Crowdsourcing for Engineering Design: Objective Evaluations and Subjective Preferences

Crowdsourcing enables designers to reach out to large numbers of people who may not have been previously considered when designing a new product, listen to their input by aggregating their preferences and evaluations over potential designs, aiming to improve ``good'' and catch ``bad'' design decisions during the early-stage design process. This approach puts human designers--be they industrial designers, engineers, marketers, or executives--at the forefront, with computational crowdsourcing systems on the backend to aggregate subjective preferences (e.g., which next-generation Brand A design best competes stylistically with next-generation Brand B designs?) or objective evaluations (e.g., which military vehicle design has the best situational awareness?). These crowdsourcing aggregation systems are built using probabilistic approaches that account for the irrationality of human behavior (i.e., violations of reflexivity, symmetry, and transitivity), approximated by modern machine learning algorithms and optimization techniques as necessitated by the scale of data (millions of data points, hundreds of thousands of dimensions). This dissertation presents research findings suggesting the unsuitability of current off-the-shelf crowdsourcing aggregation algorithms for real engineering design tasks due to the sparsity of expertise in the crowd, and methods that mitigate this limitation by incorporating appropriate information for expertise prediction. Next, we introduce and interpret a number of new probabilistic models for crowdsourced design to provide large-scale preference prediction and full design space generation, building on statistical and machine learning techniques such as sampling methods, variational inference, and deep representation learning. Finally, we show how these models and algorithms can advance crowdsourcing systems by abstracting away the underlying appropriate yet unwieldy mathematics, to easier-to-use visual interfaces practical for engineering design companies and governmental agencies engaged in complex engineering systems design.PhDDesign ScienceUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/133438/1/aburnap_1.pd