| In the first phase of an experimental study researchers and practitioners wish to eliminate factors that have a negligible effect so that those few that effect the response the most can be studied in subsequent phases. This first phase is referred to as factor screening. Ultimately the goal is to improve or optimize an industrial process or system in a systematic way through the use of experimental designs and response surface methodology.; Several existing experimental designs, including fractional factorial designs, two-stage group screening, supersaturated designs (SSD), sequential bifurcation, and iterated fractional factorial designs (IFFD) are discussed in the literature review. The available analysis methods are also discussed including regression, effects estimates, and normal probability plots. None of these methods can screen more than 20 independent variables efficiently and effectively for problems where little is known about the underlying coefficients.; Because of the need for an efficient, effective, and robust screening method for more than 20 independent variables, this research designed and demonstrated a new approach for factor screening that relies on heuristics, including genetic algorithms, to generate a design iteratively using feedback from prior observations. It was implemented in software called TSP which stands for Trocine Screening Procedure. TSP's internal analysis approach derives maximum information from each and every point to infer which factors are significant. It was shown to work equally well on cases with same signs-of-effects as with opposing signs-of-effects. Nineteen cases were used in all, implemented as simulation models with known coefficients. The average number of observations required for cases with 50 independent variables was 48 runs. The observed average errors, both in terms of failing to identify significant factors (Type II) and selecting insignificant factors (Type I), were low for most cases and acceptable for more difficult cases. Many opportunities are presented for furthering and improving this research as well. |
