A novel visualization enabled decision support framework for data-driven integrated design space exploration

Design preferences or targets are typically available at the system level. A designer is usually interested in understanding patches of the design space at component levels, across different stages and processes that correspond to such system targets or preferences. This demands a thorough design space exploration permitting both forward and inverse designs. Such exploration becomes cumbersome with a large number of variables and complex systems with many conflicting goals. Hence a decision support framework that permits seamless navigation in high dimensions, especially with a visual aspect for enhanced comprehension, is desirable. Current work proposes using a data-driven interpretable self-organizing map (iSOM) as a visual enabler in decision support systems for exploring the design space and understanding the trade-off in system goals. The novelty lies in being able to use a visual form to compare greater than three conflicting goals simultaneously while accounting for design variables. The proposed approach is demonstrated using two test problems: (i) hot rolling and cooling process chain design for the production of steel rods, and (ii) head and neck injury risk evaluations for vehicular crash-worthiness. Using the first problem, we demonstrate the capability of iSOM to support the solution space exploration of a many-goal steel manufacturing process chain problem to realize the design of a steel product, in the context of a compromise Decision Support Problem formulation. In the second problem, we demonstrate the capability of iSOM to support early stage Design Space Exploration (DSE) to identify critical injury risk regions of interest for different car crash scenarios. These two test problems illustrate the capability to carry out a forward and inverse design, by the proposed approach.