On-line Recognition of Abnormal Patterns in Bivariate Autocorrelated Process Using Random Forest

It is not uncommon that two or more related process quality characteristics are needed to be monitored simultaneously in production for most of time. Meanwhile, the observations obtained online often serially autocorrelated due high sampling frequency and dynamics. This goes against statistical I.I.D assumption using multivariate control charts, which may lead performance charts collapse soon. method based on pattern recognition as a non-statistical approach confined by this limitation, further provide useful information practitioners locate assignable causes led abnormalities. study proposed model Random Forest (RF) detect identify abnormalities bivariate process. The simulation experiment results demonstrate superior accuracy (RA) (97.96%) back propagation neural networks (BPNN) (95.69%), probability (PNN) (94.31%), support vector machine (SVM) (97.16%). When experimenting with simulated dynamic data flow, also achieved better average running length (ARL) standard deviation ARL (SRL) than those four comparative approaches cases mean shift magnitude. Therefore, we get conclusion RF promising detecting Although focused study, can extended control.