Utilizing Symbolic Representation in Synergistic Neural Networks Classifier of Control Chart Patterns

Abstract

AbstractControl Chart Patterns (CCPs) can be considered as time series. Industry widely used them in their process control. Therefore, accurate classification of these CCPs is vital as abnormalities can then be detected at the earliest stage. This work proposes a framework for neural networks based classifier of CCPs. It adopts a symbolic representation technique known as Symbolic Aggregate ApproXimation (SAX) in preprocessing. It was discovered that difficulty in classifying CCPs with high signal to noise ratio lies in differentiating among three very similar categories within their six categories. Synergism of neural networks is used as the classifier. Classification comprises two levels, the super class and individual category levels. The recurrent neural network known as Time-lag network is selected as classifiers. The proposed method yields superior performance than any previous neural network based classifiers which used the Generalized Autoregressive Conditional Heteroskedasticity (GARH) Model to generate CCPs.KeywordsClassificationControl Chart Patterns (CCPs)Neural NetworksProcess controlSymbolic Aggregate Approximation (SAX)Symbolic RepresentationTime Series