Networked control systems tolerant to faults

Control Systems in the last decade have seen a rapid increase in the integration of multi-loop distributed and networked systems, often containing a large number of dynamically interacting uncertain components. This is the result of a convergence of (a) digital network technology, (b) embedded systems, (c) failure analysis and online diagnosis. These systems, referred to as networked control systems (NCS), are inherently complex and involve a hierarchy of local and global embedded control and diagnostic structures that are difficult to model because of the complexity of their behaviours. Research developments have led to the integration of classic control problems, such as robustness, along with additional features of network properties, e.g. time-delays, packet loss, channel throughputs. This has resulted in a move away from classical definitions of uncertainty and robustness, as used in classical control system situations, to ones where consideration has to be given to the lack of the necessary information at places and times where decisions or actions have to be taken. These systems have given rise to significant new control challenges. The presence of uncertain components (e.g. having variable time-delay or of uncertain bandwidth) means that the complex NCS needs to have an architecture that can support fault-tolerant properties, i.e. it needs to be able to reconfigure autonomously and reliably, and be able to self-repair and accommodate to anomalies such as uncertain behaviours and faults. This is the subject of fault-tolerance of an NCS. The point-to-point control architectures that are prevalent in much of the NCS research do not easily support reconfiguration, subsequent to fault events. Each sub-system of the network has to be considered (designed, studied) in a degraded context for which autonomous behaviour is required: it may be partially or totally isolated due to a bad quality of service of the network. In this case, the sub-system must first run to a safe state before being able to run with more autonomy. The concept of ‘plug-and-play’, where components are added or removed with flexibility, can be important for fault-tolerant NCS. Procedures for fault detection and isolation (FDI) used for single-point control architectures typically make use of either signal-based or model-based methods. The latter use the concept of