Stochastic Analysis of Systems Exposed to Very Unlikely Faults

Abstract

When designing and operating any technical system, it is essential to take into account the possible faults that may occur during its operation. Dependability/reliability analysis lets us determine the level of redundancy that ensures continuity of service at an economically justified level of assurance. It ensures that all the faults are covered proportionately to the probability of their occurrence. Thus, the analysis tends to underemphasize the events that are very improbable, such as the simultaneous breakdown of all or almost all system components (e.g. due to a natural disaster). Such situations are addressed by risk analysis which combines the probability of events occurrence and their consequences. The use of straightforward stochastic modelling in this case is very difficult - Monte Carlo simulation requires huge runtimes to observe occurrences of such events. The solution is based on standard stochastic approach, modified by injecting artificially some unlikely events into the model. The usefulness of the proposed approach is demonstrated in two test studies: a discrete transport system and a Web based information system.