Stevens, David P.; Baker, Revenor C. (1992, ASQC) Bell Helicopter Textron, Fort Worth, TX 76101
This abstract is an edited version of the author's original.
This paper describes how experimental design, response surface methodology, optimization techniques, and decision theory can provide comprehensive tools for process optimization when the finished product has multiple characteristics of interest.
After an experimental design is selected, response surfaces are fit to values and variances of each response, assuming that response variance is not necessarily constant in the experimental design.
The first step in process optimization is to determine the significant factors affecting the responses. This is accomplished by either planning an experience or by screening experiments. A first-order model is usually used for screening. A new design is then chosen to fit a second-order model between factors and responses.
The methodology described here is a new approach to process optimization problems. The assumption of constant error variance is omitted and response surfaces are fit through variances as well as response values, assuming the design has a minimum of two l\replicates at each point. These surfaces are used to calculate expected losses and estimate response variability. Responses are simultaneously optimized by minimizing total expected loss. The model for total losses allows the generation of a process operating window using a user-supplied decision rule. This window makes it easier to implement the solution.
Decision making,Design of experiments (DOE),Manufacturing,Statistical methods,Response surface methodology (RSM),Taguchi method