Initial Situation:
Industrial robots offer great economic potential owing to their remarkable flexibility and versatility. They can be used as milling robots for metal-cutting manufacturing processes, for example. Compared with conventional milling machining centres, industrial robots are significantly more economical and enable components with a large volume to be milled. However, the robot’s low static and dynamic stiffness can result in it being displaced from the programmed path due to the process forces exerted on it. Without appropriate compensation, this may lead to the component’s shape deviating too heavily from the planned geometry.
Goals:
The aim of the ProKIRo research project is to increase the accuracy of milling robots by using modern, statistical machine learning methods and by fusing information. In order to successfully minimise the displacement of the robot by means of a compensation control or process control system, a suitable, accurate method for modelling the robot properties is required. Robot modelling has a significant impact on the quality of the process control and compensation control system. The aim therefore is to merge two established types of modelling (physically motivated and data-based robot models) and to use statistically confirmed learning methods to increase the accuracy of industrial robots in metal-cutting manufacturing processes while simultaneously validating the model forecasts.
Process:
The first task is to build a suitable test stand, which includes all the components that will subsequently be needed. While setting up the test stand, steps are to be taken to ensure that it can detect the process forces exerted, measure the displacement of the robot and process the measurement data. The displacement and the forces induced by the process are to be recorded continuously using measurement technology integrated into the robot. The functional capability is to be ensured by tests. The sets of data from the simulation and from the measurement data, which are required for the development of a self-learning modelling platform, are to be generated and recorded. These two sources of information are to be connected using stochastic information fusion mechanisms. An agent-based compensation control system is to be implemented on the basis of the preliminary work. Finally, the results are to be validated through the performance of milling tests.
