated two new methods for designing optimal fault detection and isolation filters. These methods utilize Linear Matrix Inequalities to solve for the optimal detection filter by allowing the user to specify the desired eigenstructure. The first involves a least squares approach while the second utilizes a direct parameterization of the estimator feedback gain. A Monte Carlo simulation of the proposed ice detection method shows that the improvement offered by this new design is substantial. This technology is applicable to the detection and isolation of degradation failures in all dynamical systems.