A review on input features for control chart patterns recognition

Abstract

Control chart pattern recognition (CCPR) is an essential tool for monitoring and diagnosing manufacturing process variability. It is used for recognizing manufacturing processes’ abnormality. The specific type of patterns can be predicted with improved classification accuracy and less computational time when using appropriate features set in classifiers. Various features set extracted from process data streams have been proposed by researchers as input data representations for control chart pattern recognition (CCPR). This could confuse new researchers as to which features set need to be selected. Therefore, this paper aims to compare statistical features, shape features and mixed features as used in CCPR and identifies related open issues and research trends. This review concludes that mix features for input data representation are more promising to achieve a better recognition performance in terms of accuracy compared to the statistical and shape features.