Using MPCA of spectra model for fault detection in a hot strip mill

This paper proposes a diagnostic method based on the combination of multi-way principal component analysis (MPCA) and autoregressive (AR) model extraction of power spectrum density. The method is applied to detect one type of surface damage, called pincher, in a China Steel Corporation (CSC) hot strip mill. The time-domain signal is modeled by an autoregressive process because it has less bias and variation. The results of analysis show that the performance of the SPE chart is improved and that 95% of abnormal coils are detected successfully. It is found that MPCA of power spectrum density derived from an autoregressive model has the potential to detect coils with surface damage.