A Comparative Study On The Second-Order Designs In Response Surface Methodology

Response surface methodology (RSM) is a popular statistical method, which is usedin robust parameter design and process optimization. Through scientifically designedexperiments, main factors that significant influence the quality of products will befound, and quality characteristics or performance measures will be optimized. Inpractice, which second-order design is chosen will greatly affect the adequacy andcost of the experiment, and is crucial to experimenters before its implementation. It isnot reasonable to evaluate the designs with single-valued criteria. As a result,synthetic evaluation and comparision on popular second-order designs is called for toensure the efficiency of the experiment.The paper introduces the related theory and describes the three phases of RSM asthe beginning part. On the basis of characteristics of the four designs, the differencesof some frequently used response surface designs --Central Composite Designs(including Central Composite Circumscribed design, Central Composite Inscribeddesign, and Central Composite Face-centered design) and Box-Behnken Design – arecompared from the points of prediction capability, center points required, factor leveletc.. When it comes to prediction capability, CCC, CCI, and BBD are rotatable, whichguarantees the uniformity of prediction error. In three factors, CCC and CCI are bestat prediction, and CCF is also superior to BBD. For the four factor case, BBD is arotation of CCDs, as a result of which, they are identical in performance. For thefive-variable case, BBD performs well compared to CCDs for all but a tiny fraction ofthe design space. So as to the complaxity of design, CCC and CCI put forward morerequirements to experiment conditions, while CCF and BBD are much simpler.Innovations of this paper rest in the following. First, this paper discussed thecharacteristics of the four designs systematically. Second, an evaluation system is putforward, which assesses the adequacy and efficiency of a design from the points ofprediction capability and complexity. The differences of the four designs in theseaspects were presented as conclusion.