City Research Online

Abstract

For many systems, high confidence is required that they will never suffer an accident over an extended period of operation. Statistics of accident- or problem-free operation can give factual support for this confidence. But changes, to systems or to the way they are used, create problems for this part of dependability assurance. For instance, experience of safe operation before a design improvement should be still relevant for claims of safety after the improvement; but methods in current use do not show how much it should contribute to confidence in the latter. Thus quantitative assessment after changes may ignore (or instead overrate) large amounts of evidence, distorting decision making about system acceptance or evolution. To help with this problem, we extend previous work on integrating statistical evidence, from operation, with prior confidence, based on production and verification quality, that a design is free from design faults. Our extension also takes into account evidence of operation before the change, and confidence, derived from analysis, that a change did not degrade dependability. We apply “Conservative Bayesian Inference” (CBI) to allow probabilistic reasoning without specifying detailed prior distributions for the variables of interest, a serious difficulty in current use of Bayesian methods. We show: (i) that pre-change evidence can contribute substantially towards trusting the system post-change, especially while post-change experience is still limited; (ii) how this contribution depends on the strength of the analysis showing that the change improves, or does not affect, safety, and on other parameters; (iii) the limits to the advantages that pre-change evidence can bring.

Publication Type:	Conference or Workshop Item (Paper)
Publisher Keywords:	survival probability, software correctness, similarity arguments, conservative Bayesian inference, globally at least equivalent, field testing, safety critical systems, ultra-high reliability, no worse than existing system, proven in use
Subjects:	Q Science > QA Mathematics Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Departments:	School of Science & Technology School of Science & Technology > Computer Science School of Science & Technology > Computer Science > Software Reliability
SWORD Depositor:	Symplectic Administrator

Export

Downloads