Reliability estimation of safety-critical software-based systems using bayesian networks

    Research output: Book/ReportReport


    Due to the nature of software faults and the way they cause system failures new methods are needed for the safety and reliability evaluation of software-based safety-critical automation systems in nuclear power plants. In the research project "Programmable automation system safety integrity assessment (PASSI)", belonging to the Finnish Nuclear Safety Research Programme (FINNUS, 1999-2002), various safety assessment methods and tools for software based systems are developed and evaluated. The project is financed together by the Radiation and Nuclear Safety Authority (STUK), the Ministry of Trade and Industry (KTM) and the Technical Research Centre of Finland (VTT). In this report the applicability of Bayesian networks to the reliability estimation of software-based systems is studied. The applicability is evaluated by building Bayesian network models for the systems of interest and performing simulations for these models. In the simulations hypothetical evidence is used for defining the parameter relations and for determining the ability to compensate disparate evidence in the models. Based on the experiences from modelling and simulations we are able to conclude that Bayesian networks provide a good method for the reliability estimation of software-based systems.
    Original languageEnglish
    Place of PublicationHelsinki
    PublisherRadiation and Nuclear Safety Authority STUK
    Number of pages23
    ISBN (Print)951-712-449-X
    Publication statusPublished - 2001
    MoE publication typeD4 Published development or research report or study

    Publication series



    • safety
    • safety analysis
    • reliability analysis
    • bayesian belief networks
    • automation
    • programmable systems
    • software-based systems
    • reactor protection systems
    • nuclear reactor safety


    Dive into the research topics of 'Reliability estimation of safety-critical software-based systems using bayesian networks'. Together they form a unique fingerprint.

    Cite this