Property-Based Fault Detection and Diagnosis for Safety-Critical Cyber-Physical Systems
thesisposted on 27.11.2018 by Yao Feng
In order to distinguish essays and pre-prints from academic theses, we have a separate category. These are often much longer text based documents than a paper.
A large variety of modern applications including autonomous cars, energy systems and medical devices are referred to as Cyber-Physical Systems (CPSs). Runtime monitoring of CPSs is an important method to verify if an execution of the system violates a given correctness property defined over the system state. A threshold based monitor evaluates the rejection probability, which is the probability that the system is in one of the states that violate the property, against a predefined threshold. The evolution of property automata, which model the correctness properties is driven by the internal state of the Probabilistic Hybrid Automaton (PHA), which is the mathematical dynamical model of the CPS. Particle Filter (PF), an Monte Carlo implementation of Bayes Filter, is employed for the estimation of the internal hybrid state. An interesting class of CPS are the concurrent CPSs with multiple components. The interaction among the components is studied in the thesis. An PF algorithm is proposed to reduce the complexity for a type of interaction which would cause state space explosion. An important application of runtime monitoring is Fault Detection and Diagnosis (FDD). It extends the the faults studied in FDD to a more general form, i.e., the violation of properties. In the proposed property-based fault detection and diagnosis (PB-FDD), hierarchical faults are defined as system level faults and component level faults. Furthermore, the causes for particle inconsistency which would fail the belief propagation are further studied. Besides the particle depletion occurred during particle propagation step in PF, another hypothesis is system model inconsistency, which corresponds to an unknown component level fault. Based on the distribution of the importance weights in PF, two hypothesis tests are formalized to first detect particle inconsistency and further diagnose the causes.