Development of a Self-Learning, Automatic Parameterisation of an Aerodynamic Part Feeding System

Abstract

The individualisation of customer requirements and the increasing pressure to reduce manufacturing costs due to the internationalisation of markets requires undertakings to react flexibly and to ensure long-term competitiveness through innovations in production processes. As far as lowering manufacturing costs is concerned, there is considerable potential for rationalisation in assembly work. At the same time, it is expected that the provision and supply of parts in automated assembly systems will become a bottleneck in future with regard to quality, time and costs. The reason for this is that the part feeding systems traditionally used are stretched to their limits in terms of supply performance, flexibility and process reliability. Orienting parts in feeding systems is often the most technically demanding process. Active, aerodynamic orientation systems for feeding parts have been developed at the Institute of Production Systems and Logistics. The system’s parameters have had to be configured manually by experienced technicians to date. This work takes up a great deal of time. In order to minimise this time spent by users, a genetic algorithm (GA) is developed in this paper, which enables the optimum parameter values for unknown parts to be identified automatically. To this end, the optimisation problem to be solved, namely to identify the parameter values for aerodynamic orientation, is first described mathematically. The structure of the GA and its method of functioning are then explained. A methodology to accelerate the convergence speed of the GA is presented in this context, in that the quantity of individuals in each generation and the number of test parts observed is adapted to the quality of the solution in order to shorten the time needed to find the solution.