Modelling High-Fidelity Robot Dynamics

Abstract: The field of robotics is in continuous development. Driving forces for the development are higher demands on robot accuracy and being more cost effective in the development process. To reduce costs, product development is moving towards virtual prototyping to enable early analysis and testing. This process demands realistic models and modelling is therefore of utmost importance. In the process of modelling high fidelity robot dynamics many different physical aspects have to be taken into account. Phenomena studied in this thesis stretch from where to introduce flexibilities, mechanical and dynamical coupling effects, and how to describe friction. By using a bottom up approach the effects are analysed individually to evaluate their contribution both to accuracy and computationalcomplexity. A strategy for how to model a flexible parallel linkage manipulator by introducing some crucial simplifications is presented. The elastic parameters are identified using a frequency domain identification algorithm developed in [Wernholt, 2007] and shows that the presented method works well up to a certain level of fidelity. Friction is modelled using empirically derived static and dynamic models. Evaluation of accuracy is conducted through identification of friction models for a real manipulator and it is seen that to capture all existing phenomena in low velocities  a dynamic model is needed. It is also seen that friction characteristics vary with temperature and a Kalman filter is suggested to adaptively estimate friction parameters. Finally an implementation of a flexible manipulator model using the software MapleSim is presented. The tool severely simplifies the process of modelling manipulators and enables for export to other environment such as simulation, optimization and control.