Publications
Hybrid Processing of Measurable and Subjective Data
Conventional systems surety analysis is basically restricted to measurable or physical-model-derived data. However, most analyses, including high-consequence system surety analysis, must also utilize subjective information. In order to address this need, there has been considerable effort on analytically incorporating engineering judgment. For example, Dempster-Shafer theory establishes a framework in which frequentist probability and Bayesian incorporation of new data are subsets. Although Bayesian and Dempster-Shafer methodology both allow judgment, neither derives results that can indicate the relative amounts of subjective judgment and measurable data in the results. The methodology described in this report addresses these problems through a hybrid-mathematics-based process that allows tracking of the degree of subjective information in the output, thereby providing more informative (as well as more appropriate) results. In addition, most high consequence systems offer difficult-to-analyze situations. For example, in the Sandia National Laboratories nuclear weapons program, the probability that a weapon responds safely when exposed to an abnormal environment (e.g., lightning, crush, metal-melting temperatures) must be assured to meet a specific requirement. There are also non-probabilistic DOE and DoD requirements (e.g., for determining the adequacy of positive measures). The type of processing required for these and similar situations transcends conventional probabilistic and human factors methodology. The results described herein address these situations by efficiently utilizing subjective and objective information in a hybrid mathematical structure in order to directly apply to the surety assessment of high consequence systems. The results can also improve the quality of the information currently provided to decision-makers. To this end, objective inputs are processed in a conventional manner; while subjective inputs are derived from the combined engineering judgment of experts in the appropriate disciplines. In addition to providing output constituents (including portrayal of uncertainty) corresponding to combination of these input types, their individual contributions to the resultant uncertainty are determined and provided as part of the output information. Finally, the safety assessment is complemented by a latent effects analysis, facilitated by soft-aggregation accumulation of observed operational constituents.