Reliability Assessment and Optimization under Uncertainty in the Dempster-Shafer Framework

Epistemic uncertainty has been a subject heavily discussed in recent years. Where uncertainty of estimations for long time has been disregarded, it is now viewed as an inherent property of the system model. This work demonstrates how different data types like expert estimations and manufacturer’s data with various degrees of epistemic uncertainty are acquired and aggregated in a coherent framework. Using this model we show how to improve reliability using multi-objective optimization. A Pareto-based evolutionary algorithm is applied to find a choice of nondominated solutions. A new strategy for biasing the search to desired objectives without loosing diversity is presented. The user can select a posteriori between solutions covering a wide range of the objective space but clustering in the specified area.