Authors:
Thijs Devos
1
;
2
;
Matteo Kirchner
1
;
2
;
Jan Croes
1
;
2
;
Jasper De Smet
3
and
Frank Naets
1
;
2
Affiliations:
1
DMMS Core Lab, Flanders Make, Gaston Geenslaan 8, 3001 Leuven, Belgium
;
2
LMSD Research Group, Department of Mechanical Engineering, KU Leuven, Celestijnenlaan 300, 3001 Leuven, Belgium
;
3
MotionS Core Lab, Flanders Make, Gaston Geenslaan 8, 3001 Leuven, Belgium
Keyword(s):
State Estimation, Extended Kalman Filter, Observability, Sensor Selection, Non-linear Vehicle Model.
Abstract:
This paper presents a novel automotive state estimation approach aiming to provide reliable results for multi-objective estimation applications. Because single-objective estimators typically feature simple, dedicated models, they often lack accuracy for highly dynamically coupled systems such as vehicles. Therefore, this approach features a more complex, system-level, non-linear vehicle model containing more accurate physics. Based on the assumption that the estimator targets a specific number of quantities of interest, an extensive observability analysis is performed to ensure stable estimator operation. Firstly, a novel algorithm to detect unobservable estimator states is presented, followed by a methodology for detailed analysis on which estimator states are decoupled using the linearized Jacobians. It is shown that if the unobservable states are partially decoupled and have no dependency towards the quantities of interest, an observable transformation can be carried out which sta
bilizes the estimator during operation ensuring reliable and interpretable results for the quantities of interest. The methodology is validated using an experimental vehicle case for which sensor selection was performed and demonstrates the estimator performance as well as potential limitations for unobservable vehicle states.
(More)