Rule-Based and Visual Model Evolution using G EN GE D J. PADBERG, C. ERMEL, R. BARDOHL Technische Universit¨ at Berlin, Germany email: padberg,rosi,lieske @cs.tu-berlin.de Abstract The evolution of software has become a more and more critical point in the use and maintenance of large software system. Accordingly, the con- tinuous change of the system’s design gets into focus. Here we suggest a two-fold approach combining visual architecture description with graphi- cal specification techniques. Rule-based modification – a generalization of graph transformations – at both levels can be used to describe the model evo- lution. Allowing transformations at both levels allows distinguishing evolu- tion steps with local and a global range. 1 Introduction Visual modeling techniques, e.g. ER diagrams, Petri nets, UML diagrams, etc., have been successfully applied in different application domains like computer science or (electrical) engineering. They provide an intuitive way to describe static and dynamic properties of systems, and their precision allows consistency checks and behavior simulation. In the past, visual modeling techniques for soft- ware engineering have been used mainly for the development of new software systems. Yet, their use for further development or later modification (evolution) of existing software, i.e. reengineering and continuous software engineering be- comes more and more important. Modern software systems have to be long- lasting, and they are connected and integrated with each other. In this extended abstract, we sketch a formal approach supporting the evolution of software at the design level, the model evolution. We suggest an approach of employing visual modeling techniques on two levels in order to describe the evolution of a system. The evolution steps can be regarded as rule-based modifications of the system specifications as well as its architecture. Our concepts especially consider the fol- lowing requirements of model evolution: The evolution of modern software systems is a complex task. Common modeling techniques usually are based on graphical notions which allow 1