Jonas Weigand, Magnus Volkmann, Martin Ruskowski

• Using industrial robots for machining applications in flexible manufacturing processes lacks a high accuracy. The main reason for the deviation is the flexibility of the gearbox. Secondary Encoders (SE) as an additional, high precision angle sensor offer a huge potential of detecting gearbox deviations. This paper aims to use SE to reduce gearbox compliances with a feed forward, adaptive neural control. The control network is trained with a second network for system identification. The presented algorithm is capable of online application and optimizes the robot accuracy in a nonlinear simulation.