A fuzzy-soft learning vector quantization for control chart pattern recognition

Abstract

This paper presents a supervised competitive learning network approach, called a fuzzy-soft learning vector quantization, for control chart pattern recognition. Unnatural patterns in control charts mean that there are some unnatural causes for variations in statistical process control (SPC). Hence, control chart pattern recognition becomes more important in SPC. In order to detect effectively the patterns for the six main types of control charts, Pham and Oztemel described a class of pattern recognizers for control charts based on the learning vector quantization (LVQ) such as LVQ, LVQ2 and LVQ-X etc. In this paper, we propose a new supervised LVQ for control charts based on a fuzzy-soft competitive learning network. The proposed fuzzy-soft LVQ (FS-LVQ) uses a fuzzy relaxation technique and simultaneously updates all neurons. It can increase correct recognition accuracy and also decrease the learning time. Comparisons between LVQ, LVQ-X and FS-LVQ are made. Numerical results show that the proposed FS-LVQ has better accuracy and less learning epochs for all neurons being completely learned than LVQ and LVQ-X. Overall, FS-LVQ is highly recommended to be used as a control chart pattern recognizer.