Content area
Full Text
1.
Introduction
It has been more than nine decades since Sir Ronald Fisher put design of experiments (DOE) in the hands of statisticians, thereby making the study of complex systems more efficient and effective (Fisher, 1966). During the 1920s and early 1930s, Fisher introduced principles of statistical thinking into designing experimental investigations mainly in agriculture and other related life sciences. Most methods of experimentation such as factorial design and analysis of variance (ANOVA), developed during this era, are classified as classical DOE (Fisher, 1958, 1966). The application of statistically designed experiments from agriculture to more complex industrial experimentation was driven by the development of response surface methodology (RSM) in the 1950s (Box and Wilson, 1951). Over the next 30 years, RSM and other classical DOE techniques spread throughout the chemical and process industries, mainly for the purpose of research and development. However, due to lack of computing resources and adequate training in basic statistical concepts, the application of statistically designed experiments was not widespread at manufacturing process level.
By the late 1970s, statisticians switched their attention to more standardized approaches of designed experiments, popularly known as Taguchi methods (Kackar, 1985; Taguchi, 1987, 1991; Taguchi and Wu, 1980). Developed by Dr G. Taguchi, this experimentation approach (also known as robust design) suggested highly fractionated factorial designs and orthogonal arrays (OA) along with some novel statistical methods for discrete part industries including automotive, aerospace, electronics, and semiconductors.
Furthermore, the need for simpler alternatives for designed experiments led to the development of Shainin system (SS), named after its creator - Dorian Shainin, a well-known quality consultant in the USA and Europe (Bhote, 1990, 1991; Bhote and Bhote, 2000). The SS emphasized on using observational investigations prior to experimental investigations, and searching for a dominant cause using the process of elimination and leveraging (Shainin, 1993; De Mast, 2004; Steiner et al. , 2009). Whilst the work of Fisher and Taguchi is well acclaimed in contemporary manufacturing industry, the work of Dorian Shainin is criticized for being overstated and unsubstantiated (Zeigel, 2001; Moore, 1993; Nelson, 1991; Hockman, 1994).
All these statistical experimental design approaches (classical, Taguchi, and Shainin) have their proponents and opponents, and there is sufficient literature that debates the advantage of one over the other for industrial experimentation...