Computer Science Graduate Seminar

Thursday, April 28, 2022, 10:30am

Dynamic Fault Trees: Semantics, Analysis and Applications

      Meeting ID: 997 0976 8339
      Password: 975390

Abstract

Safe and reliable systems are crucial in today’s society. Fault trees are a prominent and widely-used model to assess and improve the reliability of systems. Fault trees model how component failures propagate through a system and lead to a failure of the overall system. Dynamic fault trees (DFTs) are an extension of (static) fault trees and allow more modelling flexibility by introducing dynamic gates, spare management, functional dependencies and failure restrictions.

In this presentation, we investigate dynamic fault trees in detail and consider three main aspects: (1) the precise semantics of DFTs, (2) the analysis of DFTs by model checking techniques, and (3) the application of DFTs, for example in the railway domain.

We first specify the semantics of dynamic fault trees in terms of generalized stochastic Petri nets (GSPNs). We investigate multiple semantic questions resulting from the combination of DFT elements. Our resulting GSPN framework subsumes the major existing DFT semantics and allows to pinpoint their differences.

Secondly, we present analysis techniques for DFTs based on probabilistic model checking. We introduce several (orthogonal) optimisation techniques which exploit symmetries, irrelevant failures and independent subtrees to improve the state-space generation times. We also show an approximation algorithm based on partial state-space exploration. All presented approaches are implemented in the open-source model checker Storm and evaluated on a DFT benchmark suite. The evaluation shows that our tool Storm-dft is state-of-the-art for DFT analysis.

Third, we present the application of DFTs in the railway domain. The case study considers train routing options in railway station areas in terms of available infrastructure elements. We analyse how switch failures impact the potential train routes in a station and determine the most critical components.

 

The computer science lecturers invite interested people to join.