A role for "one-factor-at-a-time" experimentation in parameter design

Abstract

This paper explores the role of one-at-a-time experimentation in parameter design of engineering systems. The focus is on degree of improvement achieved rather than on efficiency in estimating model parameters. The performance of adaptive one-at-a-time plans is compared with the performance of orthogonal arrays through computer simulations based on data from 66 response variables in 27 full factorial experiments described in science and engineering journals and textbooks. From the simulation results, a map of the expected gains in performance is provided as a function of the degree of pure experimental error and the strength of interactions among experimental factors. When experimental error is small (less than a quarter of the factor effects) or the interactions among control factors are large (more than one-quarter of all factor effects), an adaptive one-at-a-time strategy tends to achieve greater gains than those provided by orthogonal arrays.