Robot design optimization
In the EU-project RealSim (Real-time Simulation for Design of Multi-physics Systems) DLR and KUKA worked together to develop methods and tools that ease the design of new robots or of variants of existing robots by taking into account the interaction of mechanics, electronics and software systems of a robot early in the design phase. The goal is to reduce development cost and time and improve robot performance. It is planned to include step after step the developed and demonstrated technology into the actual design process at KUKA. The figure sketches the design process. After the initial specification of the robot (desired payload, work space, etc.), a first model of the robot is created. In a robot component library, the data of older designs are available, that can be directly utilized or served as a reference. The initial design is carried out following heuristic rules, e.g., by “static” calculations to check the joint torques in different arm configurations and adjust the kinematic parameters. The initial design is followed by a MOPS design optimization phase in which a good compromise candidate is determined by minimizing the maximum of a set of (appropriately scaled) criteria for a set of typical industrial tasks. Finally, the design can be verified in a real-time simulation using the actual robot control hardware, replacing the not-yet-existing robot by a real-time simulation.
The core for robot optimization are flexible, detailed robot models. These are difficult to obtain because multi-domain models are required, containing components from multi-body systems, drive trains, electrical systems and controllers. Furthermore fast simulations are essential for optimization in rapid control prototyping and hardware in the loop simulations.A robot component library is the core for all design phases in order to describe the robot. A typical robot model is shown in the figure. It is based on the free, object-oriented modeling language Modelica. Modelica models are simulated with the commercial simulation package Dymola from the Swedish company Dynasim. In the RealSim project, Dynasim and DLR developed together new algorithms for the real-time simulation of stiff differential-algebraic equation systems in order that detailed robot models can be simulated in real-time. For the robots under consideration these algorithms give a speed-up of at least 15 with respect to standard integrators, such as the explicit or implicit Euler method with fixed step-size used in real-time simulations but also with respect to the state-of-the-art offline variable-step integrator DASSL. The basic idea is that stiff variables are discretized with the implicit Euler method, other variables with the explicit Euler method and that the symbolic algorithms of Dymola are applied on the discretized model equations.