Signal feature analysis for dynamic anomaly detection of components in embedded control systems

Abstract

Embedded Control Systems (ECS) are getting increasingly complex for the realization of Cyber-Physical Systems (CPS) with advanced autonomy (e.g. autonomous driving of cars). This compromises system dependability, especially when components developed separately are integrated. Under the circumstance, dynamic anomaly detection and risk management often become a necessary means for compensating the insufficiencies of conventional verification and validation, and architectural solutions (e.g. hardware redundancy). The aim of this work is to support the design of embedded software services for dynamic anomaly detection of components in ECS, through probabilistic inference methods (e.g. Hidden Markov Model - HMM). In particular, the work provides a method for classifying the signal features of operational sensors and thereby applies Monte-Carlo sensitivity analysis for eliciting the probabilistic properties for error estimation. Such approach, based upon a physical model, reduces the dependency on empirical data for bringing about confidence on newly developed components.