Information filter based sensor fusion for estimation of vehicle velocity

Sammanfattning: In this thesis, the possibility to estimate the velocity of a Heavy-duty vehicle (HDV) based on the Global Positioning System (GPS), an Inertial Measurement Unit (IMU) and the propeller shaft tachometer is investigated. The thesis was performed at Scania CV AB. The objective was to find an alternative to the wheel encoders that currently are used for velocity estimation. Three different sensor configurations were tested: the first (SC1) was based on GPS and an accelerometer, the second (SC2) was based on GPS, an accelerometer and a gyroscope, and the third (SC3) was based on GPS, an accelerometer and the propeller shaft tachometer. An experimental sensor architecture for collection of measurement data was built. The sensor configurations were evaluated in simulations based on measurement data collected from a test vehicle at Scania’s test track in S¨odert¨alje. An Information filter (IF) was used for decentralized fusion of sensor measurements. The sensor configurations were evaluated against the wheel encoders and a high quality GPS/IMU reference system using the Root Mean Squared Error (RMSE), Mean Signed Deviation (MSD) and maximum error. It was concluded that the sensor configurations based solely on GPS and IMU are not robust enough during GPS outages because of the IMU’s drift. An alternative source to GPS for correction of the IMU errors was thus necessary. The propeller shaft tachometer was used for this. The RMSE for this sensor configuration (SC3) was reduced with 37% and the MSD was reduced with 60% in comparison to the wheel encoder based velocity in the most extreme test performed, when the wheels slip and the GPS signal is erroneous during two instances. SC3 is thus proposed for further development. This work lays the basis for real-time implementation of the proposed sensor configuration and shows the feasibility of using the IF for decentralized multi-sensor fusion. It is also suggested to use the IF for integration of multiple sensors to create a refined and redundant velocity estimation.