Quick light-mixing for image inspection using simplex search and robust parameter design

Abstract

Finding the optimal illumination conditions for industrial machine vision is iterative and time-consuming work. This study discusses simplex search for automatic illumination and the Taguchi method for algorithm tuning. The simplex search could find the illumination inputs to obtain a fine image, but the degree of fineness and processing time varied according to the algorithm parameters. The mechanism from the inputs and the degree of fineness were complex and nonlinear, so it was hard to find the best parameters through conventional experiments. To address these issues, the Taguchi method was applied to the simplex search and used to figure out the best parameter combination with a minimum number of experiments. The targets of Taguchi analysis were finer images and fewer iterations. An L 25(55) orthogonal array was constructed for 5 parameters and 5 levels, and was filled with the experimental results. A new combination of parameters determined by Taguchi analysis was applied to retests. The retest results showed fewer iterations with a fine image that was close to the best case. The Taguchi method can reduce the amount of work required to set parameters for optimal illumination. © 2014 Springer International Publishing.