A hybrid method of artificial neural networks and simulated annealing in monitoring auto-correlated multi-attribute processes

The quality characteristics of both manufacturing and service industries include not only the variables but the attributes as well. While a substantial research have been performed on auto-correlated variables, little attempt has been fulfilled for auto-correlated attributes. Ignoring the imbedded autocorrelation structure in constructing control charts cause not only the in-control run length to decrease, but also the false alarms to increase. To overcome these shortcomings, in this research, an autoregressive vector first models the autocorrelation structure of the process data. Then, a modified Elman neural network is developed to generate simulated data using the ARTA algorithm. Next, a new methodology based upon the modified Elman neural network capabilities is developed to not only monitor the process, but also to detect the cause of process deterioration. In this network, instead of the back propagation, a simulated annealing approach is proposed as an alternative training technique that is able to search globally. At the end, some simulation experiments are performed and the performance of the proposed methodology with the ones of existing control charting methods is compared. The results of the comparison study are encouraging.