Adaptive Fault-Tolerance for Cyber-Physical Systems

Cyber-physical systems are increasingly used in lifecritical applications, where the probability of catastrophic failure has to be kept below very low levels. Massive fault-tolerance has been used to mask failure to achieve such low levels. However, fault-tolerance is expensive. We argue here that the fault-tolerance needs of an application change depending on its current position in its state space and the range of control inputs that can be applied. We illustrate the applicability of such an approach using the inverted pendulum as a case-study, and discuss its potential benefits.