Loading...
Improved-Performance Vehicle’s State Estimator Under Uncertain Model Dynamics
Avzayesh, Mohammad
Avzayesh, Mohammad
Date
2020-07
Author
Type
Thesis
Degree
Citations
Altmetric:
Files
Description
A Master of Science thesis in Mechanical Engineering by Mohammad Avzayesh entitled, “Improved-Performance Vehicle’s State Estimator Under Uncertain Model Dynamics”, submitted in July 2020. Thesis advisors is Mamoun Abdel-Hafez and Mohammad Al-Shabi. Soft copy is available (Thesis, Approval Signatures, Completion Certificate, and AUS Archives Consent Form).
Abstract
This thesis demonstrates the improved-performance state estimation of the navigational system using few fusion techniques. In this study, the Smooth Variable Structure Filter (SVSF) and some of its recent forms are examined for state estimation applications under uncertainty in the model dynamics and varying environments. The developments of the SVSF in terms of robustness and optimality are investigated in this thesis. Moreover, its combinations with different filtering strategies to acquire better accuracy whilst being robust to model uncertainty are studied. State estimation techniques, such as Extended and Unscented Kalman filters (EKF & UKF), SVSF, the combination of SVSF with EKF (EK-SVSF) and the combination of UKF with SVSF (UK-SVSF), are applied on an experimental case study, which is the Partial Discharge Localization. Their performance and behavior are validated and compared. Based on the steady state results, we propose the use of UKF and SVSF in a hybrid filter, known as the UK-SVSF, as it has the combined advantage of both filters, fast convergence and low Mean Squared Error (MSE). The UK-SVSF is tested in a navigational system, where the MIDG unit and GPS antenna sensors are used with fusion techniques to combine data from the two sensors. The results are compared to the solution from the MIDG, EKF, SVSF, UKF and EK-SVSF. The results show that in the absence of added noise and model uncertainties, all the filtering strategies have almost identical performance. However, injecting uncertainties to the system causes the performance of the UKF and the EKF to degrade and to go out of bounds. On the other hand, adding noise to the system causes the path of the SVSF to become rough with oscillatory behavior around the truth path. Conclusively, the Root Mean Squared Error (RMSE) results reveal that in the presence of both: noise and modeling uncertainties, which is referred to as the worst-case scenario (large magnitude of noise and model uncertainty, and missing information), the UK-SVSF has the superior performance; it reduces the RMSE of SVSF and UKF by 20% and 33%, respectively. Moreover, its RMSE is less than the RMSEs that are obtained from EKF and EK-SVSF by 38% and 18%, respectively.