Eylemsizlik duyucularının ve manyetometrelerin deterministik ve stokastik hata modellemesi

This thesis focuses on the deterministic and stochastic modeling and model parameterestimation of two commonly employed inertial measurement units. Eachunit comprises a tri-axial accelerometer, a tri-axial gyroscope, and a tri-axialmagnetometer. In the first part of the thesis, deterministic modeling and calibrationof the units are performed, based on real test data acquired from a ightmotion simulator. The deterministic modeling and identification of accelerometersis performed based on a traditional model. A novel technique is proposed forthe deterministic modeling of the gyroscopes, relaxing the test bed requirementand enabling their in-use calibration. This is followed by the presentation of anew sensor measurement model for magnetometers that improves the calibrationerror by modeling the orientation-dependent magnetic disturbances in a gimbaledangular position control machine. Model-based Levenberg-Marquardt and modelfreeevolutionary optimization algorithms are adopted to estimate the calibrationparameters of sensors. In the second part of the thesis, stochastic error modelingof the two inertial sensor units is addressed. Maximum likelihood estimationis employed for estimating the parameters of the different noise components ofthe sensors, after the dominant noise components are identified. Evolutionaryand gradient-based optimization algorithms are implemented to maximize thelikelihood function, namely particle swarm optimization and gradient-ascent optimization.The performance of the proposed algorithm is verified through experimentsand the results are compared to the classical Allan variance technique.The results obtained with the proposed approach have higher accuracy and requirea smaller sample data size, resulting in calibration experiments of shorterduration. Finally, the two sensor units are compared in terms of repeatability,present measurement noise, and unaided navigation performance.Keywords: Inertial sensors, deterministic error modeling, stochastic error modeling,in-field calibration, Levenberg-Marquardt algorithm, particle swarm optimization,gradient-ascent optimization, Allan variance, maximum likelihood estimation.