Automatic Implementation and Analysis for Fixed-point Controllers in Modelica using Dymola

Abstract: In model-based development, floating-point arithmetic is typically used for computations in algorithms and models. However, on many target platforms, there is no support for floating-point computations due to constraints on, e.g., price, size, power consumption, and execution speed. A solution is to use fixed-point arithmetic instead. Manually transforming floating-point code to fixedpoint code is an error-prone and time-consuming task. Therefore, this thesis has explored the possibility to automatically generate fixed-point code for the controller part of a Modelica system model using Dymola. To support this scenario, Modelica was extended with new experimental annotations and Dymola was extended with additional functionality for analysis and code generation. A complete Modelica model-toembedded code scenario was evaluated using Lego Mindstorms (NXT) as target platform and a tool chain was developed to be able to compile and run the generated code. A Modelica library with components corresponding to the NXT sensors and actuators was also developed. The work has, in addition to this thesis, resulted in two conference papers and has been used in teaching a project course (FRT090) at the Department of Automatic Control at Lund University. It was also used in another Master Thesis where the students generated a stabilizing controller for a two wheeled robot capable of carrying a human.