Tidsskrift: Journal of Chemometrics, vol. 16, p. 263–273, 2002
Open Access: none
In industry, variability in raw material quality often leads to unstable end-product quality. By sorting the raw materials into homogeneous groups, stability and quality of the end-product can be improved. The optimal splitting and the corresponding optimal process settings in each category can be found by fuzzy clustering. In the single-response case the distance between groups and objects is defined as the predicted response's distance to the target. When there are multiple responses, the distance between an object and a category can be based either on a quadratic loss function that weights the different responses or on desirability functions. Another approach is to put restrictions on the less important responses when optimizing the process variables. These three methods are discussed and illustrated by an example. The constrained approach was not suited for the example treated, while the other two approaches gave similar solutions to the optimization problem. Copyright (C) 2002 John Wiley Sons, Ltd.