Delta utility models in decision-based design

Selection of the most preferred design from a group of similar design alternatives is an important step in decision-based design. The designer need not only consider perceptions of the likelihood of a design alternative achieving predicted performance attributes, but also make tradeoffs between incommensurable attributes. The uncertainty in predicted performance attributes make it difficult for the designer to make rational decisions on the selection. This dissertation presents an original modeling framework to support selections between similar competing designs. Delta Performance Model (DPM) and Delta Utility Model (DUM) are two components of the framework. DPM is a useful model on distinguishing performance attributes which are probability distributions. And DUM is a utility model which provides preference measurements when trade-offs exist among performance attributes. The Delta Performance Model (DPM) explicitly considers uncertainty and utilizes the fact that some uncertainty and modeling error is the same for both candidate designs. It is created to produce information on the relative performance of one design with respect to another design. The output of DPM---relative performance attributes---eliminate some portions of model error and uncertainty which are the same on all alternatives, and provide a clear rank of performance attributes for similar designs. The Delta Utility Model (DUM) is a multi-attribute decision model which is used to help designer making decisions when trade-offs exist in design tasks. The utility theory is applied to quantify preferences in the model. This involves assessing single attribute delta utilities which represents the decision maker\u27s preferences concerning the possible relative attributes, and constructing a multi-attribute delta utility function which represents the decision maker\u27s preferences concerning combination and interaction of all single attribute delta utility functions. The multi-attribute delta utility function evaluates and assigns an expected utility value for each design alternative based on performance attributes of that alternative. Uncertainty is represented and managed by the probability theory in the dissertation. The bounding range of design variables and environmental factors (exogenous variables) and their distributions are assessed and estimated in the early stage of the decision making process. These random variables are incorporated and propagated in DPM and DUM through Monte Carlo simulations. Some design cases are studied and shown that the combination of DPM and DUM gives an effective tool for evaluation of similar designs. Compared to the traditional methods, the new approach explicitly considers uncertainty and utilizes the fact that some uncertainty and modeling error is the same for both candidate designs to decrease uncertainty and amplify the utility difference between design alternatives. Therefore, it allows the designer to make better decisions in situations where two or more similar designs are being considered