Lean optimization using supersaturated experimental design

In practice, product development implies studying numerous factors that affect the final product quality and define its cost. The selection of factors to study has been left to engineers. This work is an attempt for an opposite approach which does not require selecting a small subset of factors explicitly but allows us to briefly investigate most of the parameters using a limited number of experiments. Even more, we assume that the number of experiments can be smaller than the number of parameters. The paper focuses on statistical optimization using supersaturated experimental design. The authors present a new algorithm for exploring a multi-parameter system and performing a lean optimization procedure without spending a lot of efforts. The algorithm aims at making the industrial experimental process more efficient both from the resource consumption and the economic point of view. This is especially important in first stages of system analysis and has therefore practical application in industry where each experiment is very expensive and time-consuming. Numerical results demonstrate efficiency of the algorithm which has been tested for both theoretical and realistic models.