We present a methodology for establishing meaningful multivariate specification regions on incoming materials or components. The approach uses multivariate PLS regression models to extract information from databases and to relate the properties of incoming materials, and the process variables, to the quality measures on the customer's product. The specification regions are multivariate in nature and are defined in the latent variable space of the PLS model. However, they are easy to implement in practice, and they can provide much insight into what constitutes acceptable material. The approach is illustrated using a simulated process, and is then used to establish multivariate specification regions for polymer resin properties in an industrial film process.
DEVELOPING meaningful specification regions for selecting lots of raw materials entering a customer's plant is essential to ensure that the customer's desired final product quality is achievable and that a plant can operate satisfactorily. The economic advantages of being able to establish meaningful specification regions are enormous. For the supplier of materials, establishment of and meeting specification regions that consistently ensure customer satisfaction can potentially lead to a significant increase in market share. For the customer, knowing what specifications must be placed on incoming materials to ensure smooth operation of the process and to achieve the required quality of product will allow for easier selection of suppliers who can meet those specifications or for working with suppliers to achieve them. The resulting improvement in product quality should also allow for an increase in market share. In addition, by more precisely defining raw material specifications needed for a process the customer may potentially be able to expand the supplier base and accept lower cost materials that are perfectly satisfactory.