Towards Resilience by Self-Adaptation of Industrial Control Systems

Resilience is a critical quality of future Industrial Control Systems (ICS). The ability to detect and react to unanticipated attacks, bugs, and failures is crucial. Self-adaptation can provide this ability, yet it is difficult to achieve in safety-critical real-time systems, since strict safety and timing requirements must be guaranteed. Recent results indicate that automated adaptation of ICS using the IEC 61499 is possible, however it has not been analyzed how much dynamic adaptation can contribute to overall system resilience. In this paper, we analyze how dynamic adaptation can be embedded into industrial control architectures, and quantify its advantage over a traditional restart. We propose a self-adaptive architecture using the MAPE-K model and merge it with the existing models for ICS. Using measurements on a real system, we estimate the expected adaptation time of selected adaptation scenarios and calculate the loss of productivity depending on the reaction time and adaptation complexity. The results show that using current dynamic adaptation mechanisms, minor to moderate adaptations can be completed within 10 ms, while larger adaptations can take up to a second from initialisation to cleanup. The resilience gain is larger the faster the reaction is initiated, which indicates that once dynamic adaptation is available, a faster detection and decision-making becomes more important. Dynamic adaptation can provide ICS the means to evolve and react rapidly, preparing them for an agile, flexible, and resilient future.