Self-Diagnosis and Self-Planning with Constraint-based Hybrid Models

Many of today’s embedded systems – such as automobiles, automated factories or chemical plants – are an increasingly complex mixture of hardware components and embedded control software, showing both continuous (vehicle dynamics, silo fill levels) and discrete (software) behavior. As complexity grows, handling such systems, e.g. diagnosing and repairing faults, becomes harder. This problem is addressed by model-based reasoning, which enhances these systems with self-diagnosis and self-planning capabilities based on Hidden Markov Models (HMMs) of their structure and behavior. At the heart of these capabilities are the problems of estimating the internal state and automatically plan intelligent control (re)actions. However, HMMs cannot model continuous behavior. In my thesis, I address the stated problems for hybrid systems with an integrated approach combining concepts from AI (constraint optimization, HMM reasoning), fault diagnosis in hybrid systems (stochastic abstraction of continuous behavior), and hybrid systems verification (hybrid automata).