Track: Quality

Abstract

Lately, accurate recognition of Process Control Chart Patterns (CCPs) has been considered one of the significant tactics for supervising manufacturing processes in order to achieve better control and to improve products quality. In addition to the single patterns, a huge work has been done to recognize concurrent ones that are usually due to the presence of two or more single patterns.  Feature-based approaches are more efficient in pattern recognition especially in the case of concurrent patterns.  Furthermore, a selection of an optimal set of features can significantly reduce the diagnostic search process.  In this paper, a new approach based on a combination of Adaptive neuro‑fuzzy inference system (ANFIS), Wavelet analysis (WA) and Principal components analysis (PCA) is used to recognize concurrent patterns ANFIS has proven to show high accuracy in pattern recognition. WA is used in this paper to improve the characteristics of the patterns to facilitate the recognition process by adding frequency features to the original pattern. Then, thirteen statistical features are extracted and PCA reduces their number to the most important three.  ANFIS is used for training and testing the data based on the three extracted features as inputs and the patterns as target outputs.  Extensive performance evaluation was carried out under normally and various non-normally distributed data. The non-normality of the inputs is based on two Gamma and Beta distributions. Results indicate that the proposed approach performs with high accuracy even in the case of non-normally distributed patterns.