An artificial Neural Network approach to monitor and diagnose multi-attribute quality control processes

One of the existing problems of multi-attribute process monitoring is the occurrence of high number of falsealarms (type i error). another problem is an increase in the probability of not detecting defects when theprocess is monitored by a set of independent uni-attribute control charts. in this paper, we address both ofthese problems and consider monitoring correlated multi-attributes processes following multi-binomial distributionsusing two artificial neural network based models. in these processes, out-of-control observations aredue to assignable causes coming from some shifts on the mean vector of the proportion nonconforming of theattributes. model one, which is designed for positively correlated attributes, consists of three neural networks.the first network not only detects whether the process is out-of-control, but also determines the direction ofshifts in the attribute means. in this situation, the second and the third networks diagnose the process attribute/s that has/have caused the out-of-control signal due to increase or decrease in proportion nonconforming,respectively. model two is designed for negatively correlated attributes and consists of two neural networks.the first network is designed to detect whether the process is out-of-control and the second one diagnoses theattribute/s that make/s the signal. the results of five simulation studies on the performance of the proposedmethodology are encouraging.