Statistical Model Checking of Rich Models and Properties

Software is in increasing fashion embedded within safety- and business critical
processes of society. Errors in these embedded systems can lead to human
casualties or severe monetary loss. Model checking technology has proven
formal methods capable of finding and correcting errors in software. However,
software is approaching the boundary in terms of the complexity and size that
model checking can handle. Furthermore, software systems are nowadays more
frequently interacting with their environment hence accurately modelling such
systems requires modelling the environment as well - resulting in undecidability
issues for the traditional model checking approaches.
Statistical model checking has proven itself a valuable supplement to model
checking and this thesis is concerned with extending this software validation
technique to stochastic hybrid systems. The thesis consists of two parts: the first
part motivates why existing model checking technology should be supplemented
by new techniques. It also contains a brief introduction to probability theory
and concepts covered by the six papers making up the second part. The first two
papers are concerned with developing online monitoring techniques for deciding
if a simulation satisfies a property given as a WMTL[a;b] formula. The following
papers develops a framework allowing dynamical instantiation of processes,
in contrast to traditional static encoding of systems. A logic, QDMTL, is
developed to express properties of these dynamically evolving systems. The
fifth paper shows how stochastic hybrid automata are useful for modelling
biological systems and the final paper is concerned with showing how statistical
model checking is efficiently distributed. In parallel with developing the theory
contained in the papers, a substantial part of this work has been devoted to
implementation in Uppaal SMC.