| This dissertation uses polynomial chaos theory to address recursive parameter estimation in state space systems. It joins the recursive estimators with base excitation modeling concepts to determine the mass of off road vehicles, and successfully demonstrates the methods on actual vehicle data.;The recursive, polynomial chaos based estimators of this dissertation can be applied to linear and nonlinear state space systems having linear time invariant output equations. Unlike regressor model based estimators, this dissertation's estimators can be applied directly to state space systems, and in some situations, the proposed methods can be more easily tuned than state filtering methods. The new estimation techniques contribute to the solution of the vehicle mass estimation problem.;An accurate onboard estimate of vehicle mass is valuable to the optimal performance of safety systems, chassis controllers, and drivetrain controllers. These systems schedule gear shifts, actuate brakes, induce steer, schedule fuel injection, warn drivers of rollover susceptibility, etc. Since vehicle mass can vary significantly from one loading condition to the next, the estimate of vehicle mass must be updated online.;A significant number of mass estimation algorithms have been developed for on road conditions; however, the rough terrain real-time vehicle mass estimation problem remains relatively unexplored. Existing rough terrain solutions are difficult to apply in practice because they assume that the terrain profile is known, estimated, or measured, or they assume that the vehicle is equipped with an active or semi-active suspension. Instead, this dissertation adopts a base excitation approach. This approach treats the vertical accelerations of the four unsprung masses as measured inputs to the dynamic equations governing the motion of the sprung mass; the estimator uses these sprung dynamics to calculate the most likely value of the vehicle mass.;This dissertation applies the polynomial chaos estimators and base excitation concepts to experimental data from an actual vehicle. When joined with a detection algorithm, the proposed approach had a success rate of 94%: 31 predicted successes with only 2 false positives. Without the detection algorithm, the proposed approach had a success rate of 78%: 31 total successes out of 40 total experiments. |
