Design of Experiments as a Strategic Tool in Engineering Research: A Review of Applications and Best Practices

Abstract

Design of experiments (DOE) is an organized way to learn how engineering systems react to an engineered change. The strategic value is far greater than a reduction in experimental runs: an appropriate design allows effects to be identified, interaction effects to be seen, uncertainty to be quantified and the link between laboratory evidence and practical decisions to be established. This is a fundamental overview of some principles, important design families, representative engineering applications and techniques to enhance the validity of experimental results. The discussion combines factorial screening, response surface methodology, mixture experiments, restricted randomization, and sequential computer experiments. Examples of applications such as additive manufacturing, construction materials, adsorption, production of biodiesel, and the charging of batteries demonstrate how the design can be influenced by the experimental objectives and the physical constraints. An explicit synthetic numerical example illustrates the use of the CCD for the estimation of curvature, and the identification of a stationary point without attempting to validate any experiments. The main difference in the literature reviewed is the search for an attractive fitted model and the production of evidence that helps guide a repeatable engineering decision. Best practices include identifying the experimental unit, safeguarding against misleading error estimates with the help of replication, verifying the alias structure, matching models to randomization, reserving resources for confirmation, and providing publically available information to aid in re-use. DOE is best viewed as an iterative research procedure, where the process involves using statistical reasoning, physical knowledge, and clear reporting, rather than a recipe for finding the optimum condition using a software application.