A DECENTRALIZATION METHODOLOGY FOR REAL–TIME CONTROL APPLICATIONS Martin Törngren and Jan Wikander DAMEK-Mechatronics, Dept. of Machine Design, The Royal Institute of Technology, S-100 44 Stockholm, Sweden. Email: martin@damek.kth.se, Fax: +46-8-20 22 87 Abstract: Machine-embedded control applications constitute a recent area for distributed computer systems. The industrial potential is large, but research has so far not directly treated such systems, which are characterized by real-time multirate control systems and I/O bounds. An engineering methodology for evaluating different hardware structures, control-system structures and allocation approaches is outlined. It consists of the following steps: control sys- tem structuring; decentralization involving partitioning, allocation and evaluation; and execution strategy considerations. Feedback from the use of the methodology in two case studies has been incorporated into the methodology. Keywords: Allocation, control applications, decentralized control, decomposition, design, distributed control, real-time systems, design. 1. INTRODUCTION Technological progress and user needs are stimulating new areas of distributed computer control systems. In applications like vehicles, aircraft, robots and produc- tion machines, stand-alone computer controllers are being integrated to form cooperating subsystems which can together provide improved functionality, performance and flexibility, and reduced costs. Mod- ern and future machinery will therefore, and to some extent already does, include embedded distributed real-time computer control systems. Various levels of distribution/decentralization of the control system within the machine are conceptually possible, and the appropriate level of decentralization depends on a number of factors, including costs, real-time require- ments and dependability. The main purpose of this pa- per is to outline a ”decentralization” methodology for such applications. In the historic evolution of distributed control sys- tems, important advances have been made. However, despite the immense weight of research literature in related fields, there is still a lack of engineering meth- odology and tools for building advanced distributed real-time control systems. The authors of this paper believe that the reason for this, to a large extent, depends on the intrinsically multidisciplinary nature of real-time control applications. Multidisciplinarity necessitates an understanding across fields like com- puter engineering, computer science, control theory and mechanical engineering. Another problem is the wide applicability of distributed control approaches, from telecommunication and process-control, to machine-embedded control systems. This has led to a multitude of distributed system approaches and con- cepts, for applications which have widely varying characteristics, e.g. real-time requirements. It is worth noting that distributed and decentralized are used dif- ferently in computer and control engineering (compare ”distributed-parameter systems”). Distribu- tion can be seen as having three dimensions: hard- ware, data and control (Enslow, 1978). In this paper decentralization and distribution are used synony- mously according to this definition. There are a few basic hardware (resource) structures that can be considered for machine-embedded com- puter control systems. The traditional one is a central-