Magnetic field separation for current prediction in three-phase systems : Regression-based current prediction

Abstract: Current controls the motion of a manipulator. The manipulators at ABB are powered by a three-phase alternating current system where shunt resistors are utilised to measure the current to the motors. Magnetic field sensors are instead investigated to eliminate issues with power losses, the number of components and the cost of the shunt resistors. Since current produces a magnetic field, it can be measured without contact using a magnetic field sensor. However, employing non-contact magnetic field sensors in three-phase implementations introduces problems with stray magnetic fields due to the three traces being in close proximity to each other. This magnetic crosstalk will influence the sensors, hence the current measurement for each trace. In this thesis separating this influence of the magnetic fields is done through a software approach. Initially, two magnetic field sensors, a tunnel magnetoresistance sensor and a Fluxgate sensor, were tested and evaluated to gain knowledge and understanding. From the different tests, it was decided to continue with the Fluxgate sensor. Further, a partial least-squares regression was constructed to separate the magnetic field and predict the current in each trace from the Fluxgate sensor output. From a simulation created, the current could be predicted with an error of approximately 1 nA, meaning less than 0.1% when considering a simulated linear model of the Fluxgate sensor.