Scalable Reliability Analysis by Lazy Verification

Abstract

This paper presents an iterative method to analyse system reliability models. The key idea is to analyse a partial state space of a reliability model in a conservative and an optimistic manner. By considering unexplored states as being always operational or, dually, already failed, our analysis yields sound upper-and lower-bounds on the system's reliability. This approach is applied in an iterative manner until the desired precision is obtained. We present details of our approach for Booleanlogic driven Markov processes (BDMPs), an expressive fault tree variant intensively used in analysing energy systems. Based on a prototypical implementation on top of the probabilistic model checker Storm, we experimentally compare our technique to two alternative BDMP analysis techniques: discrete-event simulation obtaining statistical bounds, and a recent closed-form technique for obtaining pessimistic system lifetimes. Our experiments show that mostly only a fragment of the state space needs to be investigated enabling the reliability analysis of models that could not be handled before.S. Khan-supported by a HEC-DAAD scholarship. 1 Known as confidence intervals.

Graph Transformation

“…We will show that various formal methods can effectively be used to: a) give a formal semantics to fault-tree dialects using Petri nets [5,7] b) simplify fault trees prior to their expensive analysis using graph rewriting [4] c) prove such rewriting correct with theorem proving [2] d) analyse the simplified fault trees by probabilistic model checking [10], and e) treat gigantic models by an iterative "generate partial state-space and verify" paradigm that provides sound bounds [8,10].…”

ARCH-COMP22 Category Report: Stochastic Models

“…• All the above cited tools are "industry proof" tools, used in real studies of complex systems such as nuclear power plants, telecommunication and electrical networks... KB3 is commercially available under the name RiskSpectrum ModelBuilder. A new tool, still a prototype, is available to process FIGARO Markovian models: it is based on the STORM probabilistic model checker, cf [50,49]. This open source tool, called STORM-Figaro, is now available on github.…”