Model-Driven Development of Control System Software

The control software for large real-time systems such as radio-telescopes is generally developed from scratch on a customized basis. The cost and effort involved may be reduced by the use of existing frameworks, components and tools, however the conceptualization, design synthesis and realization of the system still requires a massive and complex engineering process. In this paper, we discuss the design of a model-driven framework that generates much of the control and monitoring software automatically based on higher-level specifications. The design of our framework is guided by the observation that for any system, independent of its specific purpose and elements, the control software performs the same set of functions: communicate, verify, coordinate and issue commands to various sub-systems (actuators); and collect, process, transmit, visualize and archive data produced by sub-systems (sensors). The machine control problem is formulated as a recursive hierarchical composition of control nodes, each of which achieves its function by coordinating its subsystems. Control software for each node is generated from specifications of its control logic, data acquisition and processing, and events detection and handling. We developed a prototype implementation of the concept for fusion reactor control, and validated its generality by reconfiguring the prototype to control the antennas of a radio-telescope. This data-driven approach to machine control software development has the promise to not only simplify initial system software development, but greatly reduce the time and effort for maintenance and evolution of the software over the system lifetime.