State-space encoding driven error resilience in control systems and circuits

Abstract

The objective of the proposed research is to develop methodologies, support algorithms and software-hardware infrastructure for detection and diagnosis of parametric failures, transient soft errors and security attacks in linear and nonlinear circuits and systems for sensing and control. This research is motivated by the proliferation of autonomous sense-and-control real-time systems, such as intelligent robots and self-driven cars, that must maintain a minimum level of performance in the presence of unavoidable electro-mechanical degradation of system-level components in the field as well as external security attacks. A key focus is on rapid recovery from the effects of such anomalies and impairments with minimal impact on system performance while maintaining low implementation overhead as opposed to traditional schemes for recovery that rely on duplication or triplication. Real-time detection and diagnosis techniques are investigated and rely on analysis of state-space encoding based check signatures. For on-line error detection and diagnosis in control systems, linear and nonlinear state space encodings of the system behavior are analyzed in real-time. Recovery is initiated using guided reinforcement learning algorithms that determine how best the system should be controlled in the presence of the diagnosed performance impairments. For cyber-physical systems, these state-space encodings are used to detect malicious security attacks and to diagnose the affected components swiftly. These checks are utilized for fast recovery from such attacks while avoiding catastrophic system failure. Further research in this area will pave the way for successful deployment of self-healing autonomous systems and resilient cyber-physical systems.