MODEL DEVELOPMENT UNDER UNCERTAINTY VIA CONJOINT ANALYSIS

Model development decisions are critical in the early phases of engineering design. Engineering models serve as representations of reality that help designers understand input/output relationships, answer 'what-if' questions, and find optimal design solutions. Upon making model development decisions, the designer commits a large percentage of the costs associated with reaching design goals/objectives. The decisions dictate cost-drivers such as experimental setups and computation time. Unfortunately, the desire to develop the most accurate model competes with the desire to reduce costs. The designer is ultimately required to make trade-offs between attributes when choosing the best model development decision. Hence it is critical to develop tools for selecting the model development decision that appropriately balances trade-offs. A framework is proposed for model development decision-making. Conjoint Analysis (CA) is implemented in order to handle trade-offs among attributes. Thus, the framework can be used to make optimal decisions based on the assessment of multiple attributes. Moreover, the framework addresses the uncertainty that exists early in model design. Imprecision in model parameters are estimated and propagated through the model. In particular, the proposed decision framework is employed to select the optimal model development decision with respect to the final phase of experimentation. Preference intervals are evaluated in order to choose which final experimentation to perform. The decision framework proves to be. useful for making model development decisions under uncertainty by considering the preference of multiple attributes and the imprecision of said attributes that is prevalent in early model development phases.