Abstract
This paper presents a new way of applying Bayesian assessment to systems, which consist of many components. Full Bayesian inference with such systems is problematic, because it is computationally hard and, far more seriously, one needs to specify a multivariate prior distribution with many counterintuitive dependencies between the probabilities of component failures. The approach taken here is one of decomposition. The system is decomposed into partial views of the systems or part thereof with different degrees of detail and then a mechanism of propagating the knowledge obtained with the more refined views back to the coarser views is applied (recalibration of coarse models). The paper describes the recalibration technique and then evaluates the accuracy of recalibrated models numerically on contrived examples using two techniques: u-plot and prequential likelihood, developed by others for software reliability growth models. The results indicate that the recalibrated predictions are often more accurate than the predictions obtained with the less detailed models, although this is not guaranteed. The techniques used to assess the accuracy of the predictions are accurate enough for one to be able to choose the model giving the most accurate prediction.
Highlights
Off-the-shelf (OTS) components containing software are used widely for both development of new systems and upgrading existing ones as part of their maintenance
Unless software is free of design faults a good reliability record in the past in a different operational environment does not guarantee that it will work reliably in the new operational environment, the past record can be used to back a priori expectations for reliable operation in the new environment
- Even if black-box is adequate as a model for the intended new operational environment, one needs to build a prior distribution for system reliability, e.g. system probability of failure on demand from the available evidence about reliability of the components used in the system
Summary
Off-the-shelf (OTS) components containing software are used widely for both development of new systems and upgrading existing (i.e. legacy) ones as part of their maintenance. This paper applies Bayesian dependability assessment when a non black-box model of a system built with components is used. - Even if black-box is adequate as a model for the intended new operational environment, one needs to build a prior distribution for system reliability, e.g. system probability of failure on demand (pfd) from the available evidence about reliability of the components used in the system In this case one would ideally like to use all the evidence available in the prior– good reliability record about the part of the system which remains unchanged and a good record, possibly in a different operational environment, about the new components added to the system. A central part in our approach is measuring the accuracy of the predictions obtained with the coarse and recalibrated view and choosing the one that is most accurate for the data at hand For this we adapt techniques developed in software reliability growth modelling and discuss their suitability to our context.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
