“…For instance, Banerjee and Ierapetritou (2002) use high-dimensional model representations to approximate the black box, Kofler (2007) uses nonlinear covariance models, Pelikan et al (2001) apply Bayesian network models, Driessen et al (2006) optimize a linear model in a trust region. Response surface methodology-based on paradigms such as neural networks, kriging and radial basis functions-has been a popular choice for the optimization of expensive black boxes in chemical engineering (Davis and Ierapetritou, 2007), global optimization (Jones et al, 1998;Regis and Shoemaker, 2005;Sherali and Desai, 2005;Huang et al, 2006;Holmström et al, 2007;Holmström, 2008) and simulation-optimization (Daughety and Turnquist, 1981;Humphrey and Wilson, 2000;Keys and Rees, 2004;Rosen and Harmonosky, 2005;Wang, 2005;Alkhamis and Ahmed, 2006;Barton and Meckesheimer, 2006;Kleijnen, 2008;Angün et al, 2009;Yalçinkaya and Bayhan, 2009).…”