Using Abstraction in Modular Verification of Synchronous Adaptive Systems

Authors Ina Schaefer, Arnd Poetzsch-Heffter



PDF
Thumbnail PDF

File

OASIcs.TrustworthySW.2006.699.pdf
  • Filesize: 375 kB
  • 14 pages

Document Identifiers

Author Details

Ina Schaefer
Arnd Poetzsch-Heffter

Cite AsGet BibTex

Ina Schaefer and Arnd Poetzsch-Heffter. Using Abstraction in Modular Verification of Synchronous Adaptive Systems. In Workshop on Trustworthy Software. Open Access Series in Informatics (OASIcs), Volume 3, pp. 1-14, Schloss Dagstuhl – Leibniz-Zentrum für Informatik (2006)
https://doi.org/10.4230/OASIcs.TrustworthySW.2006.699

Abstract

Self-adaptive embedded systems autonomously adapt to changing environment conditions to improve their functionality and to increase their dependability by downgrading functionality in case of fail- ures. However, adaptation behaviour of embedded systems significantly complicates system design and poses new challenges for guaranteeing system correctness, in particular vital in the automotive domain. Formal verification as applied in safety-critical applications must therefore be able to address not only temporal and functional properties, but also dynamic adaptation according to external and internal stimuli. In this paper, we introduce a formal semantic-based framework to model, specify and verify the functional and the adaptation behaviour of syn- chronous adaptive systems. The modelling separates functional and adap- tive behaviour to reduce the design complexity and to enable modular reasoning about both aspects independently as well as in combination. By an example, we show how to use this framework in order to verify properties of synchronous adaptive systems. Modular reasoning in com- bination with abstraction mechanisms makes automatic model checking efficiently applicable.
Keywords
  • Dependable Embedded Systems
  • Self-Adaptation
  • Abstraction
  • Modular Verification

Metrics

  • Access Statistics
  • Total Accesses (updated on a weekly basis)
    0
    PDF Downloads