Computer Automated Multi-Paradigm Modeling in Control System Design Pieter J. Mosterman 1 Institute of Robotics and Mechatronics DLR Oberpfaffenhofen, D-82230 Wessling, Germany Pieter.J.Mosterman@dlr.de, http://www.op.dlr.de/~pjm Hans Vangheluwe School of Computer Science McGill University, Montreal, Canada H3A 2A7 hv@cs.mcgill.ca, http://www.cs.mcgill.ca/~hv Abstract The complete control system design effort involves many stages during which partial design tasks are com- pleted. Each of these tasks requires different modeling paradigms and different tools. Furthermore, the de- signed embedded control system entails a wide variety of implementation technologies that all require different specification formalisms. To handle such a multitude of modeling paradigms and different support tools: (i) a unifying generic standard language can be applied, and (ii) the required modeling paradigms can be modeled by a meta model using a shared meta language. An overview of the required parts and structure of a mod- eling environment and of the two approaches is given. The advantages with respect to multi-paradigm mod- eling are discussed. 1 Introduction The analysis and design of engineered systems involves expertise from many disciplines and entails a variety of implementation technologies (e.g., embedded software, microelectromechanical systems, analog circuits, and digital circuits) and the heterogeneous nature of these systems invariably combines with an architecture of different concurrent components that interact through continuous signals or discrete message passing. The corresponding complexity has led to the use of more formal approaches to system design through realization that apply dedicated modeling formalisms to differ- ent aspects and/or components of the system. Conse- quently, the complete system specification process com- bines several modeling, design, implementation, and re- alization paradigms such as differential equation mod- eling, continuous time signal processing, and discrete event controllers. Decomposition of the entire specifica- tion task allows teams of experts to concurrently work on their domain of expertise, e.g., control law design, simulation, optimization, modeling, and verification. 1 Pieter J. Mosterman is supported by a grant from the DFG Schwerpunktprogramm KONDISK. To comprehensively handle control system design in such a heterogeneous environment, multiple ap- proaches based on different paradigms have to be com- bined. In this paper, the following definition is used [23] “A modeling paradigm is a set of requirements that governs how any system in a particular domain is to be modeled. These modeling requirements specify the types of entities and relationships that can be modeled; how best to model them; entity and/or relationship at- tributes; the number and types of views or aspects nec- essary to logically and efficiently partition the design space; how semantic information is to be captured by, and later extracted from, the models; any analysis re- quirements; and, in case of executable models, run-time requirements.” A tool that ‘understands’ each of the corresponding formalisms (i.e., has a model of them) can be used to ensure consistency between different formalisms, allow for quick adaptation to changing needs, exchange infor- mation, and efficiently provide tailored modeling envi- ronments that are maximally constrained with respect to the domain of operation. For example, a designed control law that is automatically translated into its implementation, i.e., the hardware binding. Here the control language focuses on stability and other control characteristics, whereas the implementation has to deal with issues such as schedulability, reliability, and secu- rity, which requires different analysis formalisms. If consistency and cross coupling across these languages is ensured, implementation choices (e.g., the ‘time for space’ trade-off) can be conveniently conveyed back to the control design engineer. Multi-paradigm modeling is also critical for reconfig- urable systems as the supervising mechanisms that combine with a flexible control architecture are based on different modeling formalisms (even different plant models) and need to integrate with the control architec- ture. One solution is model integrated computing [25], which allows changes in the system model/specification and translates these automatically into software (or even reconfigures hardware). Control system design is achieved by using many soft-