Systematic study to assess neural networks´ capability in classifying control chart patterns

Monitoring a process of a certain machine/device is an essential. Abnormality ought to be detected as soon as possible. This is done by means of monitoring the Control Chart Patterns (CCPs). At present, there are six well recognized patterns. A system which can accurately classify these patterns is advantageous to manufacturing processes. Many techniques in Statistical Process Control (SPC), Artificial Intelligence (AI) have been utilized in implementing of such system. Among these existing techniques, neural networks receive much attention lately due to its learning capability where precise equations and algorithms do not exist. CCPs in this are relies on Generalized Autoregressive Conditional Heteroskedasticity Model (GARH) model to generate synthetic patterns with varied degrees of noise in them. This research is the first work to systematically study the maximum level of noise in CCPs in which neural networks can classify them satisfactory within certain degrees of accuracy. Results reveal the noise levels in which neural networks can tolerate up to 90% and 95% level of accuracy. They are also discussed in terms of Signal to Noise Ratio (SNR).