Probabilistic property prediction

Uncertainty must be characterized accurately for engineering applications because it arises from a number of sources each of which are exacerbated as the complexity increases. Having limited data also intensifies concerns about the uncertainty. Historically the design of structural components for high reliability applications has required an extensive test program to validate the design and service of the component. With the development of new materials and alloys and with an increased concern for safety, validation of a design is even more difficult. In order to meet regulatory demands for components for extended service and yet reduce the costs of preproduction, an approach which incorporates extensive scientific modeling with a limited experimental effort is desired. To that end, this effort proposes a methodology, which integrates limited data with science based modeling, so that the uncertainty is represented adequately. The approach combines classical Bayesian methods and model synthesis with data. The approach is illustrated using the yield strength of a typical turbine disk alloy. Fusion of science based modeling with data greatly improves estimation and prediction, which reduces the uncertainty. Even crude models are more beneficial than statistical data analysis alone. The approach allows for a significant reduction in data in contrast to purely statistical approaches.