Using neural networks for mean shift identification and magnitude of bivariate autocorrelated processes. (1st January 2011)

• Record Type: Journal Article
• Authors: Fountoulaki, Aikaterini; Karacapilidis, Nikos; Manatakis, Manolis
• Is Part Of: International journal of quality engineering and technology. Volume 2:Number 2(2011)
• Page Start: 114
• Page End: 128
• Publication Date: 2011-01-01
• Subjects: quality control -- neural networks -- quality management; Engineering -- Quality control -- Periodicals; Systems engineering -- Periodicals; 620.0045
• DOI: 10.1504/IJQET.2011.039125
• Languages: English
• ISSNs: 1757-2177

Abstract

Various multivariate control charts have been proposed for detecting abnormal mean shift changes in a production process to avoid out-of-control signals. Such charts can detect an unusual shift, but they do not directly provide useful information about the shift magnitude or the variable – or group of variables – that has caused the out-of-control signal. The aim of this paper is to improve the control chart interpretation for the case of autocorrelated multivariate process control. In particular, autocorrelated data following the bivariate AR(1) model is used as the reference model. First, a neural-based procedure is used to detect as soon as possible whether an abnormal shift exists and which quality variable has caused the out-of-control signal. Then, four other neural networks (NNs) are trained to estimate shift's magnitude. Extensive numerically simulated examples are used to evaluate the performance of the proposed methodology.