We are developing visual steering commands that help decision makers form their preference while exploring the trade space (i.e., “shopping”) to focus in on regions/points of interest as their preference sharpens. We have noticed a basic dichotomy when using visual steering commands: (i) those that explore the trade space by broadly searching it, and (ii) those that exploit knowledge gained during trade space exploration, to guide and narrow the search. Initially, users start out by conducting a broad search, then begin exploring localized regions of the trade space to increase knowledge of the underlying relationships, finally focusing their search in a region potentially containing their most preferred designs.

Basic Sampler: The basic sampler randomly samples the input space, which is defined by the upper and lower bounds on each input variable. The sampler performs a broad search of the trade space by using a Monte Carlo simulation on the inputs of the simulation model, where each input may have a uniform, normal, or triangular distribution. These bounds can be reduced using range slider bars to “zoom in” on regions of interest as additional information becomes available.

Attractor Sampler: The attractor sampler populates new samples near a user-defined point within the trade space. A point attractor is specified in the interface with a target graphical icon, and this sampler is frequently used to try to fill “gaps” in the n-dimensional trade space. Since trade spaces may be highly modal or discontinuous, we use an evolutionary algorithm, specifically Differential Evolution, to guide the sampling process to a user defined point. The fitness of each new sample point sample is based on the normalized Euclidean distance from the specified n-dimensional point attractor.

Pareto Sampler: In engineering design, decision makers often have variables that they would like to minimize or maximize. The Pareto sampler allows a user to place a minimizing or maximizing preference on variables in the trade space, and then it will perform a Pareto search based on the user specified directions of preference. The Pareto sampler algorithm is also based on a multi-objective version of Differential Evolution, allowing users to perform a Pareto search across complex trade spaces.

[1] Stump, G.M., Lego, S., Yukish, M., Simpson, T. W., Donndelinger, J. A., Visual Steering Commands for Trade Space Exploration : User-Guided Sampling With Example. Journal of Computing and Information Science in Engineering, 2009. 9(4): p. 044501:1-10.