AlgorithmCopyright: © Leon Gorissen
In cost-efficient robot-based Laser Materials Processing (LMP) path deviations are the major obstacle to overcome before it can be adapted widely. Thus, task specific algorithms overcoming this obstacle form the main research focus in the Digital Photonic Production group.
In LMP the Tool Center Point (TCP) – the point of laser and material interaction – depends on different components, which can be clustered in three groups (compare Figure 1).
- Handling system: The handling system in this case provides six Degrees of Freedom (DOFs) through the revolute joints of a robotic system.
- Optical system: The depicted processing head provides two DOFs by means of positioning of the laser beam in an X-Y-plane using a galvanometer scanner.
- Process: The process provides four DOFs. These result from a slight allowance in the z-direction (standoff distance), rotational freedom along the optical axis (rotational symmetry of the laser beam) and cone-shaped deviation allowance from the surface normal (process stability for slight deviation of the optical axes from the surface normal).
For a complete description of a pose (position and orientation) in space, three coordinates and three rotations have to be defined, i.e. there are - in total - six degrees of freedom (DOF) in a pose in time. In this case (compare Figure 1), twelve DOFs are available, while only six DOFs are required to define the pose, i.e. we are presented with a redundant state representation.
We model the kinematic chain of all major components of LMP systems: either as physical joints or as virtual joints. This yields a complete task-specific redundancy description of the LMP state.
Based upon this description, we make use of this redundancy in order to optimize different aspects of our task: path deviation minimization, trajectory smoothness, energy consumption, processing time, material consumption. Optimization is carried out by different state-of-the-art optimization techniques: graph-based optimization, Reinforcement Learning and Deep Learning.
Research is conducted in-silico as well as in-situ, i.e. in simulation and experimentally validated:
- Kinematic and dynamic modelling of robotics, processing heads and LMP processes
- Trajectory planning for robot-based LMP, i.e. redundancy resolution strategies
- Trajectory optimization strategies