Modeling multivariate statistical process control charts by ART2 neural networks

Abstract

It is well known that artificial neural networks (ANNs) are adaptable or plastic multivariate models and then can be modeled to solve complex statistical prediction and pattern recognition problems. Multivariate statistical process control (MSPC) charts are classical multivariate quality control tools. Hotelling multivariate T/sup 2/is one of them. This paper covers both the theoretical and practical considerations of an ART2 ANN and the T/sup 2/ control chart. The main reasons why ART2 is taken as an alternative MSPC tool are its abilities to learn patterns in an unknown environment and to learn a new pattern without having to retrain all of already learned patterns, which the both learning abilities are named stability-plasticity resolvability. This paper compares identification accuracy of the two MSPC tools. Guidelines are developed for demonstration necessity.