عنوان مقاله [English]
The process control charts are most important tool of statistical process control (SPC) approach. The general control procedures of these charts only monitor charts’ samples individually and do not consider the obtained common data from successive samples as a pattern.
The existence of natural variations in the control charts is inevitable, but the appearance of significant patterns in these charts warn the special disorders in production processes and associate out-of-control situations. The natural variations usually change significant patterns from their expected correct forms; therefore, increase of qualitative sensitivities level for study of unnatural patterns in the control charts is mandatory.
In resent years, to recognize and analyze non-random patterns in the process control charts, numerous models have been presented. These models often cannot alarm the occurrence of various formation modes of cyclic and systematic patterns; since the periodic patterns have phase difference in their starting point, but most models merely have simulated one state of their occurrence.
On the other hand, few developer models of periodic patterns generating functions have applied the artificial neural networks as recognition tool. Although the neural networks are capable in patterns learning, however they have difficult architectures, time-consuming algorithms and uncertain reliability when the sensitivities of processes to the appearance of significant patterns are high.
This paper introduces a new model - based fitted cosine curve of samples for more accurate discrimination of various formation phases of cyclic and systematic patterns and more precise estimation of their corresponding parameters at different levels of sensitivity. Our proposed model compares all periodic alternatives, then selects the best of fitted cosine curve of samples, and finally determines situation of process.
The results of simulated tests indicate that the proposed model reduces cyclic and systematic patterns misclassification error and also the average of estimation error of their corresponding parameters, in comparison with developer models of periodic patterns, for various emergence states.