Estudi sobre el impacte de la caracterització i el pre-tractament del senyal inercial en la solució INS/GNSS

Abstract

The aim of this thesis is to exploit the information provided by low-cost inertial sensors in order to bridge GNSS gaps and deliver a good estimate of the attitude of the vehicle in the INS/GNSS trajectory solution with little money investment. To achieve the main objective, the work is focused on two objectives: modelling the sensor errors both at laboratory level and on-board, and exploring inertial sensor redundancy. At laboratory level, a static test was carried to characterize both sensor long term drift behaviour and accuracy of an EPSON low-cost inertial unit (MEMS based). At on-board level, a Kalman Filter was implemented and tested with both real and simulated data. A new adaptive estimation technique is presented to deal with on-board noise model changes. This proposed adaptive method based on spectral analysis provides a more realistic uncertainty figure to the Kalman Filter yielding better results. The performance of the method is assessed comparing real and simulated trajectories against non-adaptive Kalman Filter. Also data gaps due to buffer overflow are addressed and an initial alignment and calibration setup procedure is proposed. The second part of the work focuses on exploring the effects of adding sensor redundancy on the solution hybridising multiple IMUs with a GNSS. First, the previous work done in this field is reviewed, presenting the trade-off between design topologies. A common problem in redundant systems is that when a sensor is biased, the feedback loop is unable to distinguish where it comes from. A new system architecture that tackles this problem is proposed in this work. This system takes advantage of the relative forces that arise when turning to de-couple sensor biases. To test the performance of the proposed architecture, sensor simulated data has been used to obtain a trajectory solution in a controlled environment. Yet another redundant architecture has been implemented which is based on several INS/GNSS Kalman filter pairs running in parallel and blending trajectory data. Both system performances are compared in reference to single IMU.