| In many estimation problems, it is desired to estimate system states and parameters simultaneously. However, inherent to traditional estimation architectures of the past, the designer has had to make a trade-off decision between designs intended for accurate state estimation versus designs concerned with accurate parameter estimation. This research develops one solution to this trade-off decision by proposing a new architecture based on Kalman filtering (KF) and Multiple Model Adaptive Estimation (MMAE) techniques. This new architecture, the Modified-MMAE (M{dollar}sp3{dollar}AE), exploits the benefits of an MMAE designed for accurate parameter estimation, and yet performs at least as well in state estimation as an MMAE designed for accurate state estimation. The M{dollar}sp3{dollar}AE accomplishes the simultaneous estimation task by providing accurate state estimates from a single KF designed to accept accurate parameter estimates from the MMAE. Additionally, an M{dollar}sp3{dollar}AE approximate covariance analysis capability is developed, giving the designer a valuable design tool for analyzing and predicting M{dollar}sp3{dollar}AE performance before actually implementing the M{dollar}sp3{dollar}AE and conducting a time-consuming full-scale Monte Carlo performance analysis. Finally, the M{dollar}sp3{dollar}AE architecture is applied to two existing research examples to demonstrate the performance improvement over that of conventional MMAEs. The first example involves a simple second-order mechanical translational system, in which the system's natural frequency is the uncertain parameter. The second example involves a 13-state nonlinear integrated Global Positioning System/Inertial Navigation System (GPS/INS) system, in which the variance of the measurement noise affecting the GPS outputs, is the uncertain parameter. |
