Timing analysis of a robot controller Masters’ Thesis

ABB Robotics has no methods for theoretical verification of the timing behavior of the robot control system. The system is complex, about 2,5 million line of code, distributed over 15 subsystems. When changing the system, it’s hard to predict how that change will affect the temporal behavior. The system was not designed to explicitly support impact analysis. This has resulted in a “trial-and-error” approach regarding temporal issues when adding new functionality. Timing related errors can be very costly, since they might occur only under very special conditions and thus might be hard to find. This thesis proposes a solution to the problems. A set of tools and methods for verifying the temporal behavior is presented. The approach of this work is to enable the temporal analysis of the robot controller by enabling the simulation of a model. Since no existing solutions has been found, the development of a modelling language and a simulator is necessary. The model is implemented by describing the robot control system in the developed notation (ART-ML) and inserting data measured from the control system. The model and the simulator have been used to simulate the system and the results from that simulation are easily comparable with the results from the measurements. The comparison shows as expected similarities. It is not a perfect match, but with respect to the roughness and simplicity of the created model the results are promising. The created model of robot control system consists of about 100 lines of code. That is a very simple model compared to the complexity of the modelled system. Creating a more detailed model would increase the accuracy of the simulation. Adding more details to the model is very possible since the data that is extracted by the measurements contains a lot of dependencies between the tasks that can be investigated and inserted in the model. Using these tools/methods with a more accurate model, it is possible to test changes to the robot controller before implementing them. This method of simulation-based timing analysis is very general and flexible. The modelling language is capable of describing many classes of systems and the simulator is fast, enabling simulations of complex models.