Run-time anomaly detection and mitigation in information-rich cyber-physical systems

Next generation space missions require autonomous systems to operate without human intervention for long periods of times in highly dynamic environments. Such systems are vulnerable to software and/or hardware failures due to unexpected internal or external factors. Moreover, small anomalies, if not detected and isolated in a timely manner, can cascade through the system resulting in catastrophic outcomes, especially in highly dynamic missions where failsafe is not an option. This signifies the need for effective methods for integrated system health management, automated data analysis for decision-making and verification and validation. The objective of this project is to develop the scientific foundation and associated algorithmic tools for synthesis of decentralized passive and active monitors for provably correct run-time verification and validation of sensor-rich networked cyber-physical systems. The potential benefits of the proposed research include (i) reductions in the design time of next generation space systems by automating synthesis of monitoring algorithms instead of hand-coded built-in tests, (ii) reductions in system cost by the potential to replace hardware redundancy with software-based solutions, (iii) increase in the time systems operate reliably by enabling timely detection of anomalies and reducing their cascading effects.
In particular, we have pursued research in the following three thrusts and investigated the corresponding questions listed below:
(T1) Quantifying deviations from the nominal: How to model physics and adaptive/learning software systems in a unified way? How can we compute provable error bounds at run-time to quantify how well the measurements collected from the system and nominal system model match and how well the system meets its specification? How can we use such bounds to guide generation and testing of hypothesis related to anomalies?
(T2) Decentralized implementations: Given formal behavioral specifications for a cyber-physical system, how can we assess their monitorability and decentralized monitorability with a given set of sensors? Are there structural properties of systems, measurements (i.e., sensor types/locations), and specifications that facilitate decentralized anomaly detection?
(T3) Dynamic estimation strategies: Can we synthesize strategies that can utilize available actuation capabilities to improve anomaly detection by adaptively reconfiguring the system, and correlating the measurements and the prior information while minimally interfering with the system operation? Can we identify measurement and actuation capabilities that are most critical for estimation accuracy? The proposed run-time verification and validation framework for adaptive autonomous systems contributes to “robotics and autonomous systems” technical area identified in 2015 NASA Technology Roadmap, with potential crosscutting technologies for system health management, automated data analysis for decision making and verification and validation of complex adaptive systems.