New Methods for Automated Generation of Robot Programs for Assembly of Injection Devices at Novo Nordisk

Deploying collaborative robots for assembly processes has a huge potential to support companies in reducing the lead time on new products. However, assemblyprocesses can be challenging to automate, e.g., due to part complexity, variability,and validation requirements. There are two main challenges when programmingrobots for a robust assembly process. First, a kinematic assembly trajectory needsto be found. This trajectory specifies the robot assembly movements in Cartesianspace. However, this is often insufficient to realize the desired product quality.Hence, a second task is often needed: finding an optimal compensation strategyin response to inaccuracies and fluctuating uncertainties. Robot programming forproduct assemblies is often done through a kinematic virtual simulation. Eventhough this is sufficient for obtaining the kinematic trajectory, it can be insufficient for programming and finding a suitable compensation strategy, since thisrequires simulating dynamic interactions. Dynamic simulation tools for robot programming exist; however, achieving efficient and accurate simulation of dynamicinteractions in a tight-fitting assembly process is an ongoing challenge. Dynamicsimulation engines are often based on discrete surface representations, such aspolyhedral approximations, which can lead to excessive contact points that canaffect the quality of the simulation. Simulations that are based on smooth surfacerepresentations do not have these flaws. Finding contact points between smoothsurfaces is typically more computationally expensive, but once a contact point isfound, it can be tracked efficiently.

This thesis investigates robot programming for assembly processes via simulation. First, simulation approaches for programming kinematic trajectories arestudied in the context of automated generation of the assembly motion. The studypresents approaches for programming kinematic trajectories in a simulation anddemonstrates them on an industrial assembly task. Second, dynamic simulationsof compensation strategies are studied. Specifically, tight-fitting assembly processes are studied, and a simulation tool is presented with a collision modulebased on smooth surface representations. Finally, it is demonstrated that the toolcan replicate the behavior of a compensation strategy implemented on a real robot.