MULTIVARIATE DESIGN OF PROCESS EXPERIMENTS (M-DOPE)

An approach to the design of process experiments is presented for the situation where there is a large number of potentially adjustable proc ess variables, and where these variables are coupled due to process op erating constraints. Some variations of the partial least squares (PLS ) algorithm called 'selective PLS' are introduced. These algorithms al low one to combine information in past process data with current knowl edge of the process, and thereby to separate the variables into a smal l number of orthogonal groups that form the basis for experimental des igns and process optimization. The concepts are illustrated using data from an industrial mineral flotation circuit used to concentrate valu able minerals from an ore.