Reliability analysis of systems based on software and human resources

Safety-critical systems require an assessment activity to verify that they are able to perform their functions in specified use environments. This activity benefits from evaluation methods that consider these systems as a whole and not as the simple sum of their parts. Indeed, analysis of accidents involving such systems has shown that they are rarely due to the simple failure of one of their components. Accidents are the outcome of a composite causal scenario where human, software, and hardware failures combine in a complex pattern. Unfortunately, dependability analysis and evaluation of safety critical systems are usually based on techniques and methods that consider human and computer separately, and whose results can hardly be integrated. The analogies between the processes of: (1) software-reliability growth due to testing and the related fault removal; (2) improvement of man-machine interface due to preliminary operative feedback; and (3) improvement of operator performance due to learning activity; all suggest a common evaluation approach. Only the first one of these processes is currently modeled using mathematical methods. This paper extends these methods to study the reliability-growth process of other system components: operator and man-machine interfaces. To study the feasibility of the approach, this paper analyzes the results of an experiment in which the reliability of a system is evaluated using trend analysis and reliability-growth models. The evaluation concerns the graphic man-machine interface and the operators, and could easily be extended to the software control system. The experimental results show that trend analysis and reliability-growth models could be complementary to the qualitative evaluation performed within the cognitive science approach. They could offer a quantitative support especially when their information is based on analysis of average values. In this case they could assist several decisions during operator training and especially during interface design, when comparing the effect of different possible interfaces on operator behavior. Moreover they can support the share of the same tools and of the related know-how between the fields of human and software dependability