A statistical process control application in casting industry through multivaiiate quality control charts and artificial neural networks

In the study, firstly Statistical Process Control was implemented to improve quality in the production process of the Brass Factory, which operates under the MCI Coorporation; and the values of the elements were monitored with univariate and multivariate control charts. However, when the out-of-control signals were detected in the univariate and the multivariate control charts, it was not possible to find out the relevant results about the interpreting of these signals. Therefore, secondly, a model based on the Process-Oriented Basic Representation (POBREP), was proposed for the interpretation of out-of-control signals. In the proposed model, the POBREP method, a multivariate control chart and artificial neural network (ANN) were used in combination. The proposed model is a general model that can be used in brass, iron and aluminum casting processes. This model can be used in all casting processes, but needs to be customized according to the problem. The POBREP method, which is used as the basis for model formation, has been used in the literature to model geometric deviations in the manufacturing industry, but has not been used in the processes such as chemistry, petro-chemistry, casting and in the modeling of material content. In this context, the POBREP method was used in this study for the first time in a different production process and was successfully applied. It is suggested that POBREP coefficients are monitored with univariate control charts in the literature. In this study, these coefficients were monitored with Hotelling T² control chart. It has also been found that there are some typical patterns representing uncontrolled situations in the quality characteristics monitored with the Hotelling T² control chart; and the ANN models are used for automatic recognition of these patterns. Moreover, during the application of the mentioned model, a low dimensional problem which is not related to the POBREP method was encountered; and solutions for this problem were produced.