SPC of Processes with Predicted Data: Application of the Data Mining Methodology

Abstract

SPC procedures are usually designed to control stability of directly observed parameters of a process. However, when quality parameters of interest are related to reliability characteristics it is practically hardly possible to monitor such characteristics directly. Instead, we use some training data in order to build a model that is used for the prediction of the value of an unobservable variable of interest basing on the values of observed explanatory variables. Such prediction models have been developed for normally distributed characteristics, both observable and unobservable. However, when reliability is concerned the random variables of interest are usually described by non-normal distributions, and their mutual dependence may be quite complicated. In the paper we consider the model of a process when traditionally applied assumptions are violated. We show that in such a case some non-statistical prediction models proposed in the area of data-mining, such as Quinlan’s C4.5 decision tree, perform better than popular linear prediction models. However, new problems have to be considered when shifts in the levels of process parameters may influence the performance of applied classification algorithms.