Scalable allocation of safety integrity levels in automotive systems

Azevedo, Luís Pedro da Silva

Computer science
December 2015

Thesis or dissertation


Rights
© 2015 Luís Pedro da Silva Azevedo. All rights reserved. No part of this publication may be reproduced without the written permission of the copyright holder.
Abstract

The allocation of safety integrity requirements is an important problem in modern safety engineering. It is necessary to find an allocation that meets system level safety integrity targets and that is simultaneously cost-effective. As safety-critical systems grow in size and complexity, the problem becomes too difficult to be solved in the context of a manual process. Although this thesis addresses the generic problem of safety integrity requirements allocation, the automotive industry is taken as an application example.

Recently, the problem has been partially addressed with the use of model-based safety analysis techniques and exact optimisation methods. However, usually, allocation cost impacts are either not directly taken into account or simple, linear cost models are considered; furthermore, given the combinatorial nature of the problem, applicability of the exact techniques to large problems is not a given. This thesis argues that it is possible to effectively and relatively efficiently solve the allocation problem using a mixture of model-based safety analysis and metaheuristic optimisation techniques. Since suitable model-based safety analysis techniques were already known at the start of this project (e.g. HiP-HOPS), the research focuses on the optimisation task.

The thesis reviews the process of safety integrity requirements allocation and presents relevant related work. Then, the state-of-the-art of metaheuristic optimisation is analysed and a series of techniques, based on Genetic Algorithms, the Particle Swarm Optimiser and Tabu Search are developed. These techniques are applied to a set of problems based on complex engineering systems considering the use of different cost functions. The most promising method is selected for investigation of performance improvements and usability enhancements. Overall, the results show the feasibility of the approach and suggest good scalability whilst also pointing towards areas for improvement.

Publisher
Department of Computer Science, The University of Hull
Supervisor
Parker, David (David James); Papadopoulos, Yiannis
Qualification level
Doctoral
Qualification name
PhD
Language
English
Extent
5 MB
Identifier
hull:13618
QR Code