Using a neural-based procedure for manufacturing process quality monitoring in the case of serially

The aim of this work is to propose a new procedure for manufacturing process quality control in the case of serially correlated data. Initially, a frequency-domain analysis of the issue is briefly discussed, in which the drawbacks of residual-based control charts are demonstrated. Subsequently, a neural network approach for quality control, which applies the ART algorithm to serially correlated data without identifying the autocorrelation model, is discussed. Performance comparisons between the neural-based algorithm and the residual-based CUSUM chart are also presented in the paper in order to validate the proposed approach. The simulation results demonstrate that the neural-based procedure achieves improved performance over the residual-based CUSUM test. In several cases, the neural network model is far superior the residual-based CUSUM test, while for a few others the difference is negligible or the CUSUM chart performs slightly better.