Data Fusion: A decision analysis tool that quantifies geological and parametric uncertainty

摘要

Engineering projects such as siting waste facilities and performing remediation are often driven by geological and hydrogeological uncertainties. Geological understanding and hydrogeological parameters such as hydraulic conductivity are needed to achieve reliable engineering design. Information from non-invasive and minimally invasive data sets offers potential for reduction in uncertainty, but a single data type does not usually meet all needs. Data Fusion uses Bayesian statistics to update prior knowledge with information from diverse data sets as the data is acquired. Prior knowledge takes the form of first principles models (e.g., groundwater flow) and spatial continuity models for heterogeneous properties. The variability of heterogeneous properties is modeled in a form motivated by statistical physics as a Markov random field. A computer reconstruction of targets of interest is produced within a quantified statistical uncertainty. The computed uncertainty provides a rational basis for identifying data gaps for assessing data worth to optimize data acquisition. Further, the computed uncertainty provides a way to determine the confidence of achieving adequate safety margins in engineering design. Beyond design, Data Fusion provides the basis for real time computer monitoring of remediation. Working with the DOE Office of Technology (OTD), the author has developed and patented a Data Fusion Workstation system that has been used on jobs at the Hanford, Savannah River, Pantex and Fernald DOE sites. Further applications include an army depot at Letterkenney, PA and commercial industrial sites.