ABSTRACT
The design of dependable systems and reduction of economic costs have been viewed as conflicting goals. Traditional dependable design approaches replicate hardware resources to improve faulttolerance. However, the aggregation of hardware resources leads to more failure sources and higher costs. Therefore, one possible direction to construct dependable systems and reduce economic costs is to reuse hardware resources to provide compatible functionalities through system health management. We focus on the design of distributed networked control systems operating in massively networked scenarios. In these scenarios there are many different hardware resources connected to a network for different purposes. Furthermore, it has the particularitythat subsets of their system functions are replicated throughout the system structure e.g., trains or buildings.
In this framework we propose an adaptive dependable design methodology including modelling and analysis steps. The objective of the approach is to meet functional and dependability requirements by reusing hardware resources. Firstly, we identify systematically hardware resources capable of providing additional functionalities besides their nominal functionality. Secondly, we obtain reconfiguration strategies to recover from failures and identify single points of failure. Then, in order to make system recovery possible, our design methodology addresses health management strategies covering fault detection and reconfiguration implementations. Finally, we extract systematically the failure probability of the solution architecture by generating component fault trees. To this end, our approach accounts for system failures which prevent the system from recovering when it has to do so.
The originality of this approach comes from the explicit and systematic identification of non-apparent redundancies by using existing hardware resources in massively networked scenarios. Consequently, in order to construct an adaptive dependable system methodically the necessary and sufficient health management mechanisms and implementations have been characterized and evaluated.
