In engineering spaces where modeling is difficult, engineers seek a variety of well performing solutions in order to concentrate resources on promising areas of the problem space. We call this process illumination. Gaire et al have designed an algorithm specifically for illumination of problem spaces where the underlying model is computationally expensive. This algorithm, Surrogate Assisted Illumination (SAIL) uses an evolutionary algorithm called MAP-Elites to do illumination. However, SAIL introduces a Gaussian process to simulate the computationally expensive model, and Bayesian optimization for quality control of the Gaussian process. SAIL has demonstrated potential for finding a variety of well performing solutions when applied to the design of aerodynamic hulls for velomobiles. We will walk through each of the components of the SAIL algorithm, and the results of the velomobile experiment.
"Wing Design Using SAIL,"
Scholarly Horizons: University of Minnesota, Morris Undergraduate Journal: Vol. 6
, Article 5.
Available at: https://digitalcommons.morris.umn.edu/horizons/vol6/iss2/5