Practical Implementation of Multiple Model Adaptive Estimation Using Neyman-Pearson Based Hypothesis Testing and Spectral Estimation Tools
AIR FORCE INST OF TECH WRIGHT-PATTERSON AFB OH SCHOOL OF ENGINEERING
Pagination or Media Count:
This study investigates and develops various modifications to the Multiple Model Adaptive Estimation MMAE algorithm. The standard MMAE uses a bank of Kalman filters, each based on a different model of the system. Each of the filters predict the system response, based on its system model, to a given input and form the residual difference between the prediction and sensor measurements of the system response. Model differences in the input matrix, output matrix, and state transition matrix, which respectively correspond to an actuator failure, sensor failure, and an incorrectly modeled flight condition for a flight control failure application, were investigated in this research. An alternative filter bank structure is developed that uses a linear transform on the residual from a single Kalman filter to produce the equivalent residuals of the other Kalman filters in the standard MMAE. A Neyman Pearson based hypothesis testing algorithm is developed that results in significant improvement in failure detection performance when compared to the standard hypothesis testing algorithm. Hypothesis testing using spectral estimation techniques is also developed which provides superior failure identification performance at extremely small input levels.