Model-free IMU-based DP State Estimator
MetadataVis full innførsel
The main goal of this thesis has been to develop a new model-free state estimator for dynamic positioning, based on an inertial measurement unit, a gyrocompass and a position reference system. The state estimator should provide estimates for position, velocity and attitude. A literature study was given on different navigation systems and the associated sensors, where the nonlinear observer presented by Grip, Fossen, Johansen and Saberi has been described thoroughly. This model-free state estimator was modified to include a gyrocompass and to perform under varying position measurement rates. The modified nonlinear observer was tested both in simulations and experimental tests. The simulations were realized with the use of MATLAB and Simulink, and was divided in different case studies. To begin with, two accelerometer bias estimation methods were compared to see which one was best suited. Secondly, the state estimator was tested during consistent and inconsistent position measurement rates. In the next case, the dead-reckoning capabilities of the nonlinear observer was tested under the presence of simulated noise and bias on the inertial measurement unit. Lastly, a comparison between the model-free state estimator and the model-based Kalman filter was carried out. The experimental tests were realized in collaboration with Marine Technologies LLC in Egersund, Norway. The tests were conducted on a trolley using a MEMS based inertial measurement unit. The nonlinear observer was implemented on a computer, tested in different scenarioes, and compared with a reference system using a logging system. The tests were divided in three main case studies. From the first case, ideal sensor rates were considered. The second case tested the performance of the observer without gyrocompass measurements. Lastly, the performance of the observer without the position reference system was tested. The modified nonlinear observer was found to provide satisfactory estimates, in both the simulations and the experimental tests. Due to a minor implementation problem in the experimental setup, the performance of the observer was at some degree degraded. Since the raw data from the sensors were logged, additional simulations were carried out with the use of MATLAB. These simulations gave good results. Because of sensor errors on the inertial measurement unit, the performance without aiding by the position reference system was degraded. On the other hand, the observer performed with great results in the absence of gyrocompass measurements.