On scheduling tasks in reliable real-time control systems

Feedback control is one of the most common applications of real time systems. However, the design of controller frequencies, task scheduling and reliability engineering are often done separately, resulting in suboptimal results. The article provides an overview of an integrated approach to reliable real time controller design by optimizing the system control performance subject to schedulability and software reliability constraints. In control applications, the software reliability challenges goes beyond the specification, design, development and verification considerations. Advanced control techniques such as a neural net learns by example and can perform sophisticated nonlinear control. Indeed, the properties of certain advanced controllers can be difficult to analyze and verify. This problem can be addressed by using analytically redundant controllers, where the sophisticated but less reliable controller is "supervised" by a simple and reliable controller. An example of using analytically redundant controllers to enhance system reliability is the Boeing 777 flight control, where the normal controller is the new 777 controller, whereas the secondary controller is based on the well understood 747 control technology. The aircraft's state under the normal controller should be within the stability envelope of the 747 controller (Y.C. Yeh, 1995).