Ph.D. Proposal Presentation by Scott J. Duncan
Friday, November 5, 2004

(Dr. Bert Bras, Chair)

"Including Severe Uncertainty into Product Life Cycle Design Using Information-Gap Decision Theory"


Designers assessing a product’s performance over its life cycle frequently lack the data needed to adequately describe the distribution or bounds of severe parameter uncertainty, especially that associated with time-distant aspects of environmental performance. The trustworthiness of assessment results is undermined when unwarranted or unjustifiable assumptions are used to fit uncertainty into a representation needed by a normative decision theoretic framework. An alternative approach to modeling and evaluating sparse uncertainty information has recently emerged in information-gap decision theory. This theory requires no probabilistic data and still allows an assessment of how an expanding horizon of uncertainty (i.e., deviation from a norm) affects the minimally achievable reward, or output of a system model. Uncertain variables are represented as convex, nested sets with parameterized bounds that expand outward, forming the “gap” between a nominal value and the extreme deviation that could be possible. The main tool of the theory is a response plot of minimally acceptable performance versus the level of severe uncertainty to which the system is robust. This plot enables a practitioner to visualize the tradeoffs between robustness and performance to help elicit preferences for either. Although utilizing sparse uncertainty data in this way has a less powerful effect on decision maker preferences than a probabilistic analysis might, in some cases that effect is sufficient to enable selection between alternatives.

The research at hand will aim to both extend and build a method around information-gap decision theory with the goal of effectively and efficiently utilizing sparse uncertainty data from the product life cycle in environmental risk assessment for multi-criteria design. Several activities are involved in developing this method, including (1) characterizing the range of applicability of the basic theory for simple life cycle design applications, (2) applying the theory to more complex problems featuring multiple input variables, multiple performance objectives, and hierarchies through which the result of uncertainty analysis propagates, and (3) using results of the theory as a proxy for valuing uncertainty data, helping a practitioner prioritize what further data to seek. Method validation will be achieved using a variety of case studies all involving life cycle waste generation and, where possible, environmental impact.