Feature-Based Recognition of Control Chart Patterns: A Generalized Approach

Abstract

Control chart patterns (CCPs) can be associated with certain assignable causes creating problems in the manufacturing processes and thus, the recognition of CCPs can accelerate the diagnostic search, for those causes. Researches in developing CCP recognition systems have traditionally been carried out using standardized or scaled process data to ensure the generalized applicability of such systems. Whereas standardization of data requires additional efforts for estimation of the mean and standard deviation of the underlying process, scaling of process data leads to loss of distinction between the normal and stratification patterns because a stratification pattern is essentially a normal pattern with unexpected low variability. In this paper, a new approach for generalization of feature-based CCP recognition system is proposed, in which the values of extracted shape features from the control chart plot of actual data become independent of the process mean and standard deviation. Based on a set of six shape features, eight most commonly observed CCPs including stratification pattern are recognized using heuristic and artificial neural network techniques and the relative performance of those approaches is extensively studied using synthetic pattern data.