1
Uncertainty Modeling Enhancement Concepts in Quantitative Risk Assessment Methodology
Robert (Bob) Baker; A-P-T Research, Inc.; Huntsville, AL, USA
John Tatom; A-P-T Research, Inc.; Huntsville, AL, USA
Key Words: Explosives Risk, Analytical Model, Uncertainty, Statistical Distribution
Abstract
In modeling future explosives risk, many uncertainties must be considered. Since 2002, the
Department of Defense Explosives Safety Board (DDESB) has been developing and refining
models incorporating uncertainty estimation into its approved explosives risk methodology.
Various reviews of the methodology have identified quantitative risk assessment modeling
enhancements with potential to improve the estimation of explosives risk uncertainties. The
paper reports the findings of investigations into the feasibility of these enhancements to improve
analytical models used to estimate explosives risks and the uncertainties associated with these
estimates.
Introduction
Since 1997, the DDESB has developed and continuously improved its approved methodology for
estimating the risks posed by assembly and storage of bulk explosives and explosive munitions.
This risk-based methodology has been used by the DDESB to approve explosives safety site
plans in scenarios that cannot comply with Quantity Distance (QD) requirements.
In 2003, a second-generation risk uncertainty model was developed and integrated into the
DDESB methodology. This methodology was reported at the 2004 DDESB Explosive Safety
Conference (Ref. 1, 2) and was fully documented in DDESB Technical Paper (TP) 14 (Ref. 3).
During 15 years of use, characterization, evolution, and internal/external reviews, a few features
of the original uncertainty methodology have been identified as candidates for improvement.
These features include:
• Introducing large uncertainties can push the calculated Expected Fatality (Risk) estimate
unreasonably high.
• Modeling as lognormal element distributions that represent probabilities (e.g., P
e
, t, P
f|e
)
produces distributions that extend beyond the acceptable range of [0,1]. (Ref. 4)
• Assigning the point estimates generated by the science algorithms as the medians of
elemental distributions is counter to the intent of some algorithm developers. In these
cases, the risk estimate is increased without cause.
In 2017, the DDESB sponsored a Risk Assessment Program Team-led effort to explore the
viability of changes needed to implement the improvements recommended by various reviews.
This paper reports the approach and findings (Ref. 5) of the recently completed effort.
