Study and application of MIMO, LPV state observer for industrial manipulators

This thesis is carried out in collaboration with Comau S.p.A. The experimentation of state observers for the reconstruction of the angular velocity of the links of a flexible industrial manipulator has been performed, in the presence of unmodeled or uncertain parts. In a particular simplified condition in which there is only one axis moving at a time, a study was done to understand how faithfully the dynamics of the machine can be reconstructed using simple single axis models, extending them to take into account the multi-variable dynamics of the system and trying to reconstruct the action of non-linear friction as well. The observer model is braked into a nominal part that coincides with the dynamics of the center of mass of the robotic arm and a higher order part defined as “uncertainty function”, which changes the nature of the system from an isolated and single-variable system to a multi-variable system. This type of representation allows a significant reduction in model order, maintaining high accuracy in the reconstruction of the angular velocity of the center of mass. Various models of different complexities have been tested with both the asymptotic Luenberger observer and the steady-state Kalman filter. Furthermore, the difference in the reconstruction of the state is analyzed between a first approximation model and the identified model that better represents reality. The presence of friction is taken into account by a feedforward compensation or by the addition of a disturbance observer synthesized as a pole placement regulator. First the observers are tested in simulation, then on real data from a Racer 7-1.0 robot, using both a normal trajectory and a more energized one. To evaluate the quality of the reconstruction, a virtual sensor obtained from the identification of the manipulator is used and then a final test is carried out with the real Xsens gyroscope.