Robust design, robust parameter design and parameter design are common terms used to describe a collection of methods employed to optimize a process or product in terms of its mean performance and variation (noise) in performance. Parameter design is often credited to the Japanese engineer Genichi Taguchi, though the concept predates him by several decades.
Robust design addresses parameter settings. The goal of robust design is to find and adjust the fixed settings of control factors to optimize mean performance, while simultaneously making that performance insensitive to variation in the noise factor settings.
The factors that may impact product performance can be divided into two broad categories:
• Control factors, which are factors whose levels remain fixed once set by the product or process engineers, and
• Noise factors, which are factors whose levels cannot be practically controlled in operation or production.
Noise factors can often be controlled in designed experiments for purposes of parameter design, but once the product or process is in operation, the factor levels can and often will vary over time. Variation in the noise factors imparts variation to the product or process performance. On the other hand, control factors can be maintained in a relatively tight range about desired settings.
The techniques employed in robust design include crossed arrays, direct variance modeling, and combined arrays. Combined arrays are analyzed using the dual response surface approach, where one explicitly models transmitted variance; here the focus in on identifying significant noise by control factor interactions. Combined arrays also allow for combined variance modeling, where, in addition to noise variation, the process variation is a function of the control factor settings. Simulation (Monte Carlo) techniques permit an assessment of the effect of factor level variation on the response.